Home

Atmospheric / Topographic Correction for Satellite Imagery

1. 97 5 7 11 Convert atm for another Irradiance Spectrum 97 DS Memi Help cosa rra ea 99 BSL sI 99 6 Batch Processing Reference 100 Gol Usine the batch mode oasis ds aaa A ae ew Ee A 100 6 2 Batch modules keyword driven modules 2 0 0 eee eee ees 101 CONTENTS 7 8 9 Value Added Products TAL LAL FPAR Albed 54 503 cack he eee ee Be RG ee eee ee ed 1 2 Burtace enerey balanc o i so iocs apd aa he ee eee ee A NA Sensor simulation of hyper multispectral imagery Implementation Reference and Sensor Specifics 9 1 The Monochromatic atmospheric database e s ecca dos tie e k oe 011 Visible y Near Inirared region o or gos soi ack Ok ee ee He BH D12 Thermal reigi oe o oe ee es Re Re ae ee Se ae 9 13 Database update with solar irradiance o e 9 1 4 Sensor specific atmospheric database o o 9 1 5 Resample sensor specific atmospheric LUT s with another solar irradiance 92 Supported VO le types occ GAS a GR ee EEE ee ee ee eG 9 3 Preference parameters for ATCOR 2 2 ee 9 4 Job control parameters of the inn file former in 05 Problems and Hite gt lt 6 4 44 ete aan ia 9 6 Metadata files geometry and calibration o o oo o 90 1 Landsat ETMAS o s sore pma eh RS eR AER ESA aE REE OG SPOT ot hee hued Be eae oe ESE OS ER ee Be RH eee DS 0 6 3 ALOS AVNIRA2 ooo na Eadie d ees
2. Altitude profile of the dry atmosphere e e Altitude profile of the midlatitude winter atmosphere Altitude profile of the fall autumn atmosphere o ooo o Altitude profile of the 1976 US Standard 0 2 00 0080 2 eae Altitude profile of the subarctic summer atmosphere o Altitude profile of the midlatitude summer atmosphere Altitude profile of the tropical atmosphere o e 10 Chapter 1 Introduction The objective of any radiometric correction of airborne and spaceborne imagery of optical sensors is the extraction of physical earth surface parameters such as spectral albedo directional reflectance quantities emissivity and temperature To achieve this goal the influence of the atmosphere solar illumination sensor viewing geometry and terrain information have to be taken into account Although a lot of information from airborne and satellite imagery can be extracted without radio metric correction the physical model based approach as implemented in ATCOR offers advantages especially when dealing with multitemporal data and when a comparison of different sensors is re quired In addition the full potential of imaging spectrometers can only be exploited with this approach Although physical models can be quite successful to eliminate atmospheric and topographic ef fects they inherently rely on an accurate spec
3. Radiometric CALIBRATION module e e Normalized histogram of unscaled shadow function a Panel to define the parameters for interactive de shadowing Quicklook of de shadowing results ee ees Image processing options Right panel appears if a cirrus band exists Emissivity selection panel ee Options for haze pTOCessidg e Reflectance ratio panel for dark reference pixels 0 a BRDE panel gt o anosa s diop esas a a A a ee Value added panel for a flat terrain ee Value added panel for a rugged terrain a LAL FPAR ml o rr AA AAA AA AA ee JOR status WILLOW oa de a a a Pe ee ee ai U ee a ide AFCOR Tiled Process cola a oR a ee ee ee A Top gtapl c mModiles 24 s ss aca aa we ee ee hee A ee ae Ee Slops Aspect Calculation panel ci 560 e5s be eid e eee eee ee ee eS Panel of ORY VIEW ou eg a Soe SAR ae we ee Rab eo ee ee a Example of a DEM left with the corresponding sky view image right Panel of SHADOW concisa a ewe a ee Se SO Ew ee we Se Panel of DEM smoothing o dass ag 4a aa pa beh Be wR eee ee A ee Filter ano dues co s eam A wk ee a hs eRe OR Re amp Gs e RO ge wed Resampling of a reflectance spectrum 0 00002 eee eee Statistical spectral polishing lt ss s ss aaor koii ns e Radiometric spectral polishing ooo a a Spectral smile interpolation s sa so
4. RUN Resampling pnl is Quit Figure 5 44 Resampling of a reflectance spectrum 5 5 2 Spectral Polishing Statistical Filter Remove spectral artifacts in high spectral resolution imaging spectroscopy data Inputs Input file name A hyperspectral image cube usually the output of atmospheric correction x_atm bsq Sensor Spectral Response Defines the first band of the sensor response rsp This field may be left empty in that case the wavelength information from the ENVI header is used if the CHAPTER 5 DESCRIPTION OF MODULES 83 wavelengths tag is properly given if no wavelength reference is there a spectrally equidistant succession of the spetral bands is assumed Note the Savitzky Golay filter is not wavelength aware and uses always the assumption of equidistant constantly increasing bands Number of polishing bands on each side Adjacent bands to be used for calculation on each side of the target band e g factor 3 uses 7 bands for polishing 3 on each side plus central band Smoothing Factor smoothing applied stand alone or in combination with the derivative filter 0 no smoothing 1 slight smoothing filter 1 4 1 2 moderate smoothing filter 1 2 1 3 standard smoothing filter 1 1 1 4 and more standard smoothing with moving average Polishing Filter Type Four options are available for statistical spectral polishing Derivative Filter all spectral bands of the given window size are taken int
5. Radiation components in the thermal region o o e Top level graphical interface of ATCOR e Top level graphical interface of ATCOR File 2 0 0 0 a a Top level graphical interface of ATCOR New Sensor 200 4 Topographic modules io sec aon a hoe eek Be eb ee Be we Top level graphical interface of ATCOR Atmospheric Correction ATCOR panel for flat terrain imagery 020 000008 eae Image processing options Right panel appears if a cirrus band exists Panel tor DEM tiles ni eee sw ee oe Oe we BY oe ee ee we a Typical workflow of atmospheric correction 2 ee Input output image files during ATCOR processing 0 Directory structure of ATCOR gt ss sa eso Syor eua dana dao RR A e Template reference spectra from the spec_lib library Directory structure of ATCOR with hyperspectral add on Supported analytical channel filter types oosa a e Optional haze cloud water output file o a aaa o Path radiance and transmittace of a SEBASS scene derived from the ISAC method Comparison of radiance and temperature at sensor and at surface level Top level menu of the satellite ATCOR ascos e The File Memi 3 s swir e so A a A es GO ee we a 2 Band selection dialog for ENVI file display a oaoa a Display of ENVI imagery 2 4 24 oce bake aaa a a a a a Simple text edit
6. In case of thermal bands an emissivity selection panel will appear w Constant scene emissivity 0 9800 band 13 at 10 661 micron w Emissivities water vegetation e 0 98 dry veget soil e 0 97 sand asphalt e 0 96 band 13 at 10 661 micron 2 Normalized Emissivity Method NEM max emissivity specified below Define constant scene emissivity Original NEM all surface types have the same max emissivity Adjusted NEM surface types have different max emissivity Adjusted NEM max emissivity water b s900 Adjusted NEM max emissivity green vegetation b se00 Adjusted NEM max emissivity dry veget soil b s75o Adjusted NEM max emissivity asphalt sand seso wv ISAC In Scene Atmospheric Compensation 2 ISAC and NEM separate emissivity files y DONE Figure 5 28 Emissivity selection panel CHAPTER 5 DESCRIPTION OF MODULES 73 The first two options are available for instruments with only a single thermal band the NEM and ISAC options require multiple thermal bands and are not shown if there are less than 5 thermal bands The surface classes water vegetation soil etc are calculated on the fly employing the surface reflectance spectra from the reflective bands Figure 5 29 shows the panel with the two options for haze over land processing as explained in chapter 10 5 2 Figure 5 29 Options for haze processing The panel of figure 5 30 pops up when the spatially
7. Visibility km i19 1 Solar zenith degree 43 0 Ground elevation km 0 4 SPECTRA AEROSOL TYPE VISIB ESTIMATE INFLIGHT CALIBRATION Help WATER VAPOR IMAGE PROCESSING Output file already exists change name or press OVERURITE MESSAGES QUIT Figure 4 6 ATCOR panel for flat terrain imagery Note that for a new user specified sensors these LUTs have to be calculated once prior to the first call of ATCOR This is done with the module RESLUT see section 5 2 4 available under the New Sensor menu It is recommended to check the quality of the atmospheric correction before processing the image data For that purpose the SPECTRA module should be used where the surface reflectance of small user defined boxes can be evaluated and compared with library spectra compare chapter 5 3 6 In case of calibration problems the spectral calibration module available from the Tools button of Fig 4 1 and the radiometric inflight calibration may be employed before finally pro cessing the image data The AEROSOL TYPE button provides an estimate for the recommended aerosol type e g rural maritime urban derived from the scene This module also provides a visibility value for each aerosol type based on reference pixels dark vegetation in the scene The VISIB ESTIMATE button provides a visibility value for the selected aerosol type by checking dark scene pixels in the
8. 2 5um E EA d 7 10 0 3um For flat terrain imagery with constant atmospheric conditions the global radiation is a scalar quan tity and its value can be found in the log file accompanying each output reflectance file For rugged terrain imagery the global radiation accounts for the slope aspect orientation of a DEM surface element With thermal bands a ground temperature or at least a ground brightness temperature image can be derived Then the emitted surface radiation is calculated as Reur face s O T 7 11 where es is the surface emissivity 5 669 x1078 Wm K is the Stefan Boltzmann constant and T is the kinetic surface temperature For sensors with a single thermal band such as Landsat TM an assumption has to be make about the surface emissivity to obtain the surface temperature Usually s is selected in the range 0 95 1 and the corresponding temperature is a brightness temperature A choice of es 0 97 or 0 98 is often used for spectral bands in the 10 12 um region It introduces an acceptable small temperature error of about 1 2 C for surfaces in the emissivity range 0 95 1 Examples are vegetated or partially vegetated fields e 0 96 0 99 agricultural soil e 0 95 0 97 water e 0 98 and asphalt concrete e 0 95 0 96 Emissivities of various surfaces are documented in the literature Buettner and Kern 1965 Wolfe and Zissis 1985 Sutherland 1986 Salisbury and D Ar
9. Richter R Atmospheric topographic correction for airborne imagery ATCOR 4 User Guide DLR IB 565 02 12 Wessling Germany 2012 Richter R Schl pfer D and M ller A Operational atmospheric correction for imaging spectrometers accounting the smile effect IEEE Trans Geoscience Remote Sensing Vol 49 1772 1780 2011 Richter R Wang X Bachmann M and Schl pfer D Correction of cirrus effects in Sentinel 2 type of imagery Int J Remote Sensing Vol 32 2931 2941 2011 Rodger A and Lynch M J Determining atmospheric column water vapour in the 0 4 2 5 um spectral region Proceedings of the AVIRIS Workshop 2001 Pasadena CA 2001 Salisbury J W and D Aria D M Emissivity of terrestrial materials in the 8 14 um atmo spheric window Remote Sensing of Environment Vol 42 83 106 1992 Santer R et al SPOT Calibration at the La Crau Test Site France Remote Sensing of Environment Vol 41 227 237 1992 Schl pfer D Borel C C Keller J and Itten K I Atmospheric precorrected differential absorption technique to retrieve columnar water vapor Remote Sensing of Environment Vol 65 353 366 1998 References 198 76 TT 78 79 80 81 82 83 84 85 86 87 88 89 Ra 2 Schl pfer D and Richter R Geo atmospheric processing of airborne imaging spectrome try data Par
10. The left and right channel numbers for each window or absorption region may be the same Put in a zero channel number if not applicable line 30 ch1130 1 6 vector with 6 channel numbers for the 1130 nm water vapor retrieval ch1130 1 left window channel 1050 1090 nm ch1130 2 right window channel 1050 1090 nm ch1130 4 right absorption channel 1110 1155 nm ch1130 5 left window channel 1200 1250 nm ch1130 6 right window channel 1200 1250 nm The left and right channel numbers for each window or absorption region may be the same Put in a zero channel number if not applicable 1 2 ch1130 3 left absorption channel 1110 1155 nm 4 5 line 31 chth_w1 chth_al chth_a2 chth w2 bands for thermal water vapor retrieval 9 12 um region chth_w1 left window channel split window covariance variance method SWCVR chth_w2 right window channel chth_al left absorption channel chth_a2 right absorption channel line 32 e_water e_veget e_soil e_sand surface emissivities adjusted NEM channel with Tmax line 33 0 iwv_model water vapor retrieval 1 no band regression 2 band regression The choice iwv_model 0 indicates the water vapor retrieval is disabled Option 1 means the water vapor retrieval is performed for the selected bands and in case of several measurement bands the one with the smallest standard deviation is selected per 940 and 1130 nm region Finally if bot
11. 2 m m men ATCOR supports up to na 5 regions Since the sequence of moving digital low pass filters works with square filters of size 2r 2r the area A r is approximated as the corresponding square region A r 2r Step 3 it includes the spherical albedo effect on the global flux that was initially calculated with the reference background reflectance p 0 15 and is finally adapted to the scene dependent value p by correcting with the difference p pr px y p x y 1 B x y pr s 10 14 Radiation components in rugged terrain Figure 10 3 shows a sketch of the radiation components in a rugged terrain 59 Compared to the flat terrain one additional component is taken into account in the ATCOR model It is an approximation of the terrain reflected radiation It is obtained by weighting the reflected radiation in a 0 5 km surrounding of a pixel with the terrain view factor The terrain view factor is Vierrain y 1 Vsky y and the sky view factor Vs y x y is calculated from the DEM as explained below The sky view factor is normalized to 1 for a flat terrain The reflectance is calculated iteratively The first step neglects the adjacency effect and starts with a fixed terrain reflectance of PO ain 0 1 62 m d2 c9 DN x y Lp z Ov 9 Tolz Ov b x y Esrs 2 cosP z y Ex a y 2 EP z pr Dein Vierrain 2 Y The terms are defined as p x y 10 15 x y horiz
12. 35 D 9 0 black at sensor L black at sensor Tbb 8 5 grey at surface L grey at surface Thb 8 0 amp 9 10 11 12 13 8 3 19 11 12 13 Wavelength pm Wavelength ym 1 00 0 99 Y 5 0 98 2 E E wW 0 97 0 96 0 95 amp El 10 11 12 13 Wavelength gm Figure 4 17 Comparison of radiance and temperature at sensor and at surface level CHAPTER 4 WORKFLOW 47 4 10 External water vapor map Sometimes it is convenient to use an external water vapor map even if this map could be de rived from the scene If the scene is named scene bsq then the external map should be named scene_wv bsq and it must have the same number of image lines and pixels per line as the scene If this file is available in the same directory as the scene it will be automatically used during the processing and the usage is also documented in the log file scene_atm log Note this feature is only supported for instruments that are able to retrieve the water vapor column with the intrinsic channels because the prerequisite is the availability of the corresponding LUTs Chapter 5 Description of Modules For most ATCOR modules a convenient graphical user interface is available but batch jobs can also be submitted If the atcor binary atcor sav is opened by the IDL virtual machine or when atcor is typed on the IDL command line a menu with pull down buttons pops up see Figure 5 1 with a thematic grouping of modules A detailed
13. CHAPTER 10 THEORETICAL BACKGROUND 170 the average value of the reference pixels or a spatial interpolation can be applied The visibility calculated for each reference pixel range 5 190 km in ATCOR is converted into an integer called visibility index vi with range 0 182 The visibility index is closely related to the total optical thickness 6 at 550 nm the equidistant optical thickness spacing is 0 006 for a ground at sea level and smaller for increasing elevations 5 0 185 0 006 x vi 10 71 It is easy to calculate the aerosol optical thickness AOT from a known total optical thickness by subtracting the Rayleigh optical thickness and a very small trace gas optical thickness compare Fig 2 1 in chapter 2 With the MODTRAN code the AOT at 550 nm can be calculated from a given visibility VIS km as AOT exp a z b z In VIS 10 72 where z is the surface elevation and a z b z are coefficients obtained from a linear regression of In AOT versus In VIS 0 5 Total Optical Thickness 0 9 0 50 100 150 200 Visibility km Figure 10 9 Optical thickness as a function of visibility and visibility indez CHAPTER 10 THEORETICAL BACKGROUND 171 10 4 3 Water vapor retrieval A water vapor retrieval can be included after the aerosol retrieval because the aerosol retrieval does not use water vapor sensitive spectral bands but the water vapor algorithm employing bands around 940 or 1130 nm depends on aeros
14. are appended to the input sensor name The contents of the input atcor sensor xxx are copied to the output directory e0_solar_xxx spc is deleted in the output directory and replaced by the new e0_solar_xxx_kurucz2005 spc A comparison of e0_solar_xxx spc with e0_solar_xxx_kurucz2005 spc shows the influence of the change of the irradiance spectrum In addition a new atm_lib xxx_kurucz2005 is created where all the LUTs atm from the input atm_lib xxx are replaced with the resampled selected irra 99995 diance spectrum This new folder also contains a file irrad_source txt identifying the selected CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 125 Standard high resolution atmospheric database 340 2547 nm Q CHRIS Proba database smaller spectral coverage 380 1080 nm High resolution database 1 Yexport data data atcor2 3 atm_database Solar irradiance file fi export data data atcor2 3 atm_database e0_solar_kurucz2005_04nm dat Solar irradiance file f2 Yexport data data atcor2 3 sun_irradiance e0_solar_thu2003_RSL_ku2005_04nm dat High resolution database 2 export data data atcor2 3 atm_database_thu2003_RS Convert Database 1 irradiance f1 into Database 2 irradiance f2 Input database corresponding to e0_solar_kurucz2005_04nm export data data atcor2 3 atm_database Output database corresponding to e0_solar_thu2003_RSL_ku2005_04nm exp
15. tings in file preference_parameters dat 1725 1 no interpolation for 725 820 nm channels i725 1 interpolation 1760 1 no interpolation for 760 nm channels i760 1 interpolation 1940 1 no interpolation for 940 nm channels i940 1 interpolation i11400 1 no interpolation for 1400 1900 nm channels i1400 1 interpolation toarad input filename pixelsize pirelsize sz solar_zenith atmfile atmfile elev elevation vis visibility adjrange adjrange scalef scalef The keywords in brackets are optional the meaning of all keywords is described in chapter 8 Information on all missing keywords is taken from the corresponding inn file If the keyword elev is missing and the corresponding inn file contains the DEM files eleva tion slope aspect then the simulation is performed for a rugged terrain otherwise for a flat terrain compare chapter 8 cal_regress ntargets 4 outfile regression4 This program uses the rdn files to calculate a regression for the c0 cl radiometric cali bration see chapters 2 4 and 5 3 9 The above example is for the case of n 4 targets and the output file will be regression4 cal in the directory of the rdn files which are prompted with a dialog pickfile panel A graphical user interface for this program is available in the Tools pull down menu of ATCOR labeled Calibration Coefficients with Regression makeblue input filenam
16. tm_blforest_ilu bsq for possible DEM related artifacts tee Cancel DK Figure 4 8 Panel for DEM files 4 3 Survey of processing steps Figure 4 9 shows the typical workflow of atmospheric correction A detailed description of the corresponding graphicical user interface for each module is given in chapter 5 First the image is loaded with possibly some additional information DEM files Then the sensor has to be defined the radiometric calibration file and a basic atmosphere aerosol type combination e g a summer atmosphere with a rural aerosol It is recommended to check the validity of the calibration and to estimate the visibility and perhaps the atmospheric water vapor column wv before processing the image cube The SPECTRA module can be employed for this purpose see chapter 5 3 6 Reflectance spectra of scene targets can be displayed as a function of visibility and water vapor the winter fall summer and tropical atmospheres have different wv contents see appendix A and compared with field or library spectra If calibration problems exist in a few channels a copy of the calibration file can be edited in these channels to match the reference spectrum If there are problems in many channels the inflight radiometric calibration module should be used to generate a calibration file as discussed in chapter 5 3 9 When the calibration file is OK the user can continue with the image processing Depending on the
17. 5 8 March 2002 JPL Publication 03 04 Pasadena U S A Asrar G Fuchs M Kanemasu E T and Hatfield J L Estimating absorbed photosyn thetically active radiation and leaf area index from spectral reflectance in wheat Agron J Vol 76 300 306 1984 Asrar G Theory and Applications of Optical Remote Sensing J Wiley New York 1989 Baret F and Guyot G 1991 Potentials and limits of vegetation indices for LAI and APAR assessment Remote Sensing of Environment Vol 35 161 173 1991 Berk A Bernstein L S Anderson G P Acharya P K Robertson D C Chetwynd J H and Adler Golden S M MODTRAN cloud and multiple scattering upgrades with application to AVIRIS Remote Sensing of Environment Vol 65 367 375 1998 Berk A Anderson G P Acharya P K and Shettle E P gt MODTRAN5 2 0 0 User s Man ual Spectral Sciences Inc Burlington MA Air Force Research Laboratory Hanscom MA 2008 Brutsaert W On a derivable formula for long wave radiation from clear skies Water Re sources Research Vol 11 742 744 1975 Buettner K J K and Kern C D The determination of infrared emissivities of terrestrial surfaces Journal of Geophysical Research Vol 70 1329 1337 1965 Carlson T N Capehart W J and Gillies R R A new look at the simplified method for remote sensing of daily evapotranspiration Remote Sensing of Environment V
18. D Thermal radiation from the atmosphere J Geophysical Research Vol 74 5397 5403 1969 Isaacs R G Wang W C Worsham R D and Goldberg S Multiple scattering LOW TRAN and FASCODE models Applied Optics Vol 26 1272 1281 1987 Kahle A B et al Middle infrared multispectral aircraft scanner data analysis for geological applications Applied Optics Vol 19 2279 2290 1980 Kamstrup N and Hansen L B Improved calibration of Landsat 5 TM applicable for high latitude and dark areas Int J Remote Sensing Vol 24 5345 5365 2003 Chander G Markham B L and Helder D L Summary of current radiometric calibration coefficients for Landsat MSS TM ETM and EO 1 ALI sensors Remote Sens Environm Vol 113 893 903 2009 Fontenla J M Curdt W and Haberreiter M Harder J and Tian H Semiempirical Models of the Solar Atmosphere III Set of Non LTE Models for Far Ultraviolet Extreme Ultraviolet Irradiance Computation The Astrophysical Journal 707 482 502 2009 Fontenla J M Harder J Livingston W Snow M and Woods T High resolution solar spectral irradiance from extreme ultraviolett to far infrared J Geophys Res Vol 116 D20108 31pp 2011 Fraser R S Bahethi O P and Al Abbas A H The effect of the atmosphere on classifi cation of satellite observations to identify surface features Remote Sens Environm Vol
19. ENVI header of the reflectance cube The new reference center wavelengths are included in the header of the output file If the input filename is path1 image_atm bsq the output name is path1 image_atm_smcorr bsq indicating the smile corrected common wavelength grid Function parameters are filename is the full name of the surface reflectance file fpname CHAPTER 6 BATCH PROCESSING REFERENCE 106 is the full name of smile_poly_ord4 dat i e including the path number is the above option number 1 3 and if the keyword silent is set the progress about the band processing is not issued to the command line This module is also available in the interactive mode see main menu Filter Spectral Smile Interpolation Image Cube chapter 5 5 at_derpolish infile outfile nbin respfile rsp smooth lowpass adj Derivative polishing routine PARAMETERS infile file of reflectances to be filtered outfile name of output file to be created nbin number of adjacent bands to use for filtering nbin 1 KEYWORDS respfile response file used for wavelength reference default ENVI header values lowpass perform lowpass filtering only smooth smooth the outputs by a lowpass filter of size smooth after derivative filtering adj use only adjacent bands excluding current fro derivatives at_rhoapp infile calfile eOsolar outfile scale zen date Appa
20. K P Blad B L and Dusek D Multisite analyses of spectral biophysical data for corn Remote Sensing of Environment Vol 33 1 16 1990 Wiegand C L Richardson A J Escobar D E and Gerbermann A H Vegetation indices in crop assessments Remote Sensing of Environment Vol 35 105 119 1991 Wolfe W L and Zissis G J The Infrared Handbook Office of Naval Research Washing ton DC 1985 Young S J Johnson B R and Hackwell J A An in scene method for atmospheric compensation of thermal hyperspectral data J Geophys Research Vol 107 No D24 4774 4793 2002 Zhang Y Guindon B and Cihlar J An image transform to characterize and compensate for spatial variations in thin cloud contamination of Landsat images Remote Sensing of Environment Vol 82 173 187 2002 Appendix A Altitude Profile of Standard Atmospheres This chapter contains the altitude profiles of ATCOR s standard atmospheres that are based on the MODTRAN code Only the lower 5 km altitudes are shown since this region has the largest influence on the radiative transfer results and usually comprises about 90 95 of the total water vapor column For multispectral sensors without water vapor bands e g Landsat TM or SPOT the selection of the atmosphere should be coupled to the season of the image data acquisition The influence of a large error in the water vapor estimate e g 50
21. Several possibilities exist to address this problem Gillespie et al 1998 Four options are offered by the satellite version of ATCOR a constant emissivity default e 0 98 independent of surface cover type 10 12 um region for sensors with a single thermal channel For user defined sensors with multiple thermal bands the parameter itemp_band described in 4 6 1 defines the channel employed for the surface temperature calculation fixed emissivity values assigned for 3 classes for the selected surface temperature band pa rameter itemp_band described in 4 6 1 e soil 0 96 e vegetation 0 97 else e 0 98 water and undefined class The assignment to the vegetation soil class is performed on the fly in memory employing the vegetation index red and NIR bands required and the 3 class emissivity map is available file image_atm_emi3 bsq compare chapter 4 5 In the airborne ATCOR version 69 a pre classification with more emissivity classes can be used as already suggested in 55 for multispectral thermal bands the normalized emissivity method NEM or adjusted NEM are also implemented In the NEM 23 the surface temperature is calculated for all chan nels with a constant user defined emissivity and for each pixel the channel with the highest temperature is finally selected In the adjusted NEM ANEM 12 the assigned emissivity is surface cover dependent Here we define four surface cover classes water vegetation soil dry
22. Thus the quality of the required DEM will limit the final accuracy of the geo atmospheric image product in many cases For a flat terrain and larger off nadir view angles BRDF effects may have to be accounted for and the appropriate surface cover dependent BRDF model will influence the accuracy Thermal region In the thermal wavelength region beyond 8 um the surface temperature retrieval additionally depends on the correct choice of the surface emissivity In the ATCOR model the emissivity in one thermal band is based on a classification of the reflective bands if the sensor collects co registered reflective and thermal band data Depending on the surface cover classification vegetation soil sand asphalt water etc a typical emissivity value is assigned to each class 73 If the deviation of the true surface emissivity to the assumed emissivity is less than 0 02 a typical error margin then the temperatures will be accurate to about 1 1 5 K A rule of thumb is a surface temperature error of about 0 5 0 8 K per 0 01 emissivity error if the surface temperature is much higher than the boundary layer air temperature 80 An accuracy of 1 2 K can be achieved if the emissivity estimate is better than 2 12 Bibliography 10 11 12 Adler Golden S M Matthew M W Anderson G P Felde G W and Gardner J A 2002 An algorithm for de shadowing spectral imagery Proc 11th JPL Airborne Earth Science Workshop
23. Two output surface reflectance files will be generated if tiff2envi 1 an ENVI file imagel_envi_atm bsq and a TIFF file imagel_atm tif The ENVI output is needed if the spectral classification SPECL module is run otherwise it may be deleted by the user The default for a TIFF input file is tiff2envi 0 i e the ENVI file conversion is switched off and no intermediate files are created This will save disk space and some file conversion time However if the imagel ini file specifies the de hazing or de shadowing option then tiff2envi is reset to 1 to enable the creation of of the needed additional intermediate ENVI files For non TIFF files the tiff2envi keyword is ignored If each input band is in a separate TIFF file the files should be labeled consecutively e g image_bandl1 tif image _band2 tif etc The output of ATCOR will be the corresponding CHAPTER 6 BATCH PROCESSING REFERENCE 104 surface reflectance files image_band1_atm tif image_band2_atm tif etc A keyword tiff2envi can be specified as a parameter of the batch job If not specified the default is tiff2envi 0 which means no intermediate disk files will be created only the final _atm tif With the keyword tiff2envi 1 the temporary file image_envi bsq is automat ically created and it contains the input data in the ENVI bsq band sequential format The standard output name of the atmospheric correc
24. and the combination Some criteria exist to check whether the haze land removal is likely to yield good results The haze land algorithm is switched off if those criteria are not passed ihaze 1 However as these criteria might fail in certain cases there is the option of setting haze 1 which enforces the haze removal disregarding the termination criteria ihaze 0 no haze correction ihaze 1 haze land correction might be switched off if quality check criteria are not passed ihaze 2 haze over water removal requires clear water pixels ihaze 3 haze land and haze water removal ihaze 4 sun glint removal over water ihaze 1 haze land correction is executed disregarding quality checks ihaze 2 is treated as ihaze 2 ihaze 3 haze land removal is forced haze water removal needs clear water pixels Haze removal is performed for the visible bands sun glint removal for all bands iwat_shd 0 water pixels are excluded from de shadowing land default iwat_shd 1 water pixels are included in de shadowing land CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 133 The option twat_shd 1 might be useful if the internal water classification based on purely spectral critera fails and dark land pixels are classified as water which is excluded from de shadowing So this flag is only used if the de shadowing option is set and if no external water map is supplied Example scene is mage1 bsq and
25. hs data with broad band multispectral ms data As a restriction the TOA top of atmosphere or at sensor radiance calculated with the TOARAD program assumes a nadir view The hs2ms hyperspectral to multispectral program requires the hs and ms center wavelengths and the ms channel filter curves for resampling In addition noise of the ms sensor can be included as Gaussian noise with a specified amplitude either as noise equivalent radiance NER or as noise equivalent reflectance NEAp The hs contribution to a certain ms band is weighted with the value of the ms response curve at the corresponding hs wavelength compare Fig 8 1 After summing all contributions the result is normalized with the sum of the hs filter values D Lha Rs js Lms i 8 1 E RROA where L denotes at sensor or TOA radiance R the ms response function of channel i and n is the number of hs channels covered by the i th ms filter function A similar equation is used for the resampling of surface reflectance The weight factors wz for each hs channel are calculated with 116 CHAPTER 8 SENSOR SIMULATION OF HYPER MULTISPECTRAL IMAGERY 117 o o Do vail o a T 1 1 i 1 1 1 i 1 1 1 1 1 o NS T o lo T Normalized Response Function o o Ni 1 1 1 1 1 L 1 0 62 0 64 0 66 0 68 Wavelength jem L ni 0 70 Figure 8 1 Weight factors of hyperspectral bands The solid curve sho
26. line 2 A cloud reflectance threshold Te in the blue green region to define a cloud mask Pixels belong to the cloud mask if al p blue gt T or a2 p green gt Te asterisk apparent reflectance Typical values for T range from 15 35 If the cloud reflectance threshold is too high clouds will be included in the haze mask This will reduce the performance of the haze removal algorithm CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 128 line 3 A surface reflectance threshold p 1 for water in the NIR band Pixels belong to the water mask if p NIR lt pw only NIR band available line 4 A surface reflectance threshold pw2 for water in the 1600 nm region if band exists Pixels belong to the water mask if p NIR lt pu1 and pi600 lt Pw2 The defaults Pw1 5 and pwz 3 allow some margin for turbid water line 5 interpolate bands in 760 nm oxygen region 0 no 1 yes line 6 interpolate bands in 725 825 nm water region 0 no 1 yes line 7 interpolate bands in 940 1130 nm water region 0 no 1 yes line 8 smooth water vapor map box 50m 50m 0 no 1 yes The water vapor map is calculated on a pixel by pixel basis a moderate spatial smoothing 50m 50m or at least 3 3 pixels reduces the noisy appearance line 9 interpolate bands in 1400 1900 nm nm water region 0 no 1 yes line 10 cut off limit for max surface reflectance default 150 line 11 out_hew bsq file haze cloud water la
27. rugged terrain ksolflux gt 0 These are dummy values not used if ksolflux 0 or for a flat terrain line 27 2 2 ihot_mask ihot_dynr parameters for haze correction ihot_mask 1 small area haze mask ihot_mask 2 large area haze mask ihot_dynr 1 thin to medium haze levels are corrected ihot_dynr 2 thin to thick haze levels are corrected line 28 2 0 500 0 12 0 08 1 iclshad_mask thr_shad phi_unscl_max phi_scl_min istretch_type Parameters for correction of cloud building shadow effects if icl shadow gt 0 Default values are put in this line even if icl shadow 0 iclshad_mask 1 2 3 small medium large cloud shadow mask thr_shad 0 500 threshold for core shadow areas 999 means threshold is calculated from image histogram CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 135 phi_unscl max max of unscaled shadow function maz see chapters 2 5 10 5 5 phi_sclmin min of scaled shadow function see chapters 2 5 10 5 5 istretch_type 1 linear stretching 2 exponential stretching of into line 29 ch940 1 6 vector with 6 channel numbers for the 940 nm water vapor retrieval ch940 1 left window channel 850 890 nm ch940 2 right window channel 850 890 nm ch940 3 left absorption channel 920 970 nm ch940 4 right absorption channel 920 970 nm ch940 5 left window channel 1000 1040 nm ch940 6 right window channel 1000 1040 nm
28. s calibration module See section 5 7 9 for details about how to use this routine Note If several calibration targets are employed care should be taken to select targets without spectral intersections since calibration values at intersection bands are not reliable If intersections of spectra cannot be avoided a larger number of spectra should be used if possible to increase the reliability of the calibration 2 5 De shadowing Remotely sensed optical imagery of the Earth s surface is often contaminated with cloud and cloud shadow areas Surface information under cloud covered regions cannot be retrieved with optical sensors because the signal contains no radiation component being reflected from the ground In shadow areas however the ground reflected solar radiance is always a small non zero signal be cause the total radiation signal at the sensor contains a direct beam and a diffuse reflected skylight component Even if the direct solar beam is completely blocked in shadow regions the reflected diffuse flux will remain see Fig 2 8 Therefore an estimate of the fraction of direct solar irradiance for a fully or partially shadowed pixel can be the basis of a compensation process called de shadowing or shadow removal The method can be applied to shadow areas cast by clouds or buildings CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 25 Figure 2 8 Sketch of a cloud shadow geometry Figure 2 9 shows an example of removing
29. 0973 plus ozone thickness 6 0 03 plus a very small amount due to trace gases plus the contribution of residual aerosols in the higher atmosphere 2 100 km with 6 0 04 The minimum optical thickness or maximum visibility is reached if the air does not contain aerosol particles so called Rayleigh limit which corresponds to a visibility of 336 km at sea level and no aerosols in the boundary layer and higher atmosphere In this case the total optical thickness molecular and ozone is about 0 13 Since the optical thickness due to molecular scattering nitrogen and oxygen only depends on pressure level it can be calculated accurately for a known ground elevation The ozone contribution to the optical thickness usually is small at 550 nm and a climatologic geographic average can be taken This leaves the aerosol contribution as the most important component which varies strongly in space and time Therefore the aerosol optical thickness AOT at 550 nm is often used to characterize the atmosphere instead of the visibility 1 5 2 55 1 0 A AA A AE 3 L H 0 hw HO 4 J 0 8 CO CO PAra 1 gos J 80 6 a 7 z E T 7 15 E E 04 CO S 5 J 0 5 total optical thickness 550 nm sea level F 7 0 2 AOT 1 H O L H O J 0 0 11 1 ii 1 AAA ya A A a 50 1090 150 200 0 5 1 0 5 2 0 2 5 Visibility km Wavelength m Figure 2 1 Visibility AOT and total
30. 4 To perform a TOA at sensor radiance simulation for a given scene the user has to resample files from the monochromatic atmospheric database e for the altitude 99 000 m that serves as flight altitude for space sensors see chapter 4 6 After running ATCOR for a certain scene and sensor a surface reflectance cube is obtained which is input to the TOA at sensor simulation that can be performed for a flat or a mountainous terrain A detailed description of the keywords of toarad follows CHAPTER 8 SENSOR SIMULATION OF HYPER MULTISPECTRAL IMAGERY 119 Figure 8 3 Graphical user interface of program HS2MS e input datal image_atm bsq the _atm bsq indicates a surface reflectance file which is the output of an ATCOR run The input file to ATCOR was datal image bsq and toarad extracts some information from the corresponding file datal image inn for example the sensor name The output file name is datal image_toarad bsq e atmfile h99000_wv29_rura this is an example of an atmospheric look up table file with a rural aerosol and a water vapor column of 2 9 gem see chapter 9 1 If the keyword atmfile is not specified then 299000_wv10_rura will be taken e elev 500 an example of a ground elevation at 500 m above sea level If elev is not specified then elev 0 is assumed However if the keyword elev is not specified and the datal image inn file contains file names for the DE
31. 5 10 20 30 correspond to view angles of 4 4 8 8 17 6 and 26 2 re spectively View Tilt i i Incidence9 Figure 9 5 SPOT orbit geometry In addition to the tilt angle the view direction with respect to the flight path is specified Nearly all SPOT data 99 9 is recorded in the descending node i e flying from the north pole to the equator indicated by a negative value of the velocity vector for the Z component in the METADATA DIM Then a positive incidence tilt angle in METADATA DIM means the tilt direction is left of the flight direction east for the descending node This is indicated by an L in the incidence angle in VOL_LIST PDF e g incidence angle L20 6 degree A negative incidence angle means the sensor is pointing to the west coded as R right in the VOL LIST PDF e g incidence angle R20 6 degree For ATCOR the satellite azimuth as seen from the recorded image has to be specified If a denotes the scene orientation angle with respect to north see Fig 9 6 then the satellite azimuth angle py as viewed from the scene center is e gy a 270 if tilt incidence angle is positive L left case east e y a 90 if tilt incidence angle is negative R right case west Attention CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 141 view azimuth a 909 orientation angle descending L gt node view azimuth 2
32. 5 49 Apparent Reflectance Calculation 5 6 3 Resample Image Cube This routine allows to simulate a multispectral image from imaging spectroscopy data A detailed description of this routine is given in chapter 8 CHAPTER 5 DESCRIPTION OF MODULES 89 5 7 Menu Tools The Tools menu contains contains a collection of useful routines such as the calculation of the solar zenith and azimuth angles spectral classification nadir normalization for wide field of view imagery adding of a synthetic blue channel for multispectral sensors with a blue band e g SPOT which is done for the atmospherically corrected surface reflectance image spectral calibration con version of the monochromatic atmospheric database from one to another solar irradiance spectrum and BIL to BSQ conversion Figure 5 50 The tools menu 5 7 1 Solar Zenith and Azimuth The routine SOLAR_GEOMETRY is used to calculate the zenith and azimuth angle for the image location and acquisition time All explanations concerning the definition of angles are included in the panel Fig 5 51 Figure 5 51 Calculation of sun angles CHAPTER 5 DESCRIPTION OF MODULES 90 5 7 2 SPECL Spectral Reflectance Classification SPECL is a spectral classification based on reflectance data The program compares each pixel spectrum to a list of template spectra It is assigned to the class with the best match or put into the class undefined with class label 0 Figures
33. 5 554 6 0 31 1 0 201 Total ground to space water vapor Table A 6 Altitude profile of the midlatitude summer atmosphere Total ground to space water vapor content 2 92 cm or g cm altitude pressure temperature rel humidity abs humidity km mbar C g m 0 1013 26 4 75 18 9 1 904 20 4 73 13 0 2 805 14 4 74 9 3 3 715 10 4 48 4 7 4 633 3 8 35 2 2 5 559 3 0 38 1 5 Table A 7 Altitude profile of the tropical atmosphere Total ground to space water vapor content 4 11 cm or g cm 2 g Appendix B Comparison of Solar Irradiance Spectra The following two plots show the relative differences between two extraterrestrial solar irradiance sources e Kurucz 1997 distributed with MODTRAN Berk et al 2008 6 The previous high resolution monochromatic databases of ATCOR were calculated with this spectrum e Fontenla 2011 Fontenla et al 2009 2011 35 36 The new ATCOR release uses the improved quiet sun spectrum of Fontenla and co workers also referred to as low activity sun As explained in chapters 5 7 10 5 7 11 the user can convert the database of atmospheric look up ta bles from one solar irradiance source to another one provided that the spectral range and sampling distance agrees with the template spectra in the sun_irradiance directory of ATCOR Currently irradiance spectra of Kurucz 1997 Kurucz 2005 distribut
34. 6 229 249 1977 Kaufman Y J and Sendra C Algorithm for automatic atmospheric corrections to visible and near IR satellite imagery Int J Remote Sensing Vol 9 1357 1381 1988 Kaufman Y J et al The MODIS 2 1 um channel correlation with visible reflectance for use in remote sensing of aerosol IEEE Transactions on Geoscience and Remote Sensing Vol 35 1286 1298 1997 Kobayashi S and Sanga Ngoie K The integrated radiometric correction of optical remote sensing imageries Int J Remote Sensing Vol 29 5957 5985 2008 Kobayashi S and Sanga Ngoie K A comparative study of radiometric correction methods for optical remote sensing imagery the IRC vs other image based C correction methods Int J Remote Sensing Vol 30 285 314 2009 Irish R R Barker J L Goward S N and Arvidson T Characterization of the Landsat 7 ETM automated cloud cover assessment ACCA algorithm Photogr Eng Remote Sens Vol 72 1179 1188 2006 Yi C Y Haze reduction from the visible bands of Landsat TM and ETM images over a shallow water reef environment Remote Sens Environm Vol 112 1773 1783 2008 References 196 44 Liang S Falla Adl H Kalluri S Jaja J Kaufman Y J and Townshend J R G An 49 50 51 52 53 a A 55 56 57 58 operational atmospheric correction algorithm for Landsat Thematic Mappe
35. 6 2 This routine is used after smile aware atmospheric correction I applies a linear interpolation on the reflectance data in order to bring the spectral bands to a common reference in across track direction The inputs are as follows see Fig 5 47 Inputs Input Image A hyperspectral image cube usually the output of atmospheric correction in smile mode _atm bsq Smile polynomial file The file smile_poly_ord4 dat as of the sensor definition used for the smile aware atmospheric correction Options Three options for the spectral reference wavelength grid to be used for interpolation may be selected center of detector array The spectral position of the center pixel in across track direction of the detector is taken as the reference wavelength for each spectral band average over all detector columns For each spectral band the average of all smiled center wavelengths is calculated and used as the new reference wavelength nominal position ENVI header the nominal position as provided in the ENVI header file is taken as the reference Output A cube containing the spectrally interpolated image data is generated and the ENVI header is updated for options 1 and 2 CHAPTER 5 DESCRIPTION OF MODULES Figure 5 47 Spectral smile interpolation 86 CHAPTER 5 DESCRIPTION OF MODULES 87 5 6 Menu Simulation The Simulation menu provides programs for the simulation of at sensor radiance scenes based on surface reflec
36. 9 1 4 Sensor specific atmospheric database This sensor specific database is created by resampling the files of the monochromatic database with the sensor s spectral response functions employing program RESLUT see figure 9 4 An aerosol subset or all aerosol files from the monochromatic database can be selected All water vapor files belonging to the selected aerosol type e g wv04 wv10 wv20 wv29 wv40 will be resampled for 9999s hyperspectral or user defined sensors The folder with the atm files also contains a file ir rad_source txt identyfying the underlying solar irradiance spectrum In addition a file of the resampled extraterrestrial solar irradiance e g e0_solar_chris_m1 spc will be created in the corresponding sensor folder e g sensor chris_m1 9 1 5 Resample sensor specific atmospheric LUTs with another solar irradiance It is also possible to resample existing sensor specific LUTs atm files with another solar irra diance spectrum Input is a sensor from the atcor sensor folder example name xxx with the corresponding spectral response files rsp and a high resolution solar irradiance file from the atcor sun_irradiance directory example e0_solar_kurucz2005_04nm dat Output is a new sensor subdirectory example sensor xxx_kurucz2005 where the first 10 char acters of the e0_solar_kurucz2005_04nm starting after the e0_solar_
37. CO at 14 ym Fig 3 1 right shows the transmittance for three levels of water vapor columns w 0 4 1 0 2 9 cm representing dry medium and humid conditions The aerosol influence still exists but is strongly reduced compared to the solar spectral region because of the much longer wavelength So an accurate estimate of the water vapor column is required in this part of the spectrum to be able to retrieve the surface properties i e spectral emissivity and surface temperature Wo 5 i i rrr rer errr sen LCO L HO il l oap 4 0 81 f L i i 8 2 0 6 E E E ES E F a F E E 5 E 04F E H 0 0 2 AQ A AN AA 1 feria 4 8 10 12 14 8 9 10 11 12 13 14 Wavelength jem Wavelength um Figure 3 1 Atmospheric transmittance in the thermal region Similar to the solar region there are three radiation components thermal path radiance L1 i e photons emitted by the atmospheric layers emitted surface radiance L2 and reflected radiance L3 In the thermal spectral region from 8 14 um the radiance signal can be written as L Loy TELgB T 70 e F 7 3 1 27 CHAPTER 3 BASIC CONCEPTS IN THE THERMAL REGION 28 L c 4DN L Ly A A Figure 3 2 Radiation components in the thermal region L Lp L T Lgg T L3 T 1 Fr where Lpatn is the thermal path radiance i e emitted and scattered radiance of different
38. DERG SS ORE ye eS 96 4 lkonos lt sae ey ak ee ae ee OR AS RES Re Se ee 0 6 9 Guickbitd i staa 48456 be Rede dee eae eee eee Re ew 9 6 6 IRS 1C 1D AA OG IROPO 5 se sade deh oe eh ee RS ae p OR i a A ai Oe ASTER e eco en ee ite a Pee ee eng ee es ee Re 9 6 9 DMC Disaster Monitoring Constellation 4 DUG UO Rape so ks deine ee oe pO a ae EN ee BA Soe ee OG A ee eh a oe bo he ee a Boe we 9 60 12 World VieWe2 occiso Saw a Be ee ee ae G 10 Theoretical Background 10 1 Basics on radiative ramsier o i sis poko da A ER He a ee O 10 Ll Solar speciral TEKOM co bn ek eG e Pa ee oe Bw ee ee 10 1 2 Integrated Radiometric Correction IRC 004 10 1 3 Spectral solar flux reflected surface radiance 200 10 1 4 Thermal spectral revion lt o oosa ee ae a 10 2 Masks for haze cloud water snow 2 ee A 10 4 Standard atmospheric conditions soosoo e a a a 10 4 1 Constant visibility aerosol and atmospheric water vapor 10 4 2 Aerosol retrieval and visibility map o 10 4 3 Water vapor retrieval o ee 10 5 Non standard conditions e ac saca ae ee 10 5 1 Empirical methods for BRDF correction 02004 10 5 2 Haze removal over land ee 10 5 3 Haze or sun glint removal over Water ee 10 54 Cirrus removal s oo we a we ee we ee p 108 108 110 116 CONTENTS 10 5 5 De shadowing 624 8 08 44 aa
39. Daedalus These instruments almost show no spectral smile i e the channel center position and bandwidth do not depend on column pixel location Spaceborne hyperspectral instruments showing the smile effect are Hyperion and CHRIS Proba airborne instruments are for example CASI 1500 and APEX This section describes the ATCOR input files required for smile sensors There are only two changes compared to the non smile instruments e The sensor definition file e g sensor_chris_mode3 dat needs one more line see Table 4 3 containing the parameters smile 1 if smile sensor otherwise 0 and filter_type a number between 1 and 9 for the type of channel filter function compare section 4 6 1 and Fig 4 14 CHAPTER 4 WORKFLOW 41 The filter types 1 to 8 are analytical functions filter type 9 is reserved for arbitrary user defined channel filter functions the band rsp files Center wavelength and bandwidth for each channel are defined in the wavelength file wvl pertaining to the center pixel column of the detector array For each spectral channel j the channel center wavelength A j depends on the image column or pixel position x The absolute value of A j is specified in the wavelength file used to generate the spectral channel response functions and it is also included in the sensor specific solar irradiance file e g e0_solar_chris mode3 spc If n is the number of image columns the change A x
40. ENVI File Version 8 0 1 c DLR 2011 7 Show Text File Select Input Image Import ENVI BIP Image Plot Sensor Response ENYI BIL Image Plot Calibration File epnas Imagine Show System File Edit Preferences QUIT Hyperion Raw Image Figure 4 2 Top level graphical interface of ATCOR File 000 X Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Help Licens Define Sensor Parameters Create Channel Filter Files BBCALC Blackbody Function T F L RESLUT Resample Atm LUTs from Monochr Database sion 8 0 2 c DLR 2011 Figure 4 3 Top level graphical interface of ATCOR New Sensor The Topographic menu contains programs for the calculation of slope aspect images from a dig ital elevation model the skyview factor and topographic shadow Furthermore it supports the smoothing of DEMs and its related layers see chapter 5 4 000 IX Satellite AT COR File New Sensor Atm Correction Topographic Filter Simulation Tools Help Slope Aspect Licensed for Daniel Version 8 0 2 c DLR 2011 Skyview Factor Shadow Mask DEM Smoothing Figure 4 4 Topographic modules The Filter menu provides spectral filtering of single spectra reflectance emissivity radiance provided as ASCII files spectral filtering of image cubes and spectral polishing see chapter 5 5 The Simulation menu provides progra
41. Geosci Remote Sensing Vol GE 22 256 263 1984 Dozier J Bruno J and Downey P A faster solution to the horizon problem Computers amp Geosciences Vol 7 145 151 1981 Dell Endice F Nieke J Schl pfer D and Itten K I Scene based method for spatial misregistration detection in hyperspectral imagery Applied Optics Vol 46 2803 2816 2007 ERSDAC ASTER User s Guide Part II Vers 3 1 2001 Gao B C Kaufman Y J Han W and Wiscombe W J Correction of thin cirrus path radiances in the 0 4 1 9 um spectral region using the sensitive 1 375 wm cirrus detecting channel J Geophys Res Vol 103 D24 32 169 32 176 1998 Gao B C Yang P Han W Li R R and Wiscombe W J An algorithm using visible and 1 38 um channels to retrieve cirrus cloud reflectances from aircraft and satellite data IEEE Trans Geosci Remote Sens Vol 40 1659 1668 2002 Gao B C Kaufman Y J Tanre D and Li R R Distinguishing tropospheric aerosols from thin cirrus clouds for improved aerosol retrievals using the ratio of 1 38 um and 1 24 um channels Geophys Res Letters Vol 29 No 18 1890 36 1 to 36 4 2002 Gao B C Meyer K and Yang P A new concept on remote sensing of cirrus optical depth and effective ice particle size using strong water vapor absorption channels near 1 38 and 1 88 pum IEEE Trans Geosci Remote Sens Vol
42. T2 haze T1 J 4 0 T2 2 12 0 0 standard cast shadow correction Q 1 reduce over under correction in cast shadow Message Cancel Save Parameters and Return Figure 5 9 Panel to edit the ATCOR preferences CHAPTER 5 DESCRIPTION OF MODULES 56 5 2 Menu New Sensor The menu New Sensor is used to create a new sensor from calibration information if the sensor is not supported as standard sensor by ATCOR 000 X Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Help Licens Define Sensor Parameters sion 8 0 2 c DLR 2011 Create Channel Filter Files BBCALC Blackbody Function T FL RESLUT Resample Atm LUTs from Monochr Database Figure 5 10 The New Sensor Menu Fig 5 11 shows the three required steps to include a new user defined usually hyperspectral sensor to ATCOR The example uses a sensor with 96 spectral bands denoted as sensor_x96 A sub directory of atcor sensor has to be created named x96 and the three files as displayed in Fig 5 11 have to be placed in this sub directory This procedure is supported by the Function Define Sensor Parameters After execution of steps 1 and 2 the new sensor will be automatically detected when ATCOR is started Details about the sensor definition files are explained in chapter 4 5 Template files of several sensors are included in the distribution New Se
43. These rules are automatically applied if Gr 0 e g during batch processing The geometric function G needs a lower bound g to prevent a too strong reduction of reflectance values Values of G greater than 1 are set to 1 and values less than the boundary g are reset to g This means the processing works in the geometric regime from Pr to 90 and the updated reflectance is Pg PLG 10 81 where pz is the isotropic Lambert value a a 5 Br 45 degr E 9 top to bottom curves 2 ira exponent b ira exponent b 1 b 1 3 top to bottom curves b 1 2 fr 45 degr b 3 4 Ar 55 degr b 1 0 amp r 65 degr 40 50 60 70 80 90 40 50 60 70 80 90 local illumination angle degree local illumination angle degree Figure 10 13 Geometric functions for empirical BRDF correction Left Functions G eq 10 80 for different values of the exponent b Right Functions G of eq 10 80 for b 1 and different start values of Gr The lower cut off value is g 0 2 Figure 10 13 shows a graphical presentation of equation 10 80 The left part displays the function G for different values of the exponent b For b 1 the decrease with 5 is strong with a constant gradient For smaller values of b the decrease with 8 is moderate initially but the gradient increases CHAPTER 10 THEORETICAL BACKGROUND 177 with larger Currently different functions G for soil sand and vegetation can be selected in ATCOR compare the graphical user interface of Figu
44. a local minimum at The secondary peak can be determined by level slicing the normalized histogram We arbitrarily define a threshold Pr as the intersection of this slice line at the level of h 2 with the normalized histogram h for P lt P lt Pmar The approach with a main peak and a smaller secondary peak is restricted to cases where the percentage of shadow pixels in the scene is less than about 25 This applies to the fully automatic processing mode If the secondary peak at Da is not clearly defined numerically i e no local minimum found at 9 or histogram difference h 2 h lt 0 03 then Dr is defined as the intersection of the slice level 0 10 with h for lt maz More flexibility exists in the interactive mode see chapter 2 5 figure 5 25 Masking of the core shadow areas with lt dy Fig 10 20 is critical like any thresholding process a large threshold could potentially include non shadow areas a low threshold could miss shadow areas The current automatic algorithm has the three user selectable options of a small medium or large core shadow mask corresponding to thresholds set at Py 0 1 7 and Py 0 1 respec tively The default value for the fully automatic algorithm is the medium size mask In addition an interactive mode for adjusting the threshold is also available A second tunable parameter is the minimum fractional direct illumination also called depth of shadow Theoreti
45. a special treatment as discussed in chapter 5 3 11 Therefore they are not automatically created from the elevation file The skyview file and cast shadow file are optional only required in extremely steep terrain The skyview calculation can also be found under the Topographic label of Fig 4 1 Blocked Options Are Not Available For The Selected Sensor Might also apply for a reduced set of bands Either Haze or Cirrus Removal not both Blocked Options Are Not Available For The Selected Sensor ES Variable Visibility aerosol optical thickness Y Yes No Variable Visibility aerosol optical thickness Yes No Variable Water Vapor sessesesessrrererersererereerereree gt Yes No Payable laten a A nica accesos Spires Ro Haze or Sun Glint Removal cccoorsrrrrccionsrrrscancnnes Y Yes No Haze Sun GLINE Ral era ereccion canes O Yes Q No Shadow Removal Clouds Buildings oooooocccrroromo y Yes No Shadow Removal Clouds Buildings sssesesseeesssees y Yes No Value Added Products ooorrrrrrorrorrocrrorsrrorerrsss Y Yes No Value Added Products s eroorcsrrosorerrrrrrcronernerenss Q Yes No Cirrus Removal s sooorrrrrssrcrorrsrorsrrcorsrsssrsso Yes Y No Solar Flux at Ground ooorcorcccorccrrccrrorrrrrsccrssss Q Yes No Solar Flux at Ground esssesseseseseseoceseseseocososoo Y Yes No Cancel o Cancel Figure 4 7 Image processing options Right panel appears if a cirrus band exists
46. absorption regions around 1400 nm and 1900 nm is recommended because of the low signal and large influence of sensor noise However interpolation can be disabled if required i e for test purposes If enabled non linear interpolation is performed in the 1400 1900 nm regions by fitting the surface reflectance curves with a hull of a template vegetation or soil spectrum All interpolated channels are marked with an in the ENVI header of the output reflectance cube Haze or sunglint removal over water the default apparent reflectance thresholds in the NIR chan nel for clear water and haze are T clear 0 04 or 4 and To haze 0 12 or 12 respec tively Pixels with values less than T clear are defined as clear water pixels with values between the thresholds T clear and T gt haze are assigned as haze or sun glint A lower value i e T clear lt 0 04 can be can be specified but might cause a wrong classification of bright coastal water sandy bottoms and bleached coral waters If the threshold 7 clear is too high haze pixels might erroneously be put into the clear water category Cast shadow areas mountainous terrain these may contain over and or undercorrected pixels during the standard empirical BRDF correction A reduction of these pixels is tried with the following steps CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 130 e bright pixels in the NIR with DN gt DN mean o o standard dev
47. and scan angle 62 It includes the scan angle dependence of the atmospheric correction functions a nec essary feature because most airborne sensors have a large FOV up to 60 90 While satellite sensors always operate outside the atmosphere airborne instruments can operate in altitudes of a few hundred meters up to 20 km So the atmospheric database has to cover a range of altitudes Since there is no standard set of airborne instruments and the spectral radiometric performance might change from year to year due to sensor hardware modifications a monochromatic atmo spheric database was compiled based on the Modtran 5 radiative transfer code This database has to be resampled for each user defined sensor Organization of the manual Chapters 2 and 3 contain a short description of the basic concepts of atmospheric correction which will be useful for newcomers Chapter 2 discusses the solar spectral region while chapter 3 treats the thermal region Chapter 4 presents the workflow in ATCOR and chapter 5 contains a detailed description of the graphical user interface panels of the major modules Chapter 6 describes the batch processing capabilities with ATCOR It is followed by chapters on value added products available with ATCOR sensor simulation miscellaneous topics and a comprehensive chapter on the theoretical background of atmospheric correction In the appendix the altitude profile of the standard atmospheres and a short interc
48. automatically counted as cloud although they might be something else e g snow or a specular reflection from a surface If a thermal band exists the following cloud criterion must also be fulfilled 1 p SWIR1 Typ lt 225 Kelvin and exclude T gt 300 Kelvin 10 43 10 42 where Tp is the at sensor blackbody temperature in the selected thermal band Cloud over water The following criteria have to be fulfilled 0 20 lt p blue lt 0 40 p green lt p blue NIR lt p green p SWIR1 lt 0 15 NDSI lt 0 2 10 44 For optically thick clouds it is not possible to distinguish clouds over water from clouds over land if only spectral criteria are used Cloud shadow Pixels must satisfy the conditions 0 04 lt p NIR lt 0 12 and p SWIR1 lt 0 20 10 45 and they should not belong to the water class This may also include building shadow pixels Snow ice Pixels must satisfy the conditions p blue gt 0 22 and NDSI gt 0 6 and DN blue lt Tsaturation 10 46 The condition DN blue lt Tsaturation Means that saturated pixels in the blue spectral band are not included in the snow mask instead they are put into the cloud class If no blue band exists a green band around 550 nm is taken However if the blue or green band is saturated and NDSI gt 0 7 then this pixel is assigned to the snow class because of the very high probability If a green band and a SWIR2 band around 2 2
49. available sensor channels there are options to process the imagery with constant or variable visibility and atmospheric water vapor For large FOV sensors an option is available to correct for across track illumination BRDF effects This is especially useful if the image recording took place in the solar principal plane In addition a spectral polishing can be performed for the atmospher ically and or BRDF corrected data as indicated by the dotted lines of figure 4 9 The polishing requires hyperspectral imagery Finally a classification may be performed Figure 4 10 shows the input output image files associated with ATCOR processing On the left part the flat terrain case is treated on the right part the rugged terrain case In mountainous terrain the DEM DEM slope and aspect files are required Optional input are the skyview file and the shadow map the latter can also be calculated on the fly The slope and aspect files can be calculated from ATCOR s interactive menu or run as a batch job slopasp_batch see chapter 6 2 The skyview file has to be computed with the skyview program see chapter 5 4 2 CHAPTER 4 WORKFLOW 35 Define Sensor a z gt ositos Figure 4 9 Typical workflow of atmospheric correction 4 4 Directory structure of ATCOR Figure 4 11 shows the directory structure of the satellite version of ATCOR There are a num ber of sub directories with the following content The bin directory holds the ATCOR
50. averaging DEM resolution x y pixel size meters 30 0 DEM height z unit On vim yen k RYN ok MESSAGES Figure 5 38 Slope Aspect Calculation panel sky flux and 1 vsky x y determines the fraction of radiation reflected from surrounding mountains onto the considered pixel see chapter 10 1 1 This program is also available in the batch mode see chapter 6 2 Input parameters besides the DEM file are e DEM horizontal resolution in meters the x and y resolution must be the same e DEM height unit supported units are m dm and cm e Angular resolution degrees in azimuth and elevation e The undersampling factor of the DEM in pixels For large DEM s the skyview processing may be very time consuming unless an undersampling is chosen here Figure 5 39 shows the GUI panel and figure 5 40 presents a skyview image derived from a DEM image An angular azimuth elevation resolution of 10 degrees 5 degrees is recommended For large images it causes a high execution time which can be reduced by selecting an undersampling factor of 3 pixels A high angular resolution is more important than a low undersampling factor 5 4 3 Shadow Mask The calculation of the shadow map is done by ATCOR after reading the DEM files If the shadow map is computed on the fly it is kept in memory and it is not stored as a separate file If the user wants to inspect the DEM shadow map the program shadow has
51. ay 0 72 a 0 61 a2 0 65 soybean with VI SAVI Choudury et al 1994 Note Since it is difficult to take into account the parameters for different fields and different sea sons it is suggested to use a fixed set of these three parameters for multitemporal studies Then the absolute values of LAI may not be correct but the seasonal trend can be captured Plants absorb solar radiation mainly in the 0 4 0 7 wm region also called PAR region photo synthetically active radiation ASRAR 1989 The absorbed photosynthetically active radiation is called APAR and the fraction of absorbed photosynthetically active radiation is abbreviated as FPAR These terms are associated with the green phytomass and crop productivity A three parameter model can be employed to approxiate APAR and FPAR Asrar et al 1984 Asrar 1989 Wiegand et al 1990 1991 FPAR C 1 A exp B LAI 7 5 Typical values are C 1 A 1 B 0 4 Again since it is difficult to account for the crop and seasonal dependence of these parameters a constant set may be used for multitemporal datasets to get the typical FPAR course as a function of time The wavelength integrated surface reflectance in a strict sense the hemispherical directional reflectance is used as a substitute for the surface albedo bi hemispherical reflectance It is calculated as 2 5um i J p A dr oa oa 7 6 f dy 0 3um Since most satellite sensors cover only part of the 0 3 2 5 u
52. b 1 2 for soil sand Vegetation eq 10 80 with exponent b 3 4 and b 1 for A lt 720 nm and A gt 720 nm respectively i e option b in the BRDF panel see Figure 5 31 strong correction e beta_thr threshold local solar illumination angle Grp where BRDF correction starts If beta thr 0 and ibrdf gt 0 then the angle Gr is calculated in ATCOR depending on the solar zenith angle and its value can be found in the corresponding _atm log file thr g g lower boundary of BRDF correction factor see chapter 2 2 eq 2 12 and chapter 10 5 1 line 23 1 0 820 0 780 0 600 lai model a0_vi al_vi a2_vi Parameters for the LAI model to be used if ksolflux gt 0 see chapter 7 line 24 0 900 0 950 0 380 c_fpar a_fpar b_fpar parameters for the fpar model to be used if ksolflux gt 0 see chapter 7 line 25 20 0 0 83 air temperature C air emissivity see chapter 7 Parameters for the net flux calculation used for flat terrain ignored for rugged terrain line 26 20 0 0 50 0 65 15 0 6 3 t_air z0_ref tgradient p wv zh_pwv see chapter 7 t air air temperature Celsius at elevation z0_ref z0_ref reference elevation for t_air km asl tgradient air temperature gradients Celsius per 100 m height p wv mb or hPa default water vapor partial pressure at z0_ref zh_pwv km scale height of water vapor exponential decrease falls to 1 e value Parameters for the net flux calculation
53. brightness or digital number DN and the at sensor radiance Fig 2 2 L co c DN 2 7 The co and c are called radiometric calibration coefficients The radiance unit in ATCOR is mWcm sr lumt For instruments with an adjustable gain setting g the corresponding equation is L tt epN 2 8 g CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 17 L cy 0DN Figure 2 2 Schematic sketch of solar radiation components in flat terrain L path radiance La reflected radiance L3 adjacency radiation During the following discussion we will always use eq 2 7 Disregarding the adjacency component we can simplify eq 2 6 L Lpath E Lire flected Lath TpEg T co ca DN 2 9 where 7 p and E are the ground to sensor atmospheric transmittance surface reflectance and global flux on the ground respectively Solving for the surface reflectance we obtain n d co e DN Lpath 2 10 P TE The factor d takes into account the sun to earth distance d is in astronomical units because the LUT s for path radiance and global flux are calculated for d 1 in ATCOR Equation 2 9 is a key formula to atmospheric correction A number of important conclusions can now be drawn e An accurate radiometric calibration is required i e a knowledge of cy c in each spectral band e An accurate estimate of the main atmospheric parameters aerosol type visibility or optical thickness a
54. dark reference pixels If not the program switches to the constant visibility mode and vis is used as a start value An optional keyword output can be used to define the output directory and name of the output reflectance file If the keyword specifies only the output path which is recommended then all output files are written to the specified output directory and the reflectance output file name is the name of the input file with _atm bsq appended The optional keyword vis can be used to overwrite the visibility value in the inn file For a constant visibility per scene npref 0 in the inv file the input vis value is the start value that will be iterated as described in chapter 10 4 1 In case of a variable scene visibility npref 1 the vis parameter is ignored if the scene contains enough dark reference pixels If not the program switches to the constant visibility mode and vis is used as a start value A negative vis value means the value abs vis is used for processing even if it causes a large percentage of negative reflectance pixels The optional keyword tiff2envi can be used if the input file has the TIFF format tiff2envi 1 produces an automatic conversion of the input TIFF file e g imagel tif into the corre sponding ENVI file e g imagel_envi bsq This file is used as a temporary input file to ATCOR and it will be automatically deleted when the ATCOR run is finished
55. discussion of the interactive panel driven modules is given hereafter whereas a description of the batch commands can be found in chapter 6 File New Sensor Atm Correction Topographic Filter Simulation Tools Help Licensed for DLR Version 8 0 c DLR 2012 Figure 5 1 Top level menu of the satellite ATCOR 48 CHAPTER 5 DESCRIPTION OF MODULES 49 5 1 Menu File The menu File offers some collaborative tools for handling of the data and ENVI files Below a short description of the individual functions is given 000 X Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Help Display ENVI File Version 8 0 1 c DLR 2011 Show Text File Select Input Image Import EW BIP Image Plot Sensor Response ENVI BIL Image Plot Calibration File ERDAS Imagine Show System File Edit Preferences Hyperion Raw Image Figure 5 2 The File Menu 5 1 1 Display ENVI File Use this function for displaying a band sequential ENVI formatted file in a simple way Band 14 637 60000 15 4000 band 14 Band 7 530 70000 16 4000 band 7 Band 2 454 70000 13 6000 band 2 Single Band Default RGB Default CIR Figure 5 3 Band selection dialog for ENVI file display An initial dialog allows to select the band s to display either a true color CIR color or a single band mode may be selected After display the following options are available within from the menu The EN
56. eR RR ee ee eS 10 6 Summary of atmospheric correction steps o 10 61 Aleorithra Tor Hat terrain roban iaa Se Rae eS A be ws 10 6 2 Algorithm for rugged terrain c co ke Re a OR es 10 7 Spectral classification of reflectance cube ee 10 8 Accuracy of the method o s serora aad s dos do bed ee we eR ee a References A Altitude Profile of Standard Atmospheres B Comparison of Solar Irradiance Spectra 183 189 189 190 191 191 193 199 202 List of Figures 2 1 2 2 2 3 2 4 2 5 2 6 2 1 2 0 29 2 10 3 1 3 2 4 1 4 2 4 3 4 4 4 5 4 6 4 7 4 8 4 9 4 10 4 11 4 12 4 13 4 14 4 15 4 16 4 17 5 1 5 2 5 3 5 4 5 9 5 6 Visibility AOT and total optical thickness atmospheric transmittance Schematic sketch of solar radiation components in flat terrain Nadir normalization of an image with hot spot geometry BERD correction im rugged terrain imagery ceo ces soea s a BE Geometric function G for three thresholds of Br o o Wavelength shifts for an AVIRIS scene o Radiometric calibration with multiple targets using linear regression Sketch of a cloud shadow geometry o e De shadowing of an Ikonos image of Munich o e Zoomed view of central part of Figure 2 9 o Atmospheric transmittance in the thermal region o e
57. external water vapor map is scenel_wv bsq in the same folder as the input scene then it is automatically used and the internal calculation is bypassed The correction of dark areas caused by topographic shadow has been improved A positive spectral reflectance correlation factor b defines the relationship p blue b p red for the dark reference pixels If b gt 0 then the surface reflectance of the reference pixels dense dark vegetation DDV in the blue band is used to modify the aerosol type if necessary to fulfill this equation This option is retained in the 2012 version but a new option with b lt 0 is added In this case the program still calculates the visibility AOT aerosol optical thickness map based on dark reference pixels but retains the selected aerosol type The tiling capability for large scenes is improved while the previous version did not allow the first tile to consist only of background pixels the new version handles any spatial distribution of background pixels in a large scene It calculates the visibility map for the tile with the lowest percentage of background pixels and processes the remaining tiles with the same average visibility to avoid potential brightness steps at tile borders The IRC Integrated Radiometric Correction method 40 41 of combined atmospheric topographic correction in rugged terrain is included as one additional option see chapter 10 1 2 Chapter 2 Basic Concepts in the
58. file the channels in the file must have the ascending band order and the maximum number of channels is 9 CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 127 e Each band in a separate file file names must include the band numbers e g image_band1 tif image band2 tif etc Again the maximum number of bands is restricted to 9 The first band is specified to ATCOR and all bands will be processed automatically e An optional keyword tiff2envi exists for the ATCOR batch and tiling modes that produces an automatic conversion of the input TIFF file e g image tif into the corresponding ENVI file e g image_envi bsq This file is used as a temporary input file to ATCOR and it will be automatically deleted when the ATCOR run is finished Two output surface reflectance files will be generated if tiff2envi 1 an ENVI file image envi_atm bsq and a TIFF file image atm tif The ENVI output is needed if the spectral classification SPECL module is run otherwise it may be deleted by the user In the interactive ATCOR mode with GUI panels the default for a TIFF input file is tiff2envi 1 i e a corresponding input ENVI bsq band sequential file is created In the batch mode the default for a TIFF input file is tiff2envi 0 i e the ENVI file conversion is switched off and no intermediate files are created This will save disk space and some file con version time However if the image ini file
59. file can be edited to match the retrieved reflectance with the field library spectrum or the inflight calibration module may be employed see chapter 5 3 9 In most cases it is useful to check some scene derived target spectra e g water vegetation or soils before starting the processing of the image cube at the risk of blind processing and obtaining wrong results 5 lek Sodio lero Uae la Select calibration file Temperature offset Save last spectrum Red 4 Green 3 Blue f Cal file tm_standard cal Atmosphere farura Display image P Target box pixels 5 Adj range km 1 00 reference spectrum Message 3 i i i UU Visibility km 44 9 Direct plot to Screen 1 w Screen 2 Extract Spectrum from x 368 y 437 Calculate refl 35 1 0 15 wavelength um 0 2 0 YHIN 9 0 max 350 Clear screen 1 Brt T C 24 4 reflectane Contrast stretching Gaussian w Histo Eq Create Zoom Window Return 1 0 0 5 0 2 0 YMIN D 0 YMAX 40 0 Clear screen 2 Brt T C 35 5 Figure 5 21 SPECTRA module To obtain a target spectrum of the scene click at any position in the image In figure 5 21 the solid white line spectrum at the top shows a coniferous forest signature the green line represents a spruce reference spectrum taken from the spec_lib directory already resampled for the Landsat 5 TM sensor The symbols mark the TM cent
60. having ground contact 2 reflected radiation Lo from a certain pixel the direct and diffuse solar radiation incident on the pixel is reflected from the surface A certain fraction is transmitted to the sensor The sum of direct and diffuse flux on the ground is called global flux 3 reflected radiation from the neighborhood L3 scattered by the air volume into the current instantaneous direction the adjacency radiance As detailed in 65 the adjacency radiation L3 consists of two components atmospheric backscattering and volume scattering which are combined into one component in Fig 2 2 to obtain a compact description Only radiation component 2 contains information from the currently viewed pixel The task of atmospheric correction is the calculation and removal of components 1 and 3 and the retrieval of the ground reflectance from component 2 So the total radiance signal L can be written as L Lpath Lreflected F Lag L L2 L3 2 6 The path radiance decreases with wavelength It is usually very small for wavelengths greater than 800 nm The adjacency radiation depends on the reflectance or brightness difference between the currently considered pixel and the large scale 0 5 1 km neighborhood The influence of the adja cency effect also decreases with wavelength and is very small for spectral bands beyond 1 5 um 65 For each spectral band of a sensor a linear equation describes the relationship between the recorded
61. higher aerosol loadings VIS 8 15 km Each visibility iteration is supplemented with an iteration of the threshold preq which is decreased in steps of 0 005 down to pred 0 025 to include only the darkest vegetation pixels see 64 for details Currently the algorithm terminates if less than 2 reference pixels are found after these two iterations In this case the user has to employ the constant visibility option specifying the value of the visibility for the scene During batch mode operation the program takes the specified visibility from the inn file Then a check for negative reflectance pixels is performed with dark pixels in the red band 660 nm vegetation and the NIR band 850 nm water and the visibility is iteratively increased up to VIS 60 km to reduce the percentage of negative reflectance pixels below 1 of the scene pixels A corresponding notice is given in the atm log output file The third step calculates the surface reflectance in the red band as a fraction a of the NIR band reflectance Pred Q Pnir 0 1 Pnir 10 70 Similar to the empirical SWIR relationships the coefficient 0 1 is an average empirical value yielding results in close agreement with the SWIR method in many cases However deviations from the nominal value a 0 1 can vary about 30 depending on biome Before the final step of atmospheric correction takes place the visibility of non reference pixels in the scene can be set to
62. is an option to switch from one extraterrestrial solar irradiance source E A to another one E2 A The delivered high spectral resolution database of atmospheric LUTs is based on the Fontenla 2011 solar irradiance spectrum Fontenla et al 2009 2011 35 36 It represents the solar irradiance for a quiet or low activity sun and is recommended as the standard spectrum The original 0 lem7 resolution spectrum is convolved with Gaussian filter functions FWHM 0 4 nm and mapped on an equidistant 0 4 nm grid The file name of this spectrum F A is e0_solar_fonten2011_04nm dat If Ry denotes the set of quantities path radiance direct diffuse solar flux based on F A then the new set Ra with the irradiance spectrum E2 A is calculated as Ro A Ri A E2a A E1 A 9 1 Figure 9 2 presents a schematic sketch of this conversion The folder sun_irradiance contains a number of solar irradiance files that can be selected The folder of the atmospheric database DB pertaining to E A includes the corresponding irradiance file e g e0_solar_fontan2011_04nm dat and the calculated new database D B includes the Ex A file e g e0_solar_kurucz2005_04nm dat The standard or active database is named atm_database while the new database includes 10 characters from the E file name e g atm_database_kurucz2005 The ATCOR tools panel contains the program to convert from one to another spectral i
63. j of the center wavelength A j with the pixel position x can be described as a 4th order polynomial using the nm unit A x j nm ao j ailj a2 j a3 j asli t 4 1 Ac z j Ay A z j 4 2 The first left hand image pixel is x 0 the last right hand image pixel is x n 1 The coefficients a j have to be stored in an ASCII file corresponding to the band sequence The first column must contain the wavelength or band number followed by the five channel dependent coefficients beginning with ag and ending with a4 one line per channel The fixed file name is smile_poly_ord4 dat and it has to be located in the corresponding sensor sub directory In the ideal case these coefficients should be derived from laboratory measurements Since an accurate description is only required for channels in atmospheric absorption regions the 5 coefficients can be set to zero for the remaining regions but they must be provided for each channel If all 5 coefficients are set to zero for a certain channel this channel is processed in the non smile mode which will expedite the processing 2 0 512 across track FOV degree pixels per line 1 128 first last reflective band 0 35 2 55 yum 0 O first last mid IR band 2 6 7 1 wm 0 O first last thermal band 7 1 14 um 0 no tilt in flight direction 0 required dummy 1 5 1 smile sensor 5 Gaussian spectral channel filter Table 4 3 Sensor definition fil
64. on the reflectance retrieval is usually very small about 1 2 reflectance at a reflectance level of 40 for Landsat TM band 4 However for sensors with water vapor bands e g MOS B or hyperspectral sensors the water vapor content plays an important role For these sensors the database contains files with four water vapor columns 2 9 2 0 1 0 0 4 cm These are used to generate interpolated and extrapolated values for the LUTs In analogy to the files for the airborne version of ATCOR the file names are h99000_wv29_rura atm h99000_wv20_rura atm etc where the h99000 indicates the symbolic height of 99000 m used for satellites the water vapor column content wv cm or g cm7 is scaled with 10 and the aerosol type is included in the last part of the name altitude pressure temperature rel humidity abs humidity km mbar C gm 0 1013 26 4 7 5 1 9 1 904 20 4 7 3 1 3 2 805 14 4 7 4 0 9 3 715 10 4 4 8 0 5 4 633 3 8 3 5 0 2 5 559 3 0 3 8 0 1 Table A 1 Altitude profile of the dry atmosphere Total ground to space water vapor content 0 41 cm 2 org cm 199 APPENDIX A ALTITUDE PROFILE OF STANDARD ATMOSPHERES Table A 2 Altitude profile of the midlatitude winter atmosphere altitude pressure temperature rel humidity abs humidity km mbar C g m 0 1018 1 0 77 3 5 1 897 4 5 70 2 5 2 789 8 0 65 1 8 3 694 11 5 57
65. over water 5 land 6 saturated blue green band 7 snow ice 8 thin cirrus over land 9 medium cirrus over land 10 thick cirrus over land 11 thin haze over land 12 medium haze over land 13 thin haze over water 14 medium haze over water 15 cloud over land 16 cloud over water 17 water Table 10 1 Class labels in the hew file Depending on the available spectral channels it may not be possible to assign certain classes Table 10 1 contains one class for cloud over land meaning water cloud whereas the low optical thick ness cloud is put into the thin and medium thickness haze class Thin and medium haze can often be corrected successfully Of course there is no clear distinction between thick haze and cloud CHAPTER 10 THEORETICAL BACKGROUND 160 We take a pragmatic view and if the haze removal is successful in areas with thick haze then these pixels can be included in the haze mask Since this is not clear at the beginning it might be necessary to run the program twice with and without haze removal A check of the results will reveal whether the haze removal was successful ATCOR contains a number of criteria to assess the probability of a successful haze removal and will switch off the haze option if the chances are not good This automatic haze termination works in most cases but a success cannot always be guaranteed There are three cirrus classes thin medium and thick Thin and medium cirrus ca
66. processing for hyperspectral systems The button Save last spectrum upper right corner of figure 5 21 can be used to save the selected surface reflectance spectrum A sequence of target DN spectra can also be generated here that is required as input to the spectral calibration module currently only available for the hyperspectral add on module Up to 9 targets can be defined to be used in the spectral calibration They have to be labeled consecutively e g target1 target2 etc These output file names without file name extension have to be specified after pressing the button Save last spectrum For each target name three files will be generated e g target1 dat surface reflectance spectrum targetl txt a description file and target1_dn1 dat the DN spectrum The sequence of targetx _dnx dat files is used in the spectral calibration module 5 3 7 Aerosol Type The aerosol type is a parameter which is fixed for atmospheric correction This routine searches automatically for the best suited aerosol type for the currently selected image This type can then be used when selecting the atmospheric file 5 3 8 Visibility Estimate ATCOR uses the Dark Dense Vegetation DDV approach to calculate the best visibility for an image This button allows to calculate the visibility estimate for the whole image without going into the image processing Note that for variable visibility options the visibility w
67. program with all modules as listed in chapter 5 The cal directory holds all supported satellite sensors in sensor specific sub directories The sensor definition is contained in two files one contains the extraterrestrial solar irradiance e g e0_solar_aster spc the second one contains the radiomet ric calibration coefficients e g aster cal The atm_lib contains the results of the atmospheric database after resampling with the sensor specific spectral response curves The spec_lib is a di rectory containing prototype reflectance spectra of water soils vegetation asphalt etc Here the user can also put field measurements of surface reflectance spectra resampled for the appropriate sensor This is useful for inflight calibration or comparison of scene spectra with ground spectra Finally the demo_data contains some demo imagery to be able to run ATCOR immediately 4 5 Convention for file names Although file names are arbitrary it is useful to agree on some conventions to facilitate the search of files especially concerning the extensions of file names Input images to ATCOR must have the band sequential format BSQ therefore it is recommended to employ the bsq as an extension e g magel bsq Then in this example the default output file name of ATCOR will be magel1_atm bsq and a log report of the processing is available in CHAPTER 4 WORKFLOW 36 Input Image BSQ format AS Flat T
68. reference background reflectance determining the effective global flux p 0 15 is used for ATCOR S spherical albedo of the atmosphere accounts for atmospheric backscattering to the ground The geometry is described by the angles O view zenith and solar zenith and relative azimuth angles compare figure 10 1 Since p and p are not known for image data and can vary within a scene equation 10 1 has to be solved for p iteratively compare equations 10 9 10 15 In a strict sense the reflectance p used here should be called hemispherical directional reflectance factor HDRF because most surfaces show an anisotropic reflectance behavior characterized by the bidirectional reflectance distribution function BRDF Nicodemus 1970 Slater 1985 The ground is illuminated hemispherically by the direct and diffuse solar flux and the reflected radia tion is recorded from a certain direction i e hemispherical input radiation directional reflected radiation Since the reflected radiation is always measured in a small cone the term hemispherical conical reflectance factor HCRF is also used but for small instantaneous field of view sensors CHAPTER 10 THEORETICAL BACKGROUND 148 directional is a sufficiently accurate geometrical description However for simplicity we will use the abbreviation reflectance in this manual In spectral regions dominated by scattering effects the terms of equation 10 1 are calculated with
69. respect to flight direction is coded as L tilt angle gt 0 a tilt right is coded as R tilt angle lt 0 The radiometric calibration coefficients are given in the unit mWm sr inm thus they have to be multiplied with 0 1 to convert them into the unit mWem sr yum used by ATCOR 9 6 4 Ikonos Ikonos metadata files look like po_3964_metadata txt where the po indicates the project order and the following number the project number The meta data include the geographic coordinates and the solar elevation and azimuth angles The sensor can tilt into any direction and the satellite geometry as viewed from the scene center is specified with CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 142 e Nominal Collection Azimuth absolute azimuth view angle e g east 90 e Nominal Collection Elevation ATCOR s tilt angle can be calculated from equation 9 7 with the Ikonos orbit altitude 680 km The tilt angle is close to the incidence 90 elevation see Table 9 1 elevation degree incidence degree tilt degree 90 0 0 0 85 5 4 5 80 10 9 0 75 15 13 5 70 20 18 0 65 25 22 4 60 30 26 9 55 35 31 2 Table 9 1 Elevation and tilt angles for Ikonos Ikonos offers a radiometric calibration L DN calcoef where calcoef is specified in the in band ra diance unit mWem sr7 see http www spaceimaging com products ikonos spectral htm For post 22 February 200
70. specifies the de hazing or de shadowing option then tiff2envi is reset to 1 to enable the creation of of the needed additional intermediate ENVI files For non TIFF files the tiff2envi keyword is ignored Summary of output data types e byte default surface reflectance scale factor 4 0 e 16 bit signed integer scale factor gt 10 0 typically 10 or 100 e float scale factor 1 0 e The corresponding negative scale factors except byte case provide an output reflectance cube allowing negative surface reflectance values only recommended in a test phase 9 3 Preference parameters for ATCOR The preference parameters are now located in a user specific HOME directory idl rese atcor3 so multiple users of the same license can retain their personal preference settings The path of the last input image is saved separately for the flat and rugged terrain versions of ATCOR i e preference_atcor2_path txt and preference_atcor3_path txt In addition the file prefer ence_parameters dat contains a number of default parameters that can be adjusted to scene prop erties This file contains the parameters with a short description line 1 A choice to set the water vapor values for water pixels 1 average water vapor value of land pixels is assigned to water pixels 2 line average of water vapor of land pixels is assigned to water pixels Only available with wwv_model 1 see the job control parameter section 9 4
71. subtracted is computed as the DN corresponding to HOT level j minus the DN corresponding to the 2 lower histogram threshold of the HOT haze areas The de hazed new digital number is see figure 10 14 DN new DN A 10 84 A A o o Delta clear line slope a DN red band Nw o a a 60 65 70 75 BO 85 930 49 45 59 55 60 DN blue band HOT level Figure 10 14 Haze removal method Left regression between red and blue band for clear areas Right calculation of Delta A as a function of the HOT haze level example Landsat TM band 1 So the haze removal is performed before the surface reflectance calculation Two options are available the use of a large area haze mask eq 10 85 which is superior in most cases or a compact smaller area haze mask eq 10 86 HOT gt mean HOT 0 5 x stdev HOT 10 85 HOT gt mean HOT 10 86 In addition the user can select between haze removal of thin medium haze or thin to moder ately thick haze the last option is superior in most cases The algorithm only works for land pixels so the near infrared band NIR is used to exclude water pixels The current implementation provides a mask for haze over land coded with 110 in the visibility index channel The haze over water mask to be coded with 111 is difficult to determine leaving room for future research and improvements The cloud mask is coded with 112 The remaining numbers 0 to 109 of
72. the method One of the most important parameters is the available number of spectral channels during the covariance matrix and matched filter part of the algorithm The minimum requirement is a band in the near infrared region 0 8 1 0 ym The performance usually increases significantly if two additional bands at 1 6 wm and at 2 2 um are available i e a Landsat TM type of multispectral sensor Even for hyperspectral imagery these three bands around 0 85 1 6 2 2 um are sufficient for the matched filter calculation The usage of a hundred bands would not be helpful but only cause numerical problems during the inversion of the covariance matrix eq 10 100 CHAPTER 10 THEORETICAL BACKGROUND 187 Figure 10 21 Cloud shadow maps of a HyMap scene Left surface reflectance image of HyMap at Chinchon Spain 12 July 2003 Colour coding RGB 878 646 462 nm channels Center standard shadow map showing a lot of artifact shadow areas grey patches which do not appear with the core shadow approach right part Right improved cloud shadow map derived from core shadow regions Spectral channels from the visible region are merely employed for the masking of cloud regions not for the matched filter part because water vegetation dark soils and shadowed pixels all range within a few percent reflectance In addition the visible region is not very sensitive to partial shadow effects because of its larger fraction of diffuse radiation co
73. the new center wavelengths of all channels The spectral bandwidth of channels is not modified CHAPTER 5 DESCRIPTION OF MODULES 95 The DN spectra will be employed in an optimization procedure that minimizes the spikes of the derived surface reflectance spectra in the atmospheric absorption regions The first target DN file has to be entered at the top left button of the GUI panel Figure 5 57 The other target files are automatically searched and employed if the nomenclature of chapter 2 3 is employed Further input data are the sensor definition the range of bands per spectrometer solar geometry and atmospheric parameters Output is a file with the spectral channel center wavelength shifts per spectrometer and a new wavelength file containing the updated wavelengths for each channel FIRST TARGET FILE Date dd mm year 20 08 2006 Use ATCOR s SPECTRA module button save last spectrum to extract target DN files Selected SENSOR Select CALIBRATION FILE i Number of spectrometers la one spectral shift value is calculated per spectrometer Definition of Bands per Spectrometer ATMOSPHERIC FILE Select Atmosphere Visibility km 23 0 Water vapor column cm 1 00 see target txt file created by SPECTRA module button save last spectrum RUN SPECTRAL CALIBRATION Figure 5 57 SPECTRAL_CAL spectral calibration The results of the spectral shift are summarized in a file spect
74. the preference parameter file and the above three conditions Note on the water mask The NIR band surface reflectance threshold for a water mask is only used for sensors without visible bands For an instrument with visible bands better water masks are obtained with the criterion that the TOA reflectance p spectrum must have a negative gradient for bands in the visible to NIR dp A dA However for a scene with a high average ground elevation we use gt 1 2 km above sea level the TOA reflectance criterion is again replaced with the NIR surface reflectance criterion because of the distinctly smaller path radiance The water threshold for a 1600 nm band is always included as the second criterion if such a band exists Sometimes the gradient criterion with eq 9 3 is not adequate and the NIR SWIR1 reflectance thresholds yield a better water mask This may happen in urban areas containing shadow pixels cast by buildings Then the NIR SWIR1 thresholds have to be defined as negative reflectance values to overwrite the gradient criterion lt 0 for 0 4 lt lt 0 85 um 9 3 The band interpolation options are only intended for hyperspectral imagery Linear interpolation is employed in the 760 725 and 825 nm regions Non linear interpolation as a function of the veg etation index is applied in the 940 and 1130 nm parts of the spectrum to account for the leaf water content in plants nterpolation in the strong atmospheric water vapor
75. the scaled DISORT option discrete ordinate radiative transfer 51 in regions with strong atmospheric absorption the more accurate correlated k algorithm is used in combination with DISORT 5 The results are stored in look up tables LUT Since MODTRAN calculates the path radiance including the diffuse reflected ground radiation in the form Taif Egl0 p m Lpatn p Lpath 0 i_ ps Lpath 0 Tais Eg p p m 10 2 two MODTRAN runs with surface reflectance p 0 and pr 0 15 are required to calculate the diffuse ground to sensor transmittance Tg and spherical albedo s from equation 10 2 Lpatn Pr Lpath 0 7 Tdif ie Eto 10 3 E pr aie 10 4 _ p _ 0 e 1 Estoy 10 5 For image data the pixel reflectance p may differ from the background reflectance p In this case the signal at the sensor consists of three components as sketched in Fig 10 2 e component 1 scattered radiance path radiance e component 2 radiation reflected from pixel under consideration e component 3 radiation reflected from the neighborhood and scattered into the viewing di rection adjacency effect Only component 2 contains information on the surface properties of the pixel the other components have to be removed during the atmospheric correction As detailed in 65 the adjacency radia tion L3 consists of two components atmospheric backscattering and volume scattering which are combined into one component i
76. the use of a larger kernel size for the slope aspect calculation e g kernel 5 or kernel 7 instead of the default kernel 3 pixels but this approach causes a reduction of the high frequency spatial information Attention in addition to using float data before resampling it is recommended to calcu late the slope aspect maps on the original coarse spatial resolution data followed by the CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 138 high resolution resampling step for all DEM files elevation slope aspect Do not employ the sequence of resampling the elevation followed by a slope aspect calculation of the high resolution elevation map because this approach enhances artifacts Steps to reduce slope aspect striping FLOAT ELEVATION file if it is stored as integer Calculate SLOPE and also ASPECT with a low pass filter 5x5 pixels Resize SLOPE ASPECT file factor 4 larger Convolve lowpass 7x7 pixels o e WwW N RA Resize with factor 0 25 using nearest neigbor to obtain original size Note that the function DEM Smoothing 5 4 4 allows an efficient DEM smoothing Landsat and ASTER thermal band processing The thermal band s are sensitive to the atmospheric water vapor column However the thermal band atmospheric LUTs US standard mid latitude summer tropical etc only pro vide a discrete set of water vapor columns u see chapter A If the nearest u value from this set deviates more than 0
77. the visibility index channel _visindex bsq file are reserved for the coding of the optical optical thickness channel are reserved for the coding of the optical depth or visibility index itself compare chapter 10 4 2 on the derivation of the visibility map Figure 10 15 shows an example of a subset of an Ikonos scene of Dresden where the haze removal algorithm was applied More images with the results of the haze removal method are shown on ATCOR s web page http www rese ch or http www op dlr de atcor CHAPTER 10 THEORETICAL BACKGROUND 179 Figure 10 15 Subset of Ikonos image of Dresden 18 August 2002 Space Imaging Europe 2002 Left original scene right after haze removal Color coding RGB 4 2 1 NIR Green Blue bands 10 5 3 Haze or sun glint removal over water The haze removal over water uses a near infrared NIR band to estimate the spatial distribution of haze The principal method is described in 43 We use a modified version of this approach without an interactive definition of haze polygons First the water pixels are masked either using spectral criteria or taking an external water map If the scene is named scene1 bsq the external map must be named scenel_water_map bsq a 1 channel 8 bit pixel or 16 bit pixel file where water is coded with an arbitrary positive number The external map is automatically taken if it is placed in the same folder as the scene The second step is the definition of clea
78. to 11 bands The file structure is band sequential If the input file name is example bsq the output reflectance file name is example_atm bsq and the value added file name is example_atm_flx bsq the flx indicating the most important part of the calculation i e the radiation and heat fluxes 7 1 LAI FPAR Albedo Many vegetation indices have been introduced in the literature Only two are presented here because these are often used for simple parametrizations of the leaf area index LAI the fraction of absorbed photosynthetically active radiation FPAR and surface energy fluxes Baret and Guyot 1991 Choudury 1994 The normalized difference vegetation index NDVI is defined as NDVI 2850 2650 7 1 P850 P650 where peso and paso are surface reflectance values in the red 650 nm and NIR 850 nm region respectively The soil adjusted vegetation index SAVI is defined as Huete 1988 Baret and Guyot 1991 with L 0 5 ps50 p650 1 5 SAVI 7 2 ps50 peso 0 5 we The leaf area index LAI can often be approximated with an empirical three parameter relationship employing a vegetation index VI SAVI or VI NDVI VI ao a exp az LAI 7 3 108 CHAPTER 7 VALUE ADDED PRODUCTS 109 Solving for LAI we obtain 1 VI LAI Tem ag at 7 4 Sample sets of parameters are ag 0 82 a 0 78 a2 0 6 cotton with varied soil types ayg 0 68 a 0 50 a2 0 55 corn and
79. to convert database from one to another solar irradiance 125 9 4 GUI panels of the satellite version of program RESLUT 126 Ue SPOT is AA 140 0 6 Solar and view geometry cir Ra ae Boe ek eR ee ale Be Bee 141 10 1 Radiation components illumination and viewing geometry 147 10 2 Schematic sketch of solar radiation components in flat terrain 149 10 3 Radiation components in rugged terrain sky view factor 4 151 10 4 Solar illumination geometry and radiation components 152 10 5 Radiation components in the thermal region o e 156 10 6 Schematic sketch of visibility determination with reference pixel 167 10 7 Correlation of reflectance in different spectral regions 168 10 8 Rescaling of the path radiance with the blue and red band 169 10 9 Optical thickness as a function of visibility and visibility index 170 10 10Reference and measurement channels for the water vapor method 171 10 11APDA ratio with an exponential fit function for the water vapor 172 10 12Nadir normalization of an image with hot spot geometry 174 10 13Geometric functions for empirical BRDF correction 176 10 14 Haze removal method e coma ee ek Be ee a eee 178 10 15Subset of Ikonos image of Dresden 18 August 2002 040 179 10 16Haze removal
80. updated as discussed in chapter 2 3 Please refer to section 5 3 9 for further detail about how to perform an inflight calibration The radiometric calibration uses measured atmospheric parameters visibility or optical thickness from sun photometer water vapor content from sun photometer or radiosonde and ground re flectance measurements to calculate the calibration coefficients co ci of equation 2 7 for each band For details the interested reader is referred to the literature Slater et al 1987 82 Santer et al 1992 Richter 1997 Depending of the number of ground targets we distinguish three cases a single target two targets and more than two targets Calibration with a single target In the simplest case when the offset is zero co 0 a single target is sufficient to determine the calibration coefficient c L a4DN Learn Tp1Eg 1 2 17 Lpath T and Ey are taken from the appropriate LUT s of the atmospheric database p is the measured ground reflectance of target 1 and the channel or band index is omitted for brevity DNf is the digital number of the target averaged over the target area and already corrected for the adjacency effect Solving for c yields _ L Lpath Em Tp1Eg T DN DN c 2 18 Remark a bright target should be used here because for a dark target any error in the ground reflectance data will have a large impact on the accuracy of c1 Calibration with two targets I
81. values of T cir ranging from 0 01 for water vapor columns W gt 1 cm to T cir 0 15 for W lt 0 5 cm So with these thresholds tropospheric aerosols might be misclassified as cirrus in situations with dust storms but this is a necessary trade off In any case those cloud areas are CHAPTER 10 THEORETICAL BACKGROUND 183 excluded from the map of pixels employed for the aerosol retrieval which is the main purpose The cirrus and boundary layer haze removal options are exclusive i e only one of them can be selected per run 10 5 5 De shadowing Remotely sensed optical imagery of the Earth s surface is often contaminated with cloud and cloud shadow areas Surface information under cloud covered regions cannot be retrieved with optical sensors because the signal contains no radiation component being reflected from the ground In shadow areas however the ground reflected solar radiance is always a small non zero signal be cause the total radiation signal at the sensor contains a direct beam and a diffuse reflected skylight component Even if the direct solar beam is completely blocked in shadow regions the reflected diffuse flux will remain see Figure 10 18 Therefore an estimate of the fraction of direct solar irradiance for a fully or partially shadowed pixel can be the basis of a compensation process called de shadowing or shadow removal The method can be applied to shadow areas cast by clouds or buildings The propos
82. with the processing parameters must already exist 000 X Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Licensed for Daniel ATCOR2 multispectral sensors flat terrain f2 c DLR 2011 Help ATCOR3 multispectral sensors rugged terrain ATCOR2 hyperspectral sensors flat terrain ATCOR3 hyperspectral sensors rugged terrain Start ATCOR Process Tiled from inn Figure 4 5 Top level graphical interface of ATCOR Atmospheric Correction Let us start with a scene from a flat terrain area where no digital elevation model DEM is needed Then the panel of Fig 4 6 will pop up First the INPUT IMAGE FILE has to be selected AT COR requires the band sequential format BSQ for the image data with an ENVI header The TIFF format is supported with some restrictions see chapter 9 2 Next the acquisition date of the image has to be updated with the corresponding button We work from top to bottom to specify the required information The scale factor defines the multiplication factor for surface re flectance range 0 100 in the output file A scale factor of 1 yields the output as float data 4 bytes per pixel However a scale factor of 100 is recommended if the input data is 16 bit 2 bytes per pixel so a surface reflectance value of say 20 56 is coded as 2056 and is stored as a 2 byte integer which means the file size is only half of the float file size If the
83. x column direction ntx and in y line direction nty e g gt atcor2_tile input datal examples example_image bsq ntx 3 nty 2 In this case the image is split into 3 2 6 tiles each tile is processed separately finally all tiles are merged into one output file and the sub scenes are deleted The maximum size of each tile depends on the available memory for a specific machine because ATCOR performs most calcula tions in memory loading one or two complete bands of the scene A typical tile size for modern machines is 3000 3000 pixels 9 Mpixels to 5000 5000 pixels 25 Mpixels The user has to try and find out the appropiate tile size As an example with a 9 Mpixel tile size and a 30 Mpixel scene the image has to be split into 4 sub scenes Assuming that the number of image columns and lines is approximately the same one would choose the keywords ntx 2 nty 2 in this example Of course processing of much smaller tiles is also possible e g ntx 20 nty 10 but this is not recommended because of potential image border effects i e larger visibility differences for the small tiles might lead to seams at the tile borders The TIFF input data format is supported however channels in the TIFF file must have the in creasing band order and the maximum number of bands should be less than 10 If all channels are in one input TIFF file example image tif the output TIFF after atmospheric correction will also hold all ch
84. 1 2 4 608 17 5 50 0 7 5 531 23 5 47 0 4 content 0 85 cm or g cm altitude pressure temperature rel humidity abs humidity km mbar C g m 0 1013 10 0 56 5 2 1 902 3 0 47 2 8 2 802 1 0 41 1 9 3 710 5 0 40 1 4 4 628 9 0 40 1 0 5 554 14 0 40 0 6 200 Total ground to space water vapor Table A 3 Altitude profile of the fall autumn atmosphere Total ground to space water vapor content 1 14 cm or g em altitude pressure temperature rel humidity abs humidity km mbar C g m 0 1013 15 0 46 5 9 1 900 8 5 49 4 2 2 795 2 0 52 2 9 3 701 4 5 51 1 8 4 616 11 0 50 1 1 5 540 17 5 48 0 6 Table A 4 Altitude profile of the 1976 US Standard Total ground to space water vapor content 1 42 cm or g cm APPENDIX A ALTITUDE PROFILE OF STANDARD ATMOSPHERES Table A 5 Altitude profile of the subarctic summer atmosphere altitude pressure temperature rel humidity abs humidity km mbar O A Em 0 1010 14 0 75 9 1 1 896 8 5 70 6 0 2 792 3 1 70 4 2 3 700 2 3 65 2 7 4 616 7 7 60 1 7 5 541 13 1 53 1 0 content 2 08 cm or g cm altitude pressure temperature rel humidity abs humidity km mbar C g m 0 1013 21 0 76 13 9 1 902 16 5 66 9 3 2 802 12 0 55 5 9 3 710 6 0 45 3 9 4 628 0 0 39 1 9
85. 1 scenes with 11 bit data calcoef is specified as 728 727 949 and 843 blue green red NIR band respectively These values have to be converted into cl 1 calcoef bandwidth and are stored in the ikonos_2001 std cal standard calibration file 9 6 5 Quickbird The metadata files are named IMD They contain the geographic coordinates The sunEl and sunAz keywords or meanSunEl and meanSunAz specify the solar elevation and azimuth angle respectively Similar to Ikonos the sensor can tilt into any direction The satellite geometry as viewed from the scene center is specified with satEl or meanSatEl satellite elevation angle and satAz or meanSatAz absolute azimuth angle ATCOR s tilt angle can be calculated from equation 9 7 with the Quickbird orbit altitude 450 km The tilt angle is close to the incidence 90 satEl value see Table 9 2 Depending on the processing date the tilt angle may also be included in the IMD file then it is named offNadirViewAngle or meanOffNadirViewAngle elevation degree incidence degree tilt degree 90 0 0 0 85 5 4 7 80 10 9 3 75 15 14 0 70 20 18 6 65 25 23 2 60 30 27 8 55 30 32 4 Table 9 2 Elevation and tilt angles for Quickbird The Quickbird sensor uses the radiance unit Wm sr in band radiance which can be con verted into a spectral band average radiance employing the ef
86. 130 nm region are selected Finally if both regions are available the average of these two water vapor maps is taken parameter iwv_model 1 in the ini file 2 A linear regression ratio LIRR is applied to multiple bands parameter iwv_model 2 This water vapor map might be more accurate because the regression reduces sensor noise and may partially compensate calibration problems in lucky cases Although the water vapor map might be less noisy the retrieved surface reflectance spectrum will always retain any channel calibration problems Remarks CHAPTER 10 THEORETICAL BACKGROUND 173 1 The APDA algorithm is relatively fast Its disadvantage is that it is not stable numerically for very low reflectance targets water shadow regions The transmittance slope ratio method 72 might work better in these cases so it is an interesting alternative water vapor algorithm However since the required processing time is much higher than for the APDA method it is currently not implemented in the ATCOR environment In addition the method requires data with a very accurate spectral and radiometric calibration otherwise its potential advantage will be lost 2 Five water vapor grid points at 0 4 1 0 2 0 2 9 and 4 0 cm are sufficient to cover the 0 5 5 0 cm range with an accuracy of about 5 10 66 10 5 Non standard conditions The non standard situations refer to bidirectional reflectance BRDF effects and scenes with a subst
87. 2 0 3 cm from measured data e g radiosonde profile the user may generate a new thermal LUT with a water vapor column u as a linear combination of two existing LUTS with water vapor contents u1 and uz u w1u1 1 w uz Example u 2 4 cm u 2 08 cm subarctic summer us 2 92 cm mid latitude summer then w ug u u2 u1 0 619 This manipulation can be performed in the SPECTRA module after pressing the button Mixing of Atmospheres There the user has to select two existing atmospheric files defines the weight w1 and assigns a file name to the new mixed LUTs This file can be loaded from the main ATCOR panel If no atmospheric water vapor information is available but lake temperature measurements exist the user can define an appropriate temperature offset kg to match the satellite derived temperature and the water body temperature The corresponding panel Surface Radiance to Temperature Conversion pops up when the SPECTRA module or the image processing options are entered 9 6 Metadata files geometry and calibration This section explains how the geometry and calibration information of various sensor specific meta data files has to be interpreted for use in ATCOR Besides the atmospheric LUTs for the nadir view there are files for off nadir view geometries covering tilt angles from 0 to 50 increment 10 and relative azimuth angles from 0 backscatter to 180 forward scatter with an increm
88. 42 1891 1899 2004 Gillespie A R Lithologic mapping of silicate rocks using TIMS In Proc TIMS Data User s Workshop JPL Publ 83 38 Pasadena CA pp 29 44 1986 Gillespie A et al A temperature and emissivity separation algorithm for Advanced Space borne Thermal Emission and Reflection Radiometer ASTER images IEEE Trans Geosc Remote Sensing Vol 36 1113 1126 1998 Gu D and Gillespie A Topographic normalization of Landsat TM images of forest based on subpixel sun canopy sensor geometry Remote Sensing of Environment Vol 64 166 175 1998 Guanter L Richter R and Moreno J Spectral calibration of hyperspectral imagery using atmospheric absorption features Applied Optics Vol 45 2360 2370 2006 Guanter L Richter R and Kaufmann H On the application of the MODTRAN4 atmo spheric radiative transfer code to optical remote sensing accepted for publication Int J Remote Sensing 30 6 14071424 doi 10 1080 01431160802438555 2009 References 195 28 29 30 31 32 33 34 35 36 37 41 42 43 Hay J E and McKay D C Estimating solar irradiance on inclined surfaces a review and assessment of methodologies Int J Solar Energy Vol 3 203 240 1985 Huete A R A soil adjusted vegetation index SAVI Remote Sensing of Environment Vol 25 295 309 1988 Idso S B and Jackson R
89. 5 16 Plot Calibration Fil s 226644 2 54644 44 8 G8 5G Re Fae eE ES Bly show system File eee 260 sie Doe eet eA he A ee e 51 8 Edit Preferences 42 0 5 05 0 40848 824 4 2 oR S SE ee 52 Menu New Sensor 2 6 4 6 2064 ias Fe eee eR ee Re eS 5 2 1 Define Sensor Parameters i oe c csao suora maaa eee ee 5 2 2 Create Channel Filter Files o e e 5 2 3 BBCALC Blackbody Functiod s sai adie e e e p a 11 14 16 19 21 23 24 27 29 29 31 34 35 35 37 37 40 42 43 47 CONTENTS 4 5 2 4 RESLUT Resample Atm LUTS from Database 59 53 Menm Atm Correction soora p egora a a a A RR e RAI i 61 Bol The TCOR main panel serb a sda sa am miga A A D a 61 5 3 2 ATCOR2 multispectral sensors flat terrain s ooo a 61 5 3 3 ATCOR3 multispectral sensors rugged terrain 61 5 3 4 ATCOR2 User defined Sensors ooo a 62 5 3 5 ATCOR3 User defined Sensors aooo eee ee ee 62 5 3 6 SPECTRA module os 02 ta ducta omadi sa 65 met Aerosol Type a s ca ae na a A a a a 66 DS Visibility Estimate 2 saw aa ae ELA ai 66 5 3 9 Inflight radiometric calibration module 66 5 3 10 Shadow removal panels e s aoa ae eu a aaa a a N a a h 4 69 53 11 Panels for Image Processing o ote s e ace aus EE Re A o i 72 Dole Start ATCOR Process Tiled TOA iccse mea a bh wee Re 77 BA Menmu TOpoBrapllC a ess some a A AR A bee 78 DAT SA as a
90. 5 52 and 5 53 show the panel of the SPECL program and a sample output Select Sensor Y ALI w ASTER w DMC w IKONOS2 w IKONOS2_PAN w IRS AB w IRS1CD ay IRSLCD_PAN w IRSPE_AWIFS w IRSP5_LISS3 w IRSP6_LISS4 lt LANDSAT4_5 w LANDSAT v LANDSAT _PAN w MERIS w MOS_B MSS w ORBVIEW w ORBVIEW_PAN w QUICKB w QUICKB_PAN ay SAC_ZC_ZMMRS w SPOT1_3 w SPOTI_3_PAN w SPOT4 w SPOTS w SPOTS_PAN INPUT Reflectance IMAGE Vexport data data7 atcor2 3 deno_data tn_Flat tn_essen000_atn bsq OUTPUT Classification IMAGE Vexport data data7 atcor2 3 deno_data tn_Flat tn_essen1000_atn_cla bsg CONTINUE Figure 5 52 SPECL spectral classification of reflectance cube Figure 5 53 Example of classification with SPECL Left true color image of Landsat TM right result of classification 5 7 3 SPECL for User Defined Sensors This function is the same as described above in Section 5 7 2 and Figure 5 53 The only difference being that the available sensors are the ones defined in the sensor directory of the ATCOR installation i e the self defined sensors 5 7 4 Nadir normalization Wide FOV Imagery This module NADIR_REFL performs an empirical BRDF correction by normalizing the across track radiance gradients to the nadir brightness value see chapters 2 2 10 5 1 Figure 5 54 shows the corresponding GUI panel CHAPTER 5 DESCRIPTION OF MODULES 91 The nadir normalization was originally implemented for the airborne version of
91. 70 Figure 9 6 Solar and view geometry e Tricky The orientation angle is specified in the VOL_LIST PDF but does not show if the METADATA DIM is viewed with an XML browser However it is included in the META DATA DIM and can be found when the file is opened with an editor On the other hand the view angle is not included in the VOL_LIST PDF but is displayed with an XML browser applied to METADATA DIM e SPOT 4 5 imagery is usually delivered in the DIMAP format a tif file with the band sequence 3 2 1 4 NIR Red Green and 1 6 um The wavelength increasing sequence has to be created before offering the file to ATCOR Attention Old SPOT 2 imagery are usually distributed in the CAP format For this old format the SPOT leader file and the voldir pdf indicate L instrument looks to west and R instrument looks to east This is a header coding error it is just vice versa so interpret L east R west 9 6 3 ALOS AVNIR 2 The ALOS satellite has a sun synchronous orbit of 690 km and among other instruments carries the AVNIR 2 optical payload AVNIR 2 has four spectral channels blue green red NIR with a nadir resolution of 10 m The instrument has a 44 across track tilt capability Different metafiles are available one is in the DIMAP format It contains the relevant geometric and radiometric param eters The convention for the tilt and orientation angles is similar to SPOT compare Fig 9 6 i e a tilt left with
92. 77 Table 4 2 Sensor definition file instrument with thermal bands Any mid IR bands are skipped in the processing the surface temperature band itemp_band is appended after the reflective bands as the last channel of the _atm bsq output file 4 7 Spectral smile sensors Imaging systems can employ different techniques to record a scene the whiskbroom design uses a rotating or oscillating mirror to collect an image line in across track direction with one or a few detector elements per spectral band The forward direction is provided by the motion of the platform Secondly a pushbroom linear array can perform the same task without moving optical elements but the number of array lines each recording a certain spectral channel in the focal plane is limited The third imaging technique employs an area detector array where one direction collects the spatial information across track and the orthogonal direction covers the spectral di mension The advantage of the last two techniques is a longer pixel dwell time and a potentially improved signal to noise ratio SNR The drawback is a substantial increase in the spectral and radiometric characterization i e a change of the channel center wavelength across the columns of the array spectral smile spatial misregistration keystone and detector non uniformity problems 48 17 70 Typical representatives of the whiskbroom type are Landsat TM ETM HyMap AVIRIS and
93. 8 Generate Filter Files rsp Shape changes from near rectangular first bands to triangular last bands due to binning Select Type of Filter Function Red Gauss Butterworth order 1 slow drop off Butterworth order 3 between Gauss rectangular Butterworth order 4 close to rectangular Rectangular Triangular 0 46 0 48 0 50 0 52 0 54 quit Wavelength jem Figure 4 14 Supported analytical channel filter types The next two tables present examples of a sensor definition file for an instrument without thermal bands and with thermal bands respectively Line 1 is retained to be compatible with the airborne version of ATCOR Line 6 is a required dummy to be compatible with previous versions CHAPTER 4 WORKFLOW 40 5 0 6000 across track FOV degree pixels per line dummy to agree with airborne ATCOR 1 128 first last reflective band 0 35 2 55 um O first last mid IR band 2 6 7 1 wm O first last thermal band 7 1 14 um flag for tilt capability 1 yes 0 no required dummy O oo Table 4 1 Example of a sensor definition file no thermal bands 4 0 3000 cross track FOV degree pixels per line dummy to agree with airborne ATCOR 1 72 first last reflective band 0 35 2 55 um 73 73 first last mid IR band 2 6 7 1 um 74 79 first last thermal band 7 1 14 um 0 flag for tilt capability 1 yes 0 no 0 no gain settings required dummy 77 temperature band itemp_band
94. A dd a ES 78 Ba Blkyview Factor o esce atsana ma aop i a sa HOS a a Se ae eee a 78 543 Shadow Mask 264 04 a mola ess eR ee ee ee ee ee 79 DAA DEM Simoithing go deck Gk oe aa a Ae a ee we Re 8 80 Ope Mena Fale a a a hed p ed ee Re A we eae Re Bee ee Or E 82 DDL Resaniple a Spectr ciar Ree AG ee ee a oe ee 82 5 5 2 Spectral Polishing Statistical Filter gt s s ca 6046 cee RR ee ee 82 5 5 3 Spectral Polishing Radiometric Variation 0004 83 5 5 4 Spectral Smile Interpolation e 002020004 84 50 Memi Simulatie ca e eens a a i 4 A eda ee ARE ER eh hS 87 5 6 1 TOA At Sensor Radiance CUbe o e e e 87 5 6 2 At Sensor Apparent Reflectance o 87 5 6 3 Resample Image Cube e 88 oer Monu TOOTS a al a hod poa a a a ee a dd id A 89 Dol Solar Zem anid AMG e ak ek ee Pe ee ee A 89 5 7 2 SPECL Spectral Reflectance Classification 0 00004 90 5 7 3 SPECL for User Defined Sensors 2 2 00020004 90 5 7 4 Nadir normalization Wide FOV Imagery 90 5 7 5 Quick Topographic no alma Correction osos 455844448855 45 91 5 7 6 Add a Blue Spectral Channel e 92 D f 7 spectral Smile Detection ata mora aor ada Ee a aG 92 5 7 8 Spectral Calibration Atm Absorption Features 94 5 7 9 Calibration Coefficients with Regression soosoo a 96 5 7 10 Convert High Res Database New Solar Irradiance
95. ATCOR and re quires a minimum field of view of 20 Therefore it might also be of interest for some of the supported satellite sensors e g IRS 1C WiFS or MERIS INPUT IMAGE Ref lectance Radiance Vaata7 atcora2 doro_data mpo38 barL huep_berd_stn bog OUTPUT IMAGE nadir nernalizad Ydata atcor42 deno_data hynap9I berl hiyap_bari_atm_nadir beq EA wv Input Inage NOT Geocoded w Geocoded Input Image Sensor total field of view FOV degree 60 0 2 Global normalization surface cover independent wy Cover dependent nadir normalization classes bright veget nediun dark veget dry veget soil Bard selection y hot spot geometry across track angular sampling interval 1 degree A no hot spot geometry across track angular sampling interval 3 degree MANE ESE NP IA Quit Figure 5 54 Nadir normalization 5 7 5 Quick Topographic no atm Correction The quick topographic correction routine TOPOCOR tries to eliminate the slope aspect topo graphic effects and neglects the atmospheric influence The program runs very fast and the output image contains the modified digital numbers Processing with TOPOCOR can be done with or without display of images The DEM slope and aspect files have to be specified on a separate panel that pops up after the input image has been specified The topographic correction implemented here multiplies the DN value with a factor f that depends on the local solar zeni
96. Atmospheric Topographic Correction for Satellite Imagery ATCOR 2 3 User Guide Version 8 2 BETA February 2012 R Richter and D Schlapfer 1 DLR German Aerospace Center D 82234 Wessling Germany ReSe Applications Langeggweg 3 CH 9500 Wil SG Switzerland DLR IB 565 02 12 The cover image shows a true color subset of a Landsat 7 ETM scene with lake Constance at the Austrian German Swiss border top left acquired 2 June 2000 The top right image is the atmospherically corrected scene employing a haze removal over land and water The haze removal over water is one of the new features of the 2011 release The bottom part presents a zoomed view The water mask is derived from the DEM file of this scene because a unique water classification with exclusively spectral criteria is not possible in this scene A correct water land mask is critical for the quality of the separate haze removal algorithms over land and water ATCOR 2 3 User Guide Version 8 2 0 February 2012 Authors R Richter and D Schl pfer 1 DLR German Aerospace Center D 82234 Wessling Germany 2 ReSe Applications Langeggweg 3 CH 9500 Wil SG Switzerland All rights are with the authors of this manual Distribution ReSe Applications Schlapfer Langeggweg 3 CH 9500 Wil Switzerland Updates see ReSe download page www rese ch download The MODTRAN trademark is being used with the expressed permission of the owner the United States of Americ
97. CAL BACKGROUND 165 e medium snow ice probability coded 60 same as for low probability but with a more strin gent NDSI threshold of 0 6 This is the snow assignment in the hcw bsq file p blue gt 0 22 and NDSI gt 0 6 and DN blue lt Tsaturation 10 61 If no blue band exists a green band is used as a substitute If a green band and a SWIR2 band exist the rules are DN blue lt Tsaturation and p blue gt 0 22 and NDSI gt 0 6 or p green gt 0 22 and p SWIR2 p green lt 0 3 10 62 This is very similar to the snow assignment in the hcw bsq file except for the threshold for p SWIR2 p green e high snow ice probability coded 90 same as for medium probability but with a more stringent NDSI threshold of 0 7 p blue gt 0 22 and NDSI gt 0 7 and DN blue lt Tsaturation 10 63 If no blue band exists a green band is used as a substitute Again if a green band and a SWIR2 band exist the rules are DN blue lt Tsaturation and p blue gt 0 22 and NDSI gt 0 7 or p green gt 0 22 and p SWIR2 p green lt 0 2 10 64 10 4 Standard atmospheric conditions Standard conditions comprise scenes taken under a clear sky atmosphere This means the visibility aerosol optical thickness can be assumed as constant over a scene or it might vary within a certain range excluding haze and a visibility map can be calculated It also includes situations with constant or spa
98. DL command line then the ATCOR GUI selection panel pops up alternatively you may use the command restore atcor sav Disregard this panel and continue on the IDL command line with the name of the batch job module where all the input parameters have to be specified via key words Current batch programs are CHAPTER 6 BATCH PROCESSING REFERENCE 102 slopasp_batch input filename pixelsize 10 0 kernel 3 dem_unit 0 The filename should have the last four characters as _ele and the extension bsq Two output files slope and aspect are generated from the elevation file e g example DEM25m slp bsq and example _DEM25m_asp bsq The values are coded in de grees The keyword pizelsize is not required if this information is included in the map info of the ENVI header The keywords kernel and dem_unit can be omitted if the default values kernel 3 and dem_unit 0 are used The unit of pirelsize is meter For the elevation height unit three options exist dem_unit 0 height unit is meters 1 for dm 2 for cm Note Before running ATCOR with a DEM please check the results of the slope image We often encounter severe horizontal and vertical striping in the slope image in case of low quality DEMs or if coarse DEMs of 25 m have to be resampled to say 5 m Additional appropriate filtering is required in these cases A simple way might be to try a larger kernel size e g kernel 5 or kernel 7 A simp
99. Depending on the selected image processing option some additional panels may pop up Most of them are listed in chapter 5 7 4 but they are self explaining and will not be discussed here They also contain default settings which can be used in most cases When the main panel Fig 4 6 is left and the SPECTRA or IMAGE PROCESSING sections are entered all information is written to a processing initialization inn file e g image inn When reloading the input file this information is read from the inn file so a new specification of all processing parameters is not necessary Therefore this inn file can also be used for a batch processing see chapter 6 The remaining sub chapters of chapter 4 may be skipped during the first reading if the definition of user defined hyperspectral sensors is not relevant These sub chapters can be consulted if specific questions arise e g about batch job processing CHAPTER 4 WORKFLOW 34 Specify DEM Related Files Update DEM Path Path Vexport data data7 atcor2 3 deno_data tn freib_rugged Mandatory Files Elevation n brforest_30n_ele bsg DEM height z unit m w dm w cm Slope degree Jn biforest_30n_clp boa Aspect degree Jn biforest_20n_aop boa Optional Files Sky View Factor 2 tm_blforest_30m_sky bsq Cast Shadow 0 4 w Use pre calculated shadow file if existing Shadow map calculated on the fly requires more memory gt Check
100. ECTRA module has been accessed or after one of the image processing options has been selected Thus the GUI panel creates an inn file containing all input parameters The batch mode can be started after quitting the interactive session using the same IDL window It can also be started in a new IDL session after typing atcor on the IDL command line Then continue with gt atcor2_batch input datal examples example_image bsq case of flat terrain or gt atcor3_batch input datal examples example i mage bsq case of rugged terrain 2 At this stage all required input parameters are already available in the inn file in this specific case example_image inn The submitted job is a quasi batch job the corresponding IDL window is used for error and status messages and it may not be closed during the run time of the job A log file is created during processing e g example_image_atm log which contains information about the job status It contains three message levels I Info W Warning E Error followed by a two digit number between 0 and 99 and a space e g W19 followed by the appropriate information These three message levels can easily be parsed by a separate user program if desired Other information in the log file is marked with the hashmark symbol in the first column 100 CHAPTER 6 BATCH PROCESSING REFERENCE 101 In the tiling mode the user has to specify the number of tiles in
101. EORETICAL BACKGROUND 169 L Radiance L total blue band reflected radiance blue band 1 I 1 1 1 1 e E i ia ES 1 gt 1 N rd 1 e sa i x 1 1 A gt blue green red Figure 10 8 Rescaling of the path radiance with the blue and red band After subtraction of the reflected radiance from the total radiance in the blue band the remaining signal is the updated path radiance in the blue band The path radiance of bands in the blue to red region is then rescaled with interpolation average clear atmospheric conditions visibility VIS 23 km to calculate the surface reflectance in the red and NIR bands which is appropriate for situations of clear atmospheres VIS 15 40 km The second step derives a mask of dark vegetation pixels using the ratio vegetation index rvi of the red and near infrared surface reflectance rvi Pnir Prea and multiple reflectance thresholds e The mask pixels have to fulfill rvi gt 3 and Pnir gt 0 10 and Pnir lt 0 25 and Preg lt 0 04 Water pixels are automatically excluded from this mask because of the Pair gt 0 10 condition and soil pixels are excluded with the combination of all four conditions If the percentage of reference pixels is smaller than 2 of the scene the search is iterated with VIS 60 km covering the very clear conditions of visibility 40 100 km Again if the percentage is smaller than 2 the search is iterated with VIS 10 km to cover
102. ERENCE AND SENSOR SPECIFICS 122 the tilt angles 0 30 in steps of 10 is available upon request It is intended for future hyper spectral space sensors such as EnMAP A separate CHRIS Proba database is included in the distribution containing calculations for the tilt angles 0 35 and 55 for seven equidistantly spaced relative azimuth angles 0 30 180 CHRIS data acquisition is usually close within 2 to this set of tilt angles and interpolation is automatically performed While the standard database is named atm_database the CHRIS database is named atm_database_chris and it is automatically accessed if the letters chris or CHRIS are part of the user defined sensor name e g chris mode5 While the standard database uses a 0 4 nm wavelength grid the CHRIS database employs a 1 nm grid 9 1 2 Thermal region In the thermal region from 7 2 14 9 um a 1 cm grid is used i e a 10 nm wavelength spacing at 10 um using the Isaacs s 2 stream algorithm including the correlated k algorithm The Isaacs s algorithm is much faster than DISORT and yields the same results for our cases in the thermal region The spectral resolution is twice the sampling distance and a triangular weight function with 2 cm is employed So the thermal region uses an equidistant wavenumber grid In addition all files are calculated for view or scan angles from 0 nadir to 40 off nadir with a 5 inc
103. Gauss y 6 3 Rectangular wv Triangular wv 8 Shape changes from near rectangular first bands to triangular last bands due to binning Generate Filter Files rsp QUIT Figure 5 13 Spectral Filter Creation 5 2 3 BBCALC Blackbody Function This routine calculates the blackbody function as described in section 10 1 4 weighted by the spec tral response curve of the thermal band used for the temperature retrieval compare Fig 5 14 Inputs Spectral response file Select the rsp file of the spectral band in the thermal IR to be used for temperature retrieval Exponential Fit Limits The lower and the higher limit of the temperatures for which a fitting function should be created Unit of radiance output Select the unit either per micron or without normalization Outputs A file x_hs bbfit is created containing the fitting parameters 5 2 4 RESLUT Resample Atm LUTS from Database The monochromatic database of atmospheric LUTs has to be resampled for the specific channel filter functions of each sensor Details are given in chapters 4 6 1 to 9 1 4 Figure 5 15 repeats the panels of the LUT resampling program RESLUT The resampling has to be done separately for the CHAPTER 5 DESCRIPTION OF MODULES 60 Figure 5 14 Black body function calculation panel reflective and thermal region Only the required range of flight altitudes and aerosol types should be selected Figure 5 15 Panels of RESLU
104. In addition panchromatic data with a resolution of about 0 5 m is available The radiometric encoding is 11 bits per pixel The metafile for each scene contains the radiometric offset and gain values These values are given in the same unit as used by ATCOR i e mWcm sr 1 umt so they can be directly taken i e c Gain 9 12 The Offset co is usually zero 9 6 12 WorldView 2 WorldView 2 provides optical data with 8 multispectral channels in the VNIR region 428 923 nm at a spatial resolution of 1 8 m nadir with a dynamic range of 11 bits per pixel The instrument has selectable radiometric gain factors absCalFactor specified in the metafile IMD The offset co is zero for all channels and the gain c for ATCOR has to be calculated as c 0 1 AbsCalF actor FWHM 9 13 where FWHM is the effective bandwidth effectiveBandwidth in um as specified in the metafile Although the bandwidth is constant per channel the gain c might have to be updated because the absCalFactor can vary from scene to scene Additionally panchromatic images with a 0 5 m resolution are available Chapter 10 Theoretical Background Standard books on optical remote sensing contain an extensive presentation on sensors spectral signatures and atmospheric effects where the interested reader is referred to Slater 1980 80 Asrar 1989 2 Schowengerdt 2007 78 This chapter contains a description of the concepts and equation
105. KFLOW 38 alfalfa Reflectance Reflectance 0 5 1 0 1 5 2 0 0 5 1 0 15 2 0 Wavelength zm Wavelength em 40 D 52 30 a u a f Q E agricultural_soil 5 20 aa pi Pi E E 10 D 0 5 1 0 1 5 2 0 0 5 1 0 1 5 2 0 Wavelength m Wavelength m Figure 4 12 Template reference spectra from the spec_lib library The dashed spectra are resampled with the Landsat 5 TM filter curves A sensor definition file must be specified Just copy any of the existing files e g sen sor_chris_m1 dat and modify the appropriate lines see the next table A wavelength file wvl has to be specified It is a simple ASCII file with three columns band number center wavelength and bandwidth compare Fig 5 11 Center wavelength and bandwidth may be given in the nm or wm unit The first line may contain an optional header with text This wavelength file will be used to create the spectral response function for each band as a numerical table the band rsp files compare Fig 5 11 Eight analytical filter shapes can be selected from the top level graphical interface Fig 4 1 when pressing the New Sensor then Create Channel Filter Files button Then the menu of Fig 4 14 will pop up and one of these 8 filter shapes can be selected Filter numbers 1 to 4 are of the Butterworth type the slow drop off for the Butterworth order 1 is truncated at the 0 05 response and set to zero The filter type 9 parameter filter_
106. Landsat 5 TM Black Forest 12 Sept 1985 solar zen 49 0 Print Setup A solar azim 146 deg Print 0 it 0 ile type ENVI Standard data type 1 interleave bsq sensor type Unknown byte order 1 band names band 1 band 2 band 3 band 4 band 5 wavelength 0 486 0 570 0 661 0 838 1 65 a El Figure 5 5 Simple text editor to edit plain text ASCII files 5 1 3 Select Input Image This function allows to select the basis input image i e the band sequential uncorrected image data in ENVI format It is useful to define the starting point including the default paths of the images for further processing 5 1 4 Import A small number of standard formats is supported for importing data layers to be used with ATCOR ENVI BIP Image Imports and transforms an ENVI image in BIP band interleaved by pixel format to the ATCOR default BSQ band sequential format ENVI BIL Image Imports and transforms an ENVI image in BIL band interleaved by line format to the ATCOR default BSQ band sequential format Erdas Imagine Imports an uncompressed image in ERDAS Imagine format to ENVI format specifically suited for DEM data import Hyperion Raw Image Imports a Hyperion image in ENVI format 242 spectral bands for use with ATCOR The number of bands is reduced to 167 according to the sensor definition provided with the software CHAPTER 5 DESCRIPTION OF MODULES 53 5 1 5 Plot Sensor Response In the pa
107. M elevation slope and aspect then the DEM files are taken and the TOA calculation is performed for a rugged terrain If the keyword elev is specified the simulation is always performed for a flat terrain regardless of any possible DEM file names in the inn file e sz 35 5 an example of a solar zenith angle of 35 5 e vis 25 an example of a visibility of 25 km e pixelsz 4 5 an example of a pixelsize of 4 5 m CHAPTER 8 SENSOR SIMULATION OF HYPER MULTISPECTRAL IMAGERY 120 e adjrange 500 an example of an adjacency range of 500 m e scalef 10 000 scale factor for the TOA radiance The default is scalef 1 0 which provides the output file as float data of TOA radiance in units of nWcem srt umt If scalef gt 1 e g scalef 10 000 the output TOA radiance is stored as 16 bit unsigned integer multiplied with the scale factor The advantage is a smaller output file compared to the 32 bit float the drawback is that radiances will be truncated at 65 000 which might happen for bright surfaces e g snow vegetation in the NIR with scalef 10 000 see Figure 8 4 Therefore the easiest way to avoid scale problems is to use the default scalef 1 0 and have a float radiance output cube sz 30 deg albedo 0 9 0 3 0 05 19 000 1 000 9 190 Radiance mW cm sr jam 9 010 9 001 0 5 1 0 1 5 2 0 25 Wavelength am Figure 8 4 TOA radiances for three albedos and a solar zenith angle of 80 MODTRAN calculation
108. RIPTION OF MODULES 96 5 7 9 Calibration Coefficients with Regression This routine employs the rdn files obtained during the single target calibration the c1 option of ATCOR s calibration module to create a calibration file by rlinear regression 00 A CAL_REGRESS Calibration Coefficients calculated with Regression Version 1 0 The rdn files have to be generated with the Inflight Calibration Module of ATCOR using the single target option cl calculation on a per target basis Number of calibration targets rdn files w3 4 v5 v y v8 v9 First rdn radiance digital number file Ydata cal1b1 r n Output calibration file cal data regress4 cal Calculate Radiometric Calibration Figure 5 58 CAL REGRESS radiometric calibration with more than one target So for n gt 2 the single target calibration is to be performed n times and the resulting rdn files radiance versus digital number are offered as input to the cal_regress program Inputs Number of calibration targets A maximum of 9 calibration targets may be selected The files rdn should having been calculated beforehand and the need to be calculated consecutively e g calibl rdn calib2 rdn First rdn file Name of the first rdn file of the series to be used for regression Output Name Name of the calibration output to be created Output The output of this program is an ASCII fi
109. Solar Region Standard books on optical remote sensing contain an extensive presentation on sensors spectral signatures and atmospheric effects where the interested reader is referred to Slater 1980 80 Asrar 1989 2 Schowengert 1997 78 This chapter describes the basic concept of atmospheric correction Only a few simple equations 2 1 2 21 are required to understand the key issues We start with the radiation components and the relationship between the at sensor radiance and the digital number or grey level of a pixel Then we are already able to draw some important conclusions about the radiometric calibration We continue with some remarks on how to select atmospheric parameters Next is a short discussion about the thermal spectral region The remaining sections present the topics of BRDF correction spectral radiometric calibration and de shadowing For a discussion of the haze removal method the reader is referred to chapter 10 5 2 Two often used parameters for the description of the atmosphere are visibility and optical thick ness Visibility and optical thickness The visibility horizontal meteorological range is approximately the maximum horizontal distance a human eye can recognize a dark object against a bright sky The exact definition is given by the Koschmieder equation wen La LD 8eP 3 O09 3 2 1 where is the extinction coefficient unit km at 550 nm The term 0 02 in this equation
110. T for resampling the atmospheric LUTs CHAPTER 5 DESCRIPTION OF MODULES 61 5 3 Menu Atm Correction The menu 4tm Correction contains the main processing modules of ATCOR i e the panels for ATCOR2 and ATCOR3 in both used defined and pre defined usually multispectral and hy perspectral modes In this section the main panels are first shortly described Thereafter the subroutines SPECTRA and IFCALI and all panels related to them are explained O00 X Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Licensed for Danie ATCOR2 multispectral sensors flat terrain fo c DLR 2011 ATCOR3 multispectral sensors rugged terrain Help ATCOR2 hyperspectral sensors flat terrain ATCOR3 hyperspectral sensors rugged terrain Start ATCOR Process Tiled from inn Figure 5 16 The Atm Correction Menu 5 3 1 The ATCOR main panel Figure 5 17 top shows the input parameters required for ATCOR The lower part of the panel contains buttons for selecting SPECTRA determining the aerosol type employing inflight radio metric CALIBRATION and starting the image processing The processing options are shown in the separate panel as described in section 5 3 11 The trivial panels e g band selection spatial subimage etc will not be shown here The panels should be filled or clicked in the top down direction The message widget at the bottom will display hints warning
111. The exact choice of R is not critical since the adjacency influence is a second order effect Instead of the range independent weighting in eq 10 9 a range dependent function can be selected with an exponential decrease of the weighting coefficients 59 The range dependent case requires more execution time of course Except for special geometries the difference between both approaches is small because the average reflectance in a large neighborhood usually does not vary much and the influence is a second order effect p x y p x y alo pla y 10 10 The function q indicates the strength of the adjacency effect It is the ratio of the diffuse to direct ground to sensor transmittance The range dependent version of eq 10 10 is R Plz y Ple y af p x y Alreap r r ar 10 11 0 CHAPTER 10 THEORETICAL BACKGROUND 150 Here R is the range where the intensity of the adjacency effect has dropped to the 10 level i e r R 2 3x rs where r is a scale range typically r 0 2 0 4 km R 0 5 1 km p r is the reflectance at range r from the x y position and A r is the area of a circular zone from r to r dr Now we approximate the circular regions by square regions to obtain the discrete version of eq 10 11 with exponentially decreasing weighting coefficients w NR Plz y p x y alo a y gt Ziwi 10 12 i 1 1 Vi Wi ap Wi and W J A r exp r dr I 2r exp r dr 10 13
112. VI format is a raw binary file accompanied by an ASCII header hdr in ATCOR it should be stored in band sequential order CHAPTER 5 DESCRIPTION OF MODULES 50 File Display ENVI Image Displays an additional ENVI image in a new window sorry no link ing available File Show ENVI Header Displays the ENVI header of the current image in a new editable window This allows to make changes to the ENVI header Note that the file needs to be loaded from scratch if changes have been made File Band Selection Allows to select a new combination of spectral bands and updates the dis play File Display TIFF Image Loads a multi band TIFF image in a new window File Close Closes the window Edit Equalize Image Performs a histogram equalization on the three bands Edit Scale Image Applies standard linear scaling on the imagery on 5 levels Edit Scale Zoom Applies standard linear scaling on the imagery on 5 levels based on the statistics of the Zoom Window Edit No Scaling Reverts to unscaled display of the image Edit Scale to Range Scales a single band image linearly to a range entered as lower and upper limit only applicable in single band displays Edit Load Color Table Applies standard linear scaling on the imagery on 5 levels Profile Horizontal Opens a window for a horizontal profile through the image of the first band only The profile is updated for the cursor location in the zoom window whenever the zoom window is clicked Profi
113. a as represented by the United States Air Force Contents 1 Introduction 2 Basic Concepts in the Solar Region 21 Radiation components s ere aem 4 eee A a A Be eee 22 BRIDE Correction ok ao he a a a we we ee a ee Ae we 2 9 spectral calibration 2 s e opaa a 4 4 045 0 4b eae oe ok Se ae ee ae 24 Inflight radiometric calibration sa e 24505 45 be eee Se ee ES 2 5 De shadowing 066 6 bbe ee ee we 3 Basic Concepts in the Thermal Region 4 Workflow 4 1 Menus Overview 42 Prst stepe with ATCOOR oca 2544 8k doe eA RY Se ee Be wo ee a dd Durvey of processing StEpS socs aie goka oe Roe eke Ra eee ee ee Be eR 4 4 Directory structure of ATCOR oori ira ee ee oe Oe oe woe Ee a ed 4 5 Convention for lille names s ces ces ee RY ee Ae ee ke 4 6 User defined hyperspectral sensors a 4 6 1 Definition of a new sensor gt saa scoas waaau watea Re aN 4 7 Spectral smile sensors c ssa caa aoa cgp ace ara a g ae aa de Haze cloud water Mep se siasa kiadas ee E AR a a aa A 4 9 Processing of multiband thermal data e 4 10 External water vapor MAD lt i s ssa damog sa eu ER Re Or Eu eG a 5 Description of Modules Bel Memu File osie ad gan See a AA A a ee o e a 51 1 Display ENVI Pile oe css ce aog a aace a a a e ee a ae ae pLa Shopy TESIS ric ec h nakaa aina e gee we amp amp Go A R Ddy Select put Image gt i oras roca eA A ew e A 2 mlk Dapo se dama ae AE A AAA 5 1 5 Plot Sensor Response os ses cerrada e ER i
114. a file mage1_water_map bsq or mage1_hcw bsq exist in the same folder then the flag iwat_shd is ignored because an external water map always has the first priority ksolflux 0 file with value added channels not calculated ksolflux 1 value added channels are calculated flx file ishadow 0 and fshd empty string no DEM cast shadow map is used ishadow 0 and fshd valid file name pre calculated DEM cast shadow file is used ishadow 1 DEM cast shadow mask is calculated on the fly The pre calculated map avoids repeated on the fly calculations icl shadow 0 no cloud building shadow correction icl shadow gt 0 cloud building shadow correction is performed 1 de shadowed output as DN image corresponding to input DN scene 2 de shadowed output as surface reflectance image 3 de shadowed output as surface reflectance and DN image line 21 0 0 5 0 5 itriang ratio_red_swir ratio_blu_red itriang 0 average vis index of reference areas is employed for non reference pixels itriang 1 triangular interpolation of visibility index of reference areas ratio_red_swir ratio of surface reflectance of red to 2 2 um band for the reference pixels If no 2 2 um band exists but a 1 6 wm band the ratio holds for the red to 1 6 wm band If only VNIR bands exist 400 1000 nm the ratio holds for the red to NIR band ratio_blu red ratio of surface reflectance of blue band to red band for
115. ability water pixels than medium probability pixels eq 10 56 If a SWIR1 band exists the apparent NIR reflectance thresh old is relaxed first line of eq 10 57 because of the additional SWIR1 surface reflectance threshold p NIR lt 0 05 and p SWIR1 lt 0 03 or Pp NIR lt 0 03 or p SWIR1 lt 0 03 10 57 The relationships on the second line of eq 10 57 assign the water probability based on the smaller reflectance value in the NIR or SWIR1 high water probability coded 90 same as for medium probability but with lower NIR SWIR1 thresholds p NIR lt 0 03 no SWIR1 band P NIR lt 003 or p SWIR1 lt 0 02 10 58 Note the default threshold Twater swrri is 0 03 or 3 in the reflectance percent unit defined in the preference parameter file Snow ice probability The criteria for the snow ice class are described in the previous section As mentioned before if pixels are saturated in the blue green spectral bands they are counted as cloud unless the NDSI gt 0 7 The following probability rules are employed for snow e low snow ice probability coded 30 p blue gt 0 22 and NDSI gt 0 4 and DN blue lt Tsaturation 10 59 If no blue band exists a green band is used as a substitute If a green band and a SWIR2 band exist the rules are DN blue lt Tsaturation and p blue gt 0 22 NDSI gt 0 4 or p green gt 0 22 NDSI gt 0 25 p SWIR2 p green lt 0 5 10 60 CHAPTER 10 THEORETI
116. ad to problems for the haze removal as well as the cloud building shadow removal i e water is erroneously counted as land and included in the land mask Enable the output of a haze cloud water map compare Fig 4 15 to check the water mask Read chapter 4 8 There are two possibilities to define the water mask If the average ground elevation of the scene is below 1 2 km the default option for cal culating the water mask is the negative gradient criterion for the apparent reflectance in the VNIR region see eq 9 3 This yields the best water map in most cases However in some cases urban areas with shadows cast by buildings the NIR SWIR1 water surface reflectance thresholds see chapter 9 3 yield better results If the average scene eleva tion is below 1 2 km one has to define negative reflectance values to both NIR SWIR1 thresholds to overrule the default negative gradient criterion In the latter case an in crease decrease of the absolute value of the thresholds will increase decrease the number of water pixels Attention concerning de shadowing if this option is set and if the negative gradient cri terion for the water mask is enabled and if the percentage of water pixels is greater than 15 of the scene then the de shadowing algorithm assumes that the large percentage of water pixels is caused by the inadequate gradient criterion Then in the de shadowing part the NIR SWIR1 water surface reflectance thresholds are used wh
117. ain view factor range 0 1 The solar and DEM geometry is shown in figure 10 4 as well as the three solar radiation components taken into account for rugged terrain direct and circumsolar irradiance and diffuse hemispherical sky flux It can be shown that these three components are equivalent to the direct and diffuse solar flux components in flat terrain In case of a shadow pixel the direct and circumsolar components are set to zero i e the binary factor b 0 The next step iterates eq 10 15 averaging the reflected terrain radiation over a square box of 0 5 x 0 5 km If equation 10 15 is used with E E then three iterations are usually sufficient to be independent of the start value of the terrain reflectance 59 However for highly reflective surfaces e g snow and high terrain view factors more than three iterations are necessary and i a faster convergence of A ain can be achieved with a geometric series for the terrain reflected CHAPTER 10 THEORETICAL BACKGROUND 152 radiation E as proposed in 79 i pi Vierrain E Ey terrain ere 10 16 c Pterrain Vierrain The next steps include the adjacency correction eq 10 9 10 10 and the spherical albedo effect eq 10 14 Geometry of solar illumination cos 3 cos 8 cos 0 sin 0 sint cost vis sur Y 3 Direct irradiance y Circumsolar diffuse imadiance _ lsotropic diffuse lt fux Direct and diffuse rad
118. al description because they contain explanations in the panels themselves However the next section guides the ATCOR newcomer during the atmospheric correction of a sample scene The functions in the File menu allow the display of an image file the on screen display of calibration files sensor response curves etc see Fig 4 2 More details about this menu are given in chapter 5 1 File New Sensor Atm Correction Topographic Filter Simulation Tools Help Licensed for DLR Version 8 0 c DLR 2012 Figure 4 1 Top level graphical interface of ATCOR The New Sensor menu of Fig 4 1 is available if the module for hyperspectral or user defined sensors is licensed It contains routines to create spectral filter curves rectangular Gaussian etc from a 3 column ASCII file band number center wavelength bandwidth one line per channel provided by the user calculates atmospheric look up tables LUTs for new sensors and computes the radiance temperature functions for thermal bands see Fig 4 3 and chapter 5 2 The menu Atm Correction gives access to the ATCOR core processes for atmospheric correction in flat and rugged terrain For both multispectral and hyperspectral instruments It also gives access to the tiled processing It is further described in section 4 2 below and chapter 5 3 29 CHAPTER 4 WORKFLOW 30 AAA X Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Help Display
119. ames The names for the output files are entered automatically and can t be changed as ATCOR asks these files to be named exactly according to the conventions Kernel Size Box Size of the kernel in number of pixels the slope and aspect is calculated as gradient of the pixels at the edges of this box the default value is 3 pixels i e direct neighbours of center pixel DEM resolution This is the pixel size a default of 30m is assumed This needs to be entered manually DEM height unit The unit of the values in the DEM usually a DEM is stored in meters but sometimes an integer DEM is stored as dm or cm data in order to preserve disk space Outputs The two files of slope and aspect are created same size as DEM integer data 5 4 2 Skyview Factor The sky view factor of a DEM is calculated with a ray tracing program and ranges from vsky 0 to 1 with 1 indicating a full hemispherical view Data in the sky view file are scaled from 0 to 100 and coded as byte The sky view factor determines the fraction of the hemispherical diffuse CHAPTER 5 DESCRIPTION OF MODULES 79 DO Slope and Aspect Calculation V 2 0 DEM File may have 16 or 32 bit integer or float data Input DEM FILE Ysrc idl atcor atcor_23 deno_data tn rueged tn_blforest_ele bs3 QUIT SLOPE File Tsrc idl atcor atcor_23 deno_data tn rusged tn_blforest_sIp bog I OVERWRITE ASPECT File Ysrc idl atcor atcor_23 deno_data tn ruaged tn_blforest_asp bo3 Kernel size box for
120. and the terminology The full set of equations is docu mented here as implemented in ATCOR We start with the radiative transfer equation in the solar spectral region 0 4 2 5 um for a flat terrain under clear sky conditions First the equation for an infinite plane of uniform reflectance is presented Then the case of a small uniform surface embedded in a large homogeneous background of different reflectance is discussed We continue with the rugged terrain and finally discuss the equations for the thermal spectral region 8 14 jum 10 1 1 Solar spectral region For a cloud free sky and a uniform ground of reflectance p the radiance signal received at the sensor consists of scattered solar radiation and ground reflected radiation The scattered radiation component is also called path radiance It depends on the solar and viewing geometry as sketched 146 CHAPTER 10 THEORETICAL BACKGROUND 147 pme Figure 10 1 Radiation components illumination and viewing geometry in Fig 10 1 In case of a flat terrain the at sensor radiance L can be written as Asrar 1989 chapter 9 L LO 85 0 70 O 10 1 L at sensor radiance for surface reflectance p Lp path radiance Ta total ground to sensor atmospheric transmittance sum of direct Tq and diffuse Tq f transmittance Eg global flux on a horizontal surface sum of direct Eqir and diffuse Equip flux E 0 is calculated for a ground surface with p 0 Pr large scale
121. annel k L k co k ci k Figure 10 22 shows an example of de shadowing More images with the results of the de shadowing DN k 10 107 CHAPTER 10 THEORETICAL BACKGROUND 189 method can be found on ATCOR s web page http www rese ch or http www op dlr de atcor Figure 10 22 De shadowing of a Landsat 7 ETM scene Subset of a Landsat 7 ETM scene from Kenia 10 April 2001 Color coding RGB bands 4 2 1 830 560 480 nm Left original scene right after de shadowing 10 6 Summary of atmospheric correction steps Although the case of a flat terrain could be treated as a special case of a rugged terrain with the same elevation everywhere this is not an efficient solution because the rugged terrain algorithm runs 3 to 4 times slower than the flat terrain code Therefore the coding is done in separate modules as discussed below 10 6 1 Algorithm for flat terrain The complete sequence of processing for sensors with water vapor bands and a short wave IR band 1 6 or 2 2 um region consists of the following steps masking of haze cloud water and clear pixels haze removal de shadowing masking of reference pixels calculation of visibility or optical thickness for reference pixels The optical thickness for the remaining pixels can be defined as the average of the reference pixels or a spatial triangular interpolation is employed to fill the gaps Finally a moving low pass window with a box size of 3km x 3
122. annels example image_atm tif gt An IDL routine called write_atcor3_inn_file is available to users who want to generate the inn file without the ATCOR GUI Note On the IDL command line the command atcor has to be typed first to load the atcor sav file Then the atcor2_tile or atcor3_tile commands will execute the tile processing A simple trick can be used to start the atcor_tile programs directly on the IDL command line without having to type atcor first just copy the atcor sav file to atcor2_tile sav and atcor3_tile sav The same can be done for atcor2_batch sav and atcor3_batch sav For the Linux Unix operation systems a symbolic link is sufficient e g In s atcor sav atcor2_batch sav For Linux Unix users with a full IDL license a batch job can be started directly from the shell e g idl e atcor2_batch input export data data7 demo_data tm_essen bsq 6 2 Batch modules keyword driven modules Most of the modules are available in both modes interactive and batch If the atcor sav file is copied to atcor2_batch sav and atcor3_batch sav a batch job can be started immediately from the IDL command line otherwise atcor has to be typed first Hereafter a description of the batch modules and keyword driven modules is given In order to make the batch options available you first have to type atcor on the I
123. antial amount of haze and shadow areas Although bidirectional surface reflectance effects are independent of the atmospheric conditions the subject is included here because the isotropic reflector is used for the standard conditions We present some empirical methods of BRDF correc tion in flat and rugged terrain The non standard atmospheric conditions treat the haze removal and de shadowing employing spectral and statistical algorithms 10 5 1 Empirical methods for BRDF correction ATCOR offers two different methods of correcting BRDF effects The first method is mainly in tended for flat terrain and normalizes the off nadir reflectance values to the corresponding nadir values The second method is exclusively dedicated to rugged terrain imagery The reflectance values of areas with low local sun elevation angles i e large local solar zenith angles are often overcorrected by the assumption of isotropically reflecting surfaces and the method reduces these high overcorrected values depending on the illumination and or viewing angles In some cases of rugged terrain imagery it is useful to apply both methods of empirical BRDF correction 1 Nadir normalization method A simple algorithm was implemented as part of the ATCOR package to normalize the scan angle dependent brightness values to the nadir value It is recommended to apply the method to imagery after atmospheric correction i e to reflectance data However if only the across track
124. ast row repeats the concrete case for Rsolar 800 1 0 36 512 Rn Rsolar Ratm Rsurface 512 100 412 Wm a realistic reduced Rn value compared to the asphalt where Ey 800 Rsolar 800 x 1 0 12 700 Ra 700 100 600 Wm surface peso Psso NDVI G veg H veg LE veg G urb H urb LE urb full veget 0 05 0 40 0 78 77 87 435 partial veget 0 10 0 20 0 33 185 76 338 dark asphalt 0 11 0 13 0 09 228 50 322 240 306 54 bright concrete 0 35 0 40 0 07 222 48 330 240 306 5A bright concrete 0 35 0 40 0 07 164 210 37 Table 7 1 Heat fluxes for the vegetation and urban model All fluxes in Wm All radiation and heat fluxes are calculated in units of Wm They represent instantaneous flux values For applications where daily 24 h LE values are required the following equation can be used for unit conversion cm 1 LE Wm 7 27 T 786 LE We 7 27 The latent heat flux LE is frequently called evapotranspiration ET Although LE and ET are used interchangeably the unit cm day or mm day is mostly employed for ET For water surfaces the distribution of net radiation into G LE and H is difficult to determine because it depends on several other parameters Therefore G and H are set to zero here and so LE equals Rp LE Spatial maps files of air temperature and ai
125. at file see chapter 9 3 If the file name of the imagery is mage bsq the corresponding map is named image_out_hcw bsq It is a 1 channel false color coded ENVI file In principle if a certain mask of image_out_hcw bsq say haze pixels contains artifacts it may be edited and if the edited file is named image_hcw bsq it will automatically be used for the ATCOR processing This means ATCOR can repeat the processing with an improved edited haze mask The file mage_hcw bsq can also be provided by an external ATCOR independent source In any case if this files exists ATCOR will skip its internal calculations of these masks and use the pre calculated map The haze cloud water file contains the following classes see Table 4 4 e land e water e boundary layer haze two classes thin to medium haze and medium to thick haze e cirrus clouds three classes thin medium and thick cirrus provided a narrow channel around 1 38 um exists CHAPTER 4 WORKFLOW 43 image l I I 1 Edit File I I Figure 4 15 Optional haze cloud water output file e cloud over land cloud over water e snow requires a 1 6 um channel e saturated pixels using the criterion T gt 0 9 DN max where T is a threshold set at 0 9 times the maximum digital number This criterion is only used for 8 bit and 16 bit signed or unsigned data no threshold is defined for 32 bit integer or float data As an example fo
126. ated to all other bands within the same detector or spectrometer unit Optionally the polynomial coefficients can be set to zero in atmospheric absorption regions to expedite the processing Once the coefficients have been determined they are converted into the required file format and are placed in the respective sensor folder for a subsequent fully automatic radiometric and atmospheric processing Fig 5 56 shows the panel of the smile detection module The input files scene high resolution atmospheric database sensor spectral response and the output file has to be specified in the upper part of the panel Then the smile detection resolution search range and band range can be specified Next are parameters pertaining to the scene visibility solar zenith angle and ground elevation Up to 12 atmospheric or Fraunhofer feature wavelengths can be selected The Inter polation Type refers to the interpolation of the smile polynomial between feature wavelengths Extrapolation can be done by extrapolating the trend repeating the values of the last band calcu lated or by setting the polynomial to zero at the borders of the spectral range of the instrument The module will perform the calculation when pressing the Run button Results of the smile detection can be viewed with Plot Smile Here the relative smile curves with respect to the center of the detector array can be plotted or the absolute wavelength smile curve
127. ay also include soil areas For narrow band hyperspectral sensors a band close to 2 13 um is used instead of a 2 20 um band The red band is then used to calculate the visibility compare figure 10 6 as the intersection of the measured radiance with the simulated visibility dependent at sensor radiance curve Since the same visibility is employed for the blue spectral band this provides an opportunity to adjust the spectral behavior of the path radiance which is essentially the aerosol path radiance since the Rayleigh path radiance is known in the blue spectral region d PE Lilue B ToluePblue Eg blue T 10 67 The question of an automatic aerosol type calculation is addressed next CHAPTER 10 THEORETICAL BACKGROUND 168 surface reflectance A A pad gt 0 48 0 66 0 80 1 6 22 Figure 10 7 Correlation of reflectance in different spectral regions Aerosol type estimation After calculation of the scene path radiance in the blue and red region as total minus reflected radiance using the average values obtained for the dark reference pixels the ratio of Ly blue scene to Ly red scene can be compared to the corresponding ratio for the MODTRAN standard aerosols rural urban maritime desert ie Lp blue scene Ly red scene P L blue MODTRAN L red MODTRAN 10 68 The aerosol type for which the double ratio dp is closest to 1 is the best approximation for the scene It approximates the correspon
128. building shadows The scene covers part of the central area of Munich It was recorded by the Ikonos 2 sensor 17 Sept 2003 The solar zenith and azimuth angles are 46 3 and 167 3 respectively After shadow removal the scene displays a much lower contrast of course but many details can be seen that are hidden in the uncorrected scene see the zoom images of figure 2 10 The central zoom image represents the shadow map scaled between 0 and 1000 The darker the area the lower the fractional direct solar illumination i e the higher the amount of shadow Some artifacts can also be observed in Figure 2 9 e g the Isar river at the bottom right escaped the water mask entered the shadow mask and is therefore overcorrected Figure 2 9 De shadowing of an Ikonos image of Munich European Space Imaging GmbH 2003 Color coding RGB bands 4 3 2 800 660 550 nm Left original right de shadowed image The proposed de shadowing technique works for multispectral and hyperspectral imagery over land CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 26 Figure 2 10 Zoomed view of central part of Figure 2 9 Courtesy of European Space Imaging Color coding RGB bands 4 3 2 Left original center shadow map right de shadowed image acquired by satellite airborne sensors The method requires a channel in the visible and at least one spectral band in the near infrared 0 8 1 um region but performs much better if bands in the short wave
129. c coefficients for ATCOR s cal file have to be calculated as e co Lmin and c Lmax Lmin 255 9 6 7 IRS P6 The IRS P6 platform carries three optical sensors the AWiFS advanced wide field of view sensor the Liss 3 and the Liss 4 AWIiFS 60 m resolution and Liss 3 20 m have the same spectral bands green red NIR and SWIR1 at 1 6 wm the LISS 4 red band serves as the high resolution camera 5 m Similar to the IRS 1C 1D the radiometric calibration coefficients are included in the meta file the bias B Lmin and gain G Limas are specified in the unit mWem sr ywm and the nominal value of Lmin is zero For each sensor AWiFS Liss 3 Liss 4 the calibration coefficients seem to be constant with time i e independent of the scene based on laboratory calibration The radiometric coefficients for ATCOR s cal file have to be calculated as e co Lmin and c Lmax Lmin b where b 1023 for AWiFS 10 bit data encoding and b 255 for Liss 3 and Liss 4 8 bit encoding The analysis of a couple of scenes showed that a non zero bias cp is required to obtain reasonable surface reflectance spectra Therefore typical average bias values are included in the standard cal file for each sensor A fine tuning of the calibration coefficients may be necessary to obtain better agreement between scene derived surface reflectance spectra and library or ground measured spectra 9 6 8 ASTER ASTER
130. c sensor with m 1 band The program supports the four cases of resampling mentioned above i e solar or thermal at sensor radiance surface reflectance or emissivity The processing of hyperspectral thermal data is currently not supported by the satellite version of ATCOR because of the lack of commercial sensors with these channels However this part is already implemented in the airborne ATCOR A detailed description of the keywords of program toarad follows CHAPTER 8 SENSOR SIMULATION OF HYPER MULTISPECTRAL IMAGERY 118 NER noise ms sensor hs sensor Radiance Resampling At Sensor Radiance Ca atmospheric parameters surface oo reflectance Reflectance Resampling Figure 8 2 Sensor simulation in the solar region Keywords for the batch program toarad On the IDL command line program names can be written in lower case letters so as an example toarad instead of TOARAD is used synonymously in this context If toarad is submitted as a batch job the following keywords can be specified e toarad input filename pixelsize pixelsize sz solar_zenith atmfile atmfile elev elevation vis visibility adjrange adjrange scalef scalef The input file name must include the path and the keywords in brackets indicate optional pa rameters If a keyword is set it will overwrite the corresponding parameter from the inn file compare chapters 6 2 and 9
131. cally it can be zero i e a completely shadowed pixel receiving only diffuse solar illumination However a too low estimate close to zero will boost the surface reflectance especially for channels in the 1 5 2 5 um region eq 10 104 since the diffuse solar radiation term Ei is very small Therefore small positive values of 7 are recommended The range of Piin is typically from 0 05 to 0 1 with the default set at O 0 08 The third tunable parameter is mar providing the range of stretching of the unscaled shadow function into the scaled function The default of 7 is the location of the maximum of the histogram of 9 but it could be set at a greater value if the corrected image is too dark in the expanded shadow regions which indicates the histogram maximum does not represent fully illuminated areas The advantage of the presented method is its fast processing performance because it relies exclu sively on spectral calculations and avoids time consuming geometric cloud shadow pattern consid erations The drawback is that useful geometric information is neglected In some cases it is useful to have the de shadowed digital number DN image in addition to the surface reflectance product This facilitates a comparison with the originally recorded DN imagery The conversion from reflectance to the corresponding at sensor radiance is performed with eq 10 1 Then eq 10 6 is employed to compute the de shadowed DN image for ch
132. coefficients zeroborder 2 set smile coefficients to 0 at spectral borders first last channel 1 repeat smile coefficients outside of interpolated values range search range default 20 nm overwrite silently overwrites the older output at_smoothdem infile dist outfile median DEM smoothing routine PARAMETERS infile input data cube single band ENVI image dist size of smoothing filter outfile name of output file KEYWORDS median use median filter instead of default low pass filter Chapter 7 Value Added Products Asa by product of atmospheric correction a number of useful quantities can readily be calculated The first group of value added products include vegetation indices based on surface reflectance instead of at sensor radiance simple parametrizations of the leaf area index and wavelength integrated reflectance albedo The second group comprises quantities relevant for surface energy balance investigations which are a useful supplement for studies in landscape ecology and related fields e g as input for regional modeling of evapotranspiration These include global radiation on the ground absorbed solar radiation net radiation and heat fluxes Emphasis is put on simple models based on the reflectance temperature cube derived during the atmospheric correction No additional data with the exception of air temperature is taken into account All value added products are written to a file with up
133. cosh Pre Y z CHAPTER 10 THEORETICAL BACKGROUND 154 where T is the total ground to sensor transmittance and Eqir Eajr are the direct irradiance and diffuse solar flux on the ground respectively So the ATCOR version of IRC contains some improvements with respect to the original method the path radiance varies spatially mainly caused by terrain height variations possibly also due to visibility variations and the sky view factor can be provided from a ray tracing analysis instead of the local slope angle Note the IRC method usually performs well However due to the statistical evaluation of the regression analysis unphysically large gt 1 reflectance unit or small lt 0 surface reflectance values might happen for some pixels usually in areas with topographic shadow or low local sun elevations 10 1 3 Spectral solar flux reflected surface radiance The spectral solar fluxes on the ground can be calculated by setting the parameter irrad0 1 in the inn file or using the graphical user interface The fluxes depend on solar geometry terrain elevation topography and atmospheric conditions For a flat terrain ATCOR provides spectra of the direct diffuse and global flux for the selected visibility water vapor In case of variable visibility water vapor the spectra are calculated for the average scene visibility water vapor The direct flux is just the beam irradiance on the ground times the cosine of the local solar ze
134. ctor 1000 The scaled value 1000 indicates full solar irradiance smaller values a corresponding fractional value e an update of the path radiance in the blue to red spectral region is performed if required provided a blue spectral band exists e water vapor retrieval using the previously calculated visibility map If the scene contains no reference areas the user has to specify a constant visibility that enters the water vapor calculation e reflectance spectrum retrieval with pixel based water vapor and visibility map Iterations for adjacency effect and spherical albedo are included For the adjacency correction the reflectance of cloud pixels is replaced with the scene average reflectance to avoid an overcor rection of the adjacency effect e temperature emissivity retrieval if thermal bands exist 10 6 2 Algorithm for rugged terrain The algorithm for rugged terrain basically consists of the same processing step as in the flat terrain but every step has to take into account some or all DEM information CHAPTER 10 THEORETICAL BACKGROUND 191 During the calculation of the visibility index map the DEM information elevation slope aspect skyview factor is taken into account The retrieval of the water vapor map has to include the terrain elevation e The empirical BRDF correction is based on the local illumination map local solar zenith angle derived from the slope aspect and shadow channels e The retrieval of the sp
135. ctory atcor2 3 cal landsat4_5 sensor text as defined in atcor24 3 bin sensor dat line 5 1 0 gain setting Any positive value is accepted this gain setting g is used to replace the c in the corresponding cal file with c g where g is the same for all channels line 6 calfile calibration file name line 7 0 9500 0 iemiss dem_unit surface emissivity DEM height unit iemiss surface emissivity option or value disregarded if no thermal band exists CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 131 iemiss 0 invokes e 0 98 to be consistent with the definition of earlier ATCOR versions Since iemiss 1 is reserved for the cover dependent emissivity setting below e 1 0 has to be ap proximated as iemiss 0 999 or iemiss 0 9999 In case of multiple thermal bands this e holds for the thermal band itemp_band employed for the surface temperature evaluation see chapter 4 6 iemiss 1 fixed values of surface emissivity 0 98 water 0 97 vegetation 0 96 soil iemiss 2 same as iemiss 3 the iemiss 2 option of ATCORA is not supported here iemiss 3 NEM or ANEM method requires multiple thermal bands see chapter 10 1 4 iemiss 4 ISAC method requires multiple thermal bands see chapter 10 1 4 iemiss 5 both NEM and ISAC but ISAC is currently only supported for flat terrain imagery dem unit 0 m 1 dm 2 cm DEM height unit line 8 fele DEM
136. d There is not a single figure that can be given to summarize the accuracy for all situations because the radiometric accuracy of the method depends on several factors the calibration accuracy of CHAPTER 10 THEORETICAL BACKGROUND 192 the sensor the quality of geometric co registration of the spectral bands the algorithm for ortho rectification relying on auxiliary information such as attitude and GPS DGPS the accuracy of the radiative transfer code MODTRAN 5 the correct choice of atmospheric input parameters the terrain type flat or rugged and the surface cover Solar region In the solar region wavelength lt 2 5 um assuming a flat terrain and avoiding the specular and backscattering regions an accuracy of the retrieved surface reflectance of 2 for reflectance lt 10 and 4 reflectance units for reflectance gt 40 can be achieved 62 For rugged terrain the most important parameter is an adequate spatial resolution of the DEM or DSM digital surface model and the exact ortho rectification of the imagery It would be desirable to have a DEM of a quarter of the sensor s spatial resolution or at least the resolution of the sensor s footprint which is seldom available 59 Even in the latter case errors in the matching of imagery and DEM can lead to large relative reflectance errors exceeding 100 for critical geometries principal plane e g a mountain ridge with half a pixel offset between imagery and DEM 59
137. d water haze cloud etc is a useful product and the quality of the atmospheric correction may depend on the correct class assignment at least for some classes The previous haze cloud water land pixel classifyer is a binary decision a pixel belongs to a certain class or not In reality the decision is typically not unique and a class assignment has only a certain probability As the absolute probability of a class assignment is very difficult to assess we define three probability levels low medium and high coded 30 60 90 respectively These numbers might be interpreted as a percent probability but the numbers are relative and arbitrary Currently there are three quality layers cloud water and snow which are solely calculated with spectral criteria The quality file is written if the corresponding flag is set to 2 see chapter 9 3 and figure 5 9 in chapter 4 Cloud probability e low cloud probability coded 30 p blue gt 0 15 and p red gt 0 15 and p NIR p red lt 2 and p NIR gt 0 8 p red and p NIR p SWIR1 gt 1 and NDSI lt 0 7 or DN blue gt Tsaturation 10 53 where p blue is the apparent reflectance in a blue band and DN blue is the corresponding digital number If no blue band is available a green band around 550 nm is taken as a substitute If no green band exists a red band around 650 nm is taken Note that saturated pixels in visible bands are counted as cloud although they might be s
138. d angle Br 65 Right illumination map cos To avoid a misclassification of these bright areas the reflectance values have to be reduced Fig 2 4 center part In ATCOR empirical geometry dependent functions are used for this purpose For details the interested reader is referred to chapter 10 5 1 but the main idea is also discussed here In the simplest cases the empirical BRDF correction employs only the local solar zenith angle 5 and a threshold Gr to reduce the overcorrected surface reflectance pz with a factor G compare Figure 2 5 Pcorrected pLG 2 11 where G cosf cosBr gt g 2 12 The exponent b is in the range 1 3 to 1 see chapter 10 5 1 for details So the reflectance is only decreased for local solar zenith angles 8 gt Br until the minimum value G g is reached The parameters r and g may be scene dependent and can be set by the user In many cases the default g 0 25 works fine and only the threshold angle Gr has to be adapted Choice of Br The threshold illumination angle Gr should have some margin to the solar zenith angle to retain the original natural variation of pixels with illumination angles close to the solar zenith angle The threshold angle can be specified by the user and the following empirical rules are recommended e Br 0 20 if O lt 45 e If 45 lt O lt 20 then Br 0 15 e If 0 gt 55 then Gr 0 10 These rules are automatically applied if Gr 0 e g durin
139. d blue band CHAPTER 10 THEORETICAL BACKGROUND 167 L Lp t Preg Eq 7 modelling atm LUT sun geometry measurement Visibility km Figure 10 6 Schematic sketch of visibility determination with reference pixel can be employed to estimate the visibility automatically Kaufman et al 1997 For this purpose we use a modified version of the original idea for the following algorithm If a SWIR band exists the SWIR reflectance is calculated assuming a visibility of 23 km instead of the original version of top of atmosphere reflectance Then water pixels are excluded by employing only those pixels with SWIR reflectance values above 1 and an NDVI gt 0 1 For the 2 2 um band the upper threshold of the reflectance of the dark pixels is selected as 5 If the number of reference pixels is less then 1 of the image pixels then the upper threshold is increased to 10 or finally 12 Ifa 1 6 um band exists but no 2 2 wm band the corresponding upper thresholds are selected as 10 and 15 or finally 18 respectively The reflectance ratios for the red and blue band are then calculated as Pred 0 5 p22 and Polue 0 5 Pred 10 65 Prea 0 25 Pie and Pre 0 5 Pred 10 66 This situation is sketched in figure 10 7 The correlation factor of 0 5 between the 2 2 um and the red region is not a universal constant but may typically vary between 0 4 and 0 6 The correlation actually also works for dark soils So the dark pixels m
140. d brightness value for the nadir region i e reflectance or radiance then the nadir normalized brightness value of a pixel with column number j is calculated as Keo nadir baorm 3 d 5 fzli 10 79 Figure 10 12 Nadir normalization of an image with hot spot geometry Left reflectance image without BRDF correction Right after empirical BRDF correction where the function fa is obtained with three processing steps e The first step is the averaging over each interval 3 or 1 It yields a function f with m 1 grid points for the m off nadir intervals plus the nadir interval e Two cases are distinguished now if the image is not geocoded an interpolation from function fi m 1 to a function fa ncols is performed where ncols is the number of column pixels of the image If the image is geocoded an interpolation from the 3 grid to the 1 grid is performed no hot spot case e The third step is a filter with a moving average window applied to the fo function The following cases are distinguished if the image is not geocoded the window is 9 pixels without hot spot and 3 pixels with hot spot option If the image is geocoded the moving window extends over a 5 angular interval no hot spot and over a 3 interval with hot spot option Figure 10 12 shows part of a HyMap image acquired 3 June 1999 Barrax Spain 12 09 UTC containing the hot spot geometry The solar azimuth was 181 and the sensor scan line azi
141. d using the available DEM information on height z slope and aspect i e local solar illumination angle 3 and atmospheric conditions visibility water vapor VIS The direct flux on the ground is Edir x y d x y Eo Tsun VIS 2 y z cosB x y 10 26 CHAPTER 10 THEORETICAL BACKGROUND 155 where Eo Tsun are extraterrestrial solar irradiance and sun to ground transmittance respectively and b is the topographic shadow mask 0 shadow 1 sunlit pixel The diffuse flux in mountainous terrain accounts for the adjacency effect and multiple reflection effects from the surrounding topography Using the terrain view factor V from the last section and the effective terrain reflectance p Vi x y p x y and p V x y p x y the diffuse flux is approximated as Bane a y Edif flat b Tsun 2 Y 2 cosB cos0s F 1 b a y Tam lE Y z Voy 3 y Edir fla 1 Y 2 Edif plat z y 2 pe z y 1 alz y 10 27 The first line describes the anisotropic and isotropic components of the diffuse flux the second line accounts for multiple terrain reflection effects Related quantities to the global spectral solar flux on the ground are the wavelength integrated global flux and the absorbed solar flux Wm These play a role in the surface energy balance and they are available as part of the value added channels see chapter 7 2 equations 7 9 7 10 Surface reflected radiance The ground reflected or gro
142. d with 1000 Fraction of photosynthetically active radiation FPAR scaled with 1000 Surface albedo integrated reflectance from 0 3 2 5 um per cent 10 Absorbed solar radiation flux Rsolar Wm Global radiation E Wm omitted for constant visibility in flat terrain because it is a scalar which is put into the log file The next channels are only available in case of at least one thermal band Thermal air surface flux difference Rinerm Ratm Rsur face Wm Ground heat flux G Wm Sensible heat flux H Wm7 Latent heat LE Wm7 Net radiation Ry Wm7 Chapter 8 Sensor simulation of hyper multispectral imagery After atmospheric correction the surface reflectance and temperature emissivity cubes can be used to simulate new products which might be of interest e at sensor radiance cubes in the solar region 0 4 2 5 wm for different solar geometries and atmospheric conditions e at sensor radiance cubes in the thermal region 8 14 um for different atmospheric conditions e g for satellite sensor studies e resampling of the surface reflectance cube to an existing or new multispectral sensor e resampling of the surface emissivity cube to an existing or new multispectral sensor This is a convenient way to simulate realistic data for a spaceborne version of an airborne instru ment to obtain radiance data under differenct atmsopheric conditions or to compare hyperspectral
143. dependent emissivity can be selected as options The simple case of a 3 class emissivity file vegetation e 0 97 soil e 0 96 others e 0 98 is calculated on the fly Its output file name is mage1_atm_emi3 bsq the 3 indicating the 3 classes In case of standard sensors with multiple thermal bands e g ASTER the spectral emissivity channels are not computed and ATCOR uses only one of the thermal bands band 13 in case of ASTER If ASTER band 13 is offered as a single channel input file to ATCOR the emissivity is set to a constant value of e 0 98 for the surface brightness temperature calculation If all reflective bands and the thermal bands of ASTER are geocoded and offered as a 14 channel file to ATCOR then the 3 class emissivity option is also available If a user is interested in derived surface emissivity data the corresponding instrument e g ASTER has to be defined as a user specified sensor see chapter 4 6 CHAPTER 4 WORKFLOW 37 atcor _ bin ASTER Landsat TM SPOT atm_lib ASTER Landsat TM SPOT spec_lib ASTER Landsat TM SPOT demo_data Figure 4 11 Directory structure of ATCOR 4 6 User defined hyperspectral sensors Examples of satellite hyperspectral sensors are Hyperion and CHRIS Proba Since the channel center wavelength of these instruments may change from scene to scene they have to be treated as user specified sensors and a flexible interface has been implemented to enable the calcu
144. digital elevation model DEM Then the slope and aspect maps are calculated The next step is the calculation of the sky view factor see chapter 10 1 1 The original paper uses the simple equation based solely on the slope angle but with ATCOR a more accurate calculation based on a ray tracing can also be used in case of a steep terrain Then the following quantities are computed keeping the original notation of Kobayashi in most cases a 20 ho 277 10 20 Here 0 is the solar zenith angle in radian TF s denotes the slope map in radian then the simple version of the skyview is obtained with h 1 s m 10 21 The cosine of the local solar zenith illumination angle 5 is given in eq 10 17 Then the surface radiance for each channel Ls is calculated by subtracting the path radiance Lp from the at sensor radiance L Ls x y L z y Lp z y 2 10 22 In the ATCOR version of the IRC algorithm the path radiance varies spatially particularly due to the DEM height variation while a constant value per channel is used in the original IRC paper Then a regression analysis per channel of Ls versus cos is applied to calculate the slope m and intercept b After defining C m b the topographic correction map A is calculated cos0s C ho A z y 10 23 ED cosBla y Me y ho ae Finally the surface reflectance p is computed according to a L x y z A x y p x y Ary 10 24 T z y z 1 ai 2
145. ding MODTRAN aerosol type However some fine tuning is subsequently performed to be able to modify the wavelength behavior of the path ra diance compared to the standard aerosol types If Ly blue scene deviates more than 5 from Lp blue MODTRAN then Lp blue scene is used as the valid path radiance In addition the path radiance for any other bands in the blue to red region is linearly re scaled with the factor Lp blue scene Ly blue MODTRAN see Figure 10 8 Here the path radiance in the red band is used as a fixed tie point For wavelengths greater than 700 nm a posssible typical 10 differ ence in path radiance between the selected aerosol type after fine tuning and the actual aerosol is usually not important because path radiance contributes only a small fraction to the total radiance If the sensor has no blue spectral band but a green band than the green band is substituted and for the dense dark vegetation the surface reflectance relationship is used p green 1 3 p red 10 69 Now eq 10 67 is again employed for the green band instead of the blue band to calculate the path radiance the best match to a MODTRAN aerosol type and possibly a fine tuning of the path radiance Aerosol retrieval for VNIR sensors If no SWIR bands exist but at least a red band around 660 nm and a NIR band around 850 nm a different approach has to be taken see reference 64 for details It starts with the assumption of CHAPTER 10 TH
146. e The program computes a synthetic blue channel for sensors without a blue spectral band If CHAPTER 6 BATCH PROCESSING REFERENCE 105 it is started from the the interactive panel Tools a dialog pickfile box pops up prompting for the input file name If makeblue is started on the IDL command line the input file name must be fully qualified i e the path has to be included The input has to be atmospherically corrected data i e an atm bsq file where the blue band is missing e g imagery of SPOT IRS 1D DMC The output file contains a synthetic blue band as the first image channel calculated from the green red and NIR bands on a pixel by pixel basis The output file name is _blue_atm bsq This product is a useful supplement to be able to create a true color image of the scene The calculation of the surface reflectance in the synthetic blue band consists of three steps 1 The blue band reflectance index 1 is extrapolated from the green index 2 and red index 3 band reflectance p pa p2 p3 A2 M A3 A2 6 1 where the center wavelength of the blue band is taken as A 480 nm This represents the typical spectral behavior of soils and some artificial surfaces asphalt concrete 2 The red and NIR index 4 bands are employed to compute the ratio vegetation index VI pa p3 Pixels with VI gt 2 5 and p4 gt 10 are defined as vegetation and the blue band reflectance is calc
147. e smile sensor without thermal bands Imagery from smile sensors must be processed in the raw geometry IGM Image Geometry Map to preserve the original image columns During the surface reflectance retrieval the atmospheric to pographic correction is performed on a per column basis i e to each image column its appropriate center wavelength bandwidth is associated For the nadir pixel the channel center and bandwidth are defined in the sensor s wavelength file wvl for off nadir pixels the center wavelength is shifted according to eq 4 1 Currently the change in spectral bandwidth with image column is assumed to be negligible The per column processing typically implies a factor of 8 increase in processing time The following steps are to be performed CHAPTER 4 WORKFLOW 42 1 Define a sensor wvl cal rsp files RESLUT using the original set of wavelengths pre launch values as provided with the data from the data provider 2 Run the smile detection tool compare Section 5 7 7 using the sensor defined in 1 and ap propriate absorption features to derive the polynomial coefficients smile_poly_ord4 dat for smile correction in step 3 alternatively enter the smile polynomial factors from laboratory calibration Note if two detectors are in the sensor system this should be done separately for VNIR and SWIR option repeat values resolution 0 02 nm Combine the two files for VNIR and SWIR manually into one fi
148. e if the wind was coming from the sea If the wind direction was toward the sea and the air mass is of continental origin the rural urban or desert aerosol would make sense depending on the geographical location If in doubt the rural continental aerosol is generally a good choice The aerosol type also determines the wavelength behavior of the path radiance Of course nature can produce any transitions or mixtures of these basic four types However ATCOR is able to adapt the wavelength course of the path radiance to the current situation provided spectral bands exist in the blue to red region and the scene contains reference areas of known reflectance behavior The interested reader may read chapter 10 4 2 for details Visibility estimation Two options are available in ATCOR e An interactive estimation in the SPECTRA module compare chapter 5 The spectra of different targets in the scene can be displayed as a function of visibility A comparison with reference spectra from libraries determines the visibility In addition dark targets like vegetation in the blue to red spectrum or water in the red to NIR can be used to estimate the visibility e An automatic calculation of the visibility can be performed if the scene contains dark reference pixels The interested reader is referred to chapter 10 4 2 for details Water vapor column The water vapor content can be automatically computed if the sensor has spectral bands in water vapor re
149. e is converted into the corresponding ENVI elevation file shadow_batch input filename pirelsize 10 0 solze 30 5 solaz 160 8 dem_unit 0 The keywords have the same meaning as for skyview_batch solze is the solar zenith angle degr and solaz is the solar azimuth angle degr In particular filename should have the last four characters as _ele and the bsq extension The output file replaces the ending _ele with the ending shd e g example DEM25m_zen31_azil61_shd bsq The rounded zenith and azimuth angles will be included in the shd file name Note The shadow and skyview calculations can be omitted in gently undulated terrain Example for maximum slopes of 25 and a solar zenith angle of 40 no DEM shadow is possible Also the local trigonometric sky view factor employed if the _sky bsq file is missing is sufficiently accurate compare figure 10 3 A TIFF input elevation file is converted into the corresponding ENVI elevation file atcor2_batch input filename output file vis vis tiff2envi tiff2envi or atcor2_tile input filename ntx 3 nty 2 output file vis vis tiff2envi tiff2envi The 2 in atcor2_batch means the code for flat terrain i e no DEM is employed The file name must be fully qualified i e it includes the path e g data2 project1 imagel bsq The file should have the band sequential BSQ file structure A corresponding in
150. e mask if HOT gt mean HOT 0 5 o HOT Then the HOT histogram of all haze pixels is calculated Pixels with values less than 40 of the cumulative histogram are assigned to thin medium haze pixels with higher values to medium thick haze This distinction is arbitrary and has no effect on the subsequent processing Haze over water Pixels must belong to the water mask and the NIR apparent reflectance p NIR must be greater than the NIR clear water threshold Tetear atery 1 rg defined in the preference parameter file chapter 9 3 Thin haze over water is defined as TelearwateryIR 2 NIR gt 0 06 10 51 Medium haze over water is defined as 0 06 lt p NIR lt Ta 10 52 where To default 0 12 is another editable parameter in the preference file The method of haze removal over water is described in chapter 10 5 3 The same technique is also employed to remove sun glint CHAPTER 10 THEORETICAL BACKGROUND 163 10 3 Quality layers The previous section defined a coarse pixel classification which is useful for an atmospheric cor rection In addition it supports an assessment of the quality of the processing For example a large error in the radiometric calibration could cause a scene classification with all pixels labeled as water In this case a user can immediately identify the problem Of course a more detailed assessment is possible with an analysis of the reflectance spectra Nevertheless the classification map lan
151. e pixels red NIR bands 10 4 2 Aerosol retrieval and visibility map If a sensor has the appropriate spectral bands the aerosol type and visibility or optical thickness of the atmosphere can be derived provided the scene contains reference areas of known reflectance be havior Kaufman and Sendra 1988 Kaufman et al 1997 The minimum requirements are spectral bands in the red and near IR If the scene contains dense dark vegetation coniferous type the re flectance values in the red band can be obtained from a correlation with the SWIR band reflectance as detailed below The visibility of each reference pixel can then be calculated in the red band as the intersection of the modeled at sensor radiance curve with the measured radiance see figure 10 6 The measured radiance for a reference pixel of digital number DN is L cy amp DN which is a constant value indicated by the dashed line in figure 10 6 The curve indicates the modeled radiance It employs the reflectance of the reference surface e g Pref 0 02 and uses values of path radiance atmospheric transmittance and global flux for the current solar and viewing geometry stored in precalculated LUTs Automatic masking of reference areas 1 6 or 2 2 um band required or at least red NIR bands If the sensor has a SWIR band at 1 6 or 2 2 um then the scene can be searched for dark pixels in this band and a correlation of the SWIR reflectance with the reflectance in the red an
152. e water vapor content They have the extension atm Example h99000_wv04_rura atm represents a file with the symbolic altitude 99 000 m water vapor column 0 4 cm and the rural aerosol CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 123 Each atm file contains the look up tables for the visibility range 5 120 km solar zenith angles 0 70 and ground elevations O 2500 m increment 500 m The files phasefct bin in the atcor bin directory contain the path radiance as a function of the scattering angle for the standard multispectral instruments with tilt capability e g SPOT Ikonos Quickbird ALOS AVNIR 2 A small field of view is assumed for all hyperspectral or multispectral instruments i e the specified tilt angle and solar zenith angle holds for the whole scene 9 1 3 Database update with solar irradiance In the solar region any high spectral resolution database of LUTs is based on the specification of an extraterrestrial spectral solar irradiance because the values of path radiance direct and diffuse solar fluxes depend on solar irradiance Other quantities direct and diffuse atmospheric transmittances and spherical albedo are independent of the solar spectrum ATCOR s standard atmospheric database is calculated for a certain irradiance F A and the corresponding file e0_solar_zzx dat is included in the directory atm_database Beginning with the ATCOR 2011 release there
153. eatures for very low water vapor values This water vapor threshold can be set by the user see chapter 9 3 The file zxx_out_hew bsq haze cloud water corresponding to a scene xrz bsq contains three relative levels of cirrus optical thickness thin medium and high The corresponding thresholds are arbitrarily set depending on the statistics mean standard deviation of the apparent reflectance p 1 38um map The file xxx_out_hcw bsq is intended as a visual aid or quicklook therefore the cirrus level maps of different scenes cannot be compared quantitatively As an example a certain scene setting could be e thin cirrus thickness color coded as light yellow with 0 010 lt p cirrus lt 0 015 e medium thickness color coded as darker yellow with 0 015 lt p cirrus lt 0 025 e high thickness color coded as bright yellow with p cirrus gt 0 025 reflectance units In addition to the 1 38 um cirrus channel another channel index w1 around 1 24 wm or as a substitute a NIR channel from the 800 to 900 nm region is employed with a ratio criterion to define cirrus pixels p cirrus p w1 gt T cir 10 97 Reference 21 proposes a threshold of T cir 0 3 to distinguish tropospheric aerosols due to dust storms from cirrus clouds However in the absence of dust storms this threshold is too high and predicted no cirrus in a number of test scenes containing a lot of cirrus clouds Therefore we use much lower
154. ectral reflectance cube consists of the steps 1 three iterations for terrain reflectance 2 empirical BRDF correction depending on illumination map if enabled 3 adjacency correction including proper treatment of cloud areas 4 spherical albedo correction e The retrieval of surface temperature and emissivity includes the maps of visibility index water vapor if water vapor bands exist elevation and scan angle No slope aspect correction is performed in the thermal region 10 7 Spectral classification of reflectance cube The SPECL code performs a spectral preclassification of the reflectance cube based upon template spectra at the Landsat Thematic Mapper reference wavelengths i e 0 48 0 56 0 66 0 83 1 6 2 2 um and returns a map of class indices for each pixel The template spectra consist of typical vegetation covers soil sand and water see figure 10 24 If the spectral reflectance signature agrees within a 10 margin at the reference wavelengths with one of the class template spectra it is put into this class otherwise it belongs to the class undefined rm e 50 green vegetation 0 bright sand h my 40 L r nm q e SA EF rel yellow veg Y sand bare soil v 30 a S E 20 E 7 A T E asphalt man made E Pa water o ES 0 5 1 0 15 2 0 0 5 1 0 1 5 2 0 Wavalength sm Wavalength sm Figure 10 24 Examples of template reflectance spectra employed by the SPECL code 10 8 Accuracy of the metho
155. ed de shadowing technique works for multispectral and hyperspectral imagery over land acquired by satellite airborne sensors The method requires a channel in the visible and at least one spectral band in the near infrared 0 8 1 um region but performs much better if bands in the short wave infrared region around 1 6 and 2 2 wm are available as well The algorithm consists of these major components i the calculation of the covariance matrix and zero reflectance matched filter vector ii the derivation of the unscaled and scaled shadow function iii a histogram thresh olding of the unscaled shadow function to define the core shadow areas iv a region growing to include the surroundings of the core shadow areas for a smooth shadow clear transition and v the de shadowing of the pixels in the final shadow mask Details are published in 63 diffase direct attenuated solar beam Figure 10 18 Sketch of a cloud shadow geometry The method starts with a calculation of the surface reflectance image cube p p A where three spectral bands around A 0 85 1 6 and 2 2 um are selected These bands from the near and shortwave infrared region are very sensitive to cloud shadow effects because the direct part CHAPTER 10 THEORETICAL BACKGROUND 184 of the downwelling solar radiation flux at the ground level is typically 80 or more of the total downwelling flux Channels in the blue to red region 0 4 0 7 um are not used for the d
156. ed threshold angle can be calculated from eq 2 12 with b 1 Br arccos PS arccos 2 15 In many cases a separate treatment of BRDF effects for soil rock and vegetation provides better results For this purpose two modes of BRDF correction are available for vegetation compare the graphical user interface of Figure 5 31 The first mode is superior in most cases Reference 67 contains a comparison of different topographic correction methods for several Landsat TM ETM and SPOT 5 scenes from different areas The proposed empirical ATCOR approach performed best in most of these cases but no method ranked first in all cases 2 3 Spectral calibration This section can be skipped if data processing is only performed for imagery of broad band sensors Sensor calibration problems may pertain to spectral properties i e the channel center positions CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 22 and or bandwidths might have changed compared to laboratory measurements or the radiometric properties i e the offset co and slope c1 coefficients relating the digital number DN to the at sensor radiance L co c x DN Any spectral mis calibration can usually only be detected from narrow band hyperspectral imagery as discussed in this section For multispectral imagery spectral calibration problems are difficult or impossible to detect and an update is generally only performed with respect to the radiometric calibration coeffic
157. ed with MODTRAN 6 and Fontenla 2011 are offered The plots show the detailed information line structure contained in the Fontenla spectrum The curves with 2 8 nm and 10 nm represent results based on a moving average of the the 0 4 nm data over 7 and 25 spectral points respectively 202 APPENDIX B COMPARISON OF SOLAR IRRADIANCE SPECTRA Relative Difference Relative Difference black FWHM 0 4 nm green FWHM 2 8 nm red PWHM 19 0 nm 100x K1997 F2011 F2011 0 4 0 5 0 6 0 7 0 8 0 9 Wavelength jam black FWHM 0 4 nm green FWHM 2 8 nm red FWHM 10 0 nm 100x K1997 F2011 F2011 1 0 Ez 1 4 1 5 1 8 2 0 ZE Wavelength jam 203 giaa 2 4
158. efault limit is set to a high value 150 reflectance to avoid a truncation in spectra of high reflectance surfaces e g snow and or surfaces with specular reflectance The previous default cut off before 2012 was set to 90 Note on factor b Factor b is a relative saturation factor applied to the maximum radiometric encoding e g for 8 bit data and b 1 all pixels with DN 255 will be marked as saturated in the _out_hcw bsq file color coding red Setting b 0 9 implies pixels with DN gt 230 will be considered as saturated or in the non linear radiometric range close to saturation This factor is only used for 8 and 16 bit signed or unsigned data not for float or 32 bit data CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 129 Note on the cloud mask The criterion al or a2 is also coupled with the conditions p NIR p red lt 2 and p NIR p SWIR1 gt 1 and NDSI lt 0 7 where NDSI is the normalized difference snow index A quality or probability mask for clouds is generated with the above three conditions and different apparent reflectance thresholds Te in the blue green spectral region For Te 15 we define a low probability cloud Te 25 a medium probability and Te 35 a high probability cloud The result is put into a cloud map file named image_quality bsq if the input scene is named image bsq The cloud mask in the image_out_hcw bsq file is based on the user defined threshold Te in
159. elevation file name empty line for a flat terrain calculation line 9 fslp DEM slope file name empty line for a flat terrain calculation line 10 fasp DEM aspect file name empty line for a flat terrain calculation line 11 fsky DEM skyview file name empty line for a flat terrain calculation line 12 fshd DEM cast shadow file name empty line for a flat terrain calculation rugged terrain empty if calculated on the fly line 13 atmfile atmospheric LUT file name reflective region e If the automatic aerosol type retrieval is intended for batch jobs the usual aerosol identifier in the file name e g rura has to be replaced with auto Example file name without path is aamsrura atm replace it with aamsauto atm The program then uses all aerosol types for the ms mid latitude summer atmosphere in the aerosol type estimate and selects the one with the closest match compare chapter 10 4 2 In the example of the ms case four aerosol types rural urban maritime desert are checked In case of the tr tropical atmosphere only three aerosol types rural urban maritime will be found in the atmospheric library The automatic aerosol type retrieval requires the parameter npref 1 variable visibility see line 20 below If npref 0 it is reset to npref 1 In the interactive mode the user can just press the Aerosol Type button on ATCOR s main panel to execute the aeros
160. ent of 30 the phase_fct bin files in the ATCOR directory A corresponding tilt azimuth angle interpolation of the LUTs is automatically done 9 6 1 Landsat 7 ETM For ETM two types of metadata files are distributed The first type contains the min max radiance for each band e g Lmax_band1 191 6 Lmin_band1 6 2 from which the radiometric offset cy and slope or gain c for ATCOR s cal file are calculated as e co 0 1 x Lmin and c 0 1 Lmax Lmin 255 CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 139 The factor 0 1 converts from the ETM radiance unit Wm sr yum into the ATCOR radiance unit mWem sr um The second type of meta file fast format metafile fst specifies the bias and gain values directly line with biases and gains in ascending band number order again the factor 0 1 is required to convert into the ATCOR radiance unit co 0 1 bias and c 0 1 gain Different types of processing are used in the NLABS and LPGS environments see the comprehensive article 34 The general equations to convert the digital number DN into an at sensor radiance are L B G DN 9 4 B Lmin Imax lisa Gis 9 5 Qmaxz Qmin Imax Lmin 9 6 Qmax Qmin ee where B bias G gain and Qmin 1 Qmax 255 for LPGS processing starting at December 8 2008 while Qmin 0 and Qmax 255 is used for the previous NLAPS processing The facto
161. enu provides spectral filtering of single spectra reflectance emissivity radiance provided as ASCII files spectral filtering of image cubes and spectral polishing 000 X Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Help Resample a Spectrum Licensed for Daniel Spectral Polishing Statistical Filter Spectral Polishing Radiometric Variation Spectral Smile Interpolation Image Cube Figure 5 43 Filter modules 5 5 1 Resample a Spectrum This program serves for the general purpose of resampling It requires an ASCII file with two columns as input The first column is wavelength nm or um unit the second is reflectance or emissivity or something else e g spectral radiance The reflectance range can be 0 1 the intrinsic reflectance unit or the percent range 0 100 Figure 5 44 shows the graphical user interface The input spectrum is high spectral resolution data After specifying the first band of the sensor the resampling is performed for all spectral bands and the result is written to an output file again with two columns wavelength and resampled spectrum Pick Input Spectrum Vexport data data7 atcor2 3 spec_L1b full_resolution alfalfa dat Pick Response File first band of sensor Vexport data data atcor2 3 sensor Chris_nodel_RE band01 rsp Dutput Filename resampled spectrum Vexport data data7 atcor2 3 5pec_11b full_resolution alfalFa_Chris_model_RE dat
162. er Pixels 0 keep values do not replace 1 use land average 2 use line average of land pixels Water Vapor Map over Land Smooth with box size meter Bo o 270 no band interpolation in 760 nm region 21 interpolate bands in 760 nm oxygen region 2 0 no band interpolation in 725 825 nm region Q 1 interpolate bands in 725 and 925 nm water vapor region 0 no band interpolation in 940 1130 nm region lt 1 interpolate bands in 940 and 1130 nm water vapor region Q 0 no band interpolation in 1400 1900 nm region 1 interpolate bands in 1400 and 1900 nm water vapor region 0 standard water vapor algorithm 1 nonlinear influence of vegetation included Q 0 do not write _out_hcw file haze cloud water land 41 write _out_hcw file lt 2 hcw quality file Cloud reflectance threshold 2 in the blue green region cloud mask 125 0 Water reflectance threshold 2 in the NIR region water mask Water reflectance threshold 2 in the 1600 nm region water mask Maximum surface reflectance cut off limit 90 0 Water vapor threshold to switch off cirrus algorithm cm 30 10 Define saturation DN saturated gt b DN max with b 0 9 to 1 0 b i0 90 Start Stop Regions for Interpolation a negative value restores defaults 940 nm 885 0 to 11000 0 1130 nm 1069 0 to 180 0 1400 nm 1300 0 to 490 0 1900 nm 730 0 to 970 0 Haze Sun Glint over water apparent NIR reflectance Ti clear
163. er wavelengths A reference spectrum can be loaded when clicking the corresponding button at the top right of the panel The bottom graphics shows soil and water spectra taken from the scene and a soil reference spectrum from the spec_lib library CHAPTER 5 DESCRIPTION OF MODULES 66 An exact match of scene spectra and library spectra cannot be expected but the trends spectral shapes should be consistent The parameter Visibility can be set in the corresponding widget near the center of the panel and its influence on the retrieved spectrum will be shown when placing the target box in the scene To vary the atmospheric water vapor content the user should switch to the main panel and select a different atmosphere e g switch from the mid latitude summer atmosphere with a column wv 2 9 cm to a tropical atmosphere with wv 4 1 cm compare chapter 6 Most multispectral satellite sensors do not have spectral bands placed in water vapor regions so the retrieved reflectance signature will show a rather small dependence on the water vapor estimate For sensors with water vapor bands e g Hyperion or MOS B the retrieved reflectance spectrum strongly depends on the water vapor content In this case the water vapor map can be derived during the image processing see the panel of figure 5 17 as part of the reflectance retrieval Note that the water vapor is not retrieved from the image pixel for correction in the SPECTRA module as it is done in the image
164. ermal Sensor Temperature Band Number Spectral band used for temperature retrieval algorithm Optional according to sensor type Smile Sensor Response Type required for convolution when sensor definition is not given explicetely may be Butterworth Order 1 slow drop off Butterworth Order 2 close to Gauss Butterworth Order 3 between Gauss Rect Butterworth Order 4 close to Rect angular Gaussian Rectangular Triangular Decreasing Binning from Rectangular to Triangular or Arbitrary Outputs A new sensor_ dat file and possibly sensor directory is created ATTENTION this routine requires write access to the sensor directory of the ATCOR installation 5 2 2 Create Channel Filter Files ATCOR requires the spectral response of all bands of a sensor being present as a spectral filter file response file rsp For spectroscopic instruments the band characteristics are often only available by band center and width at FWHM Full width half maximum This function creates the response curves from the latter information compare Fig 5 13 CHAPTER 5 DESCRIPTION OF MODULES Aaa IN ATCOR Sensor Definition Selected Sensor src_idl atcor atcor_23 sensor hyperion167 sensor_hyperion16 dat New Rename Delete Sensor Type 2 Standard y Smile Sensor w Thermal Sensor Number of cross Track Pixels 256 First last Reflective Band to A First last Mid IR Band lb to
165. errain Rugged Terrain haze cloud water illumination AOT water vapor T Y cloud shadow T y p T cube T y value added Figure 4 10 Input output image files during ATCOR processing image1_atm log Once an image is processed with ATCOR all input parameters are saved in a file inn that is automatically loaded should the image be processed again The recommended extension for the radiometric calibration files is cal Other extensions are ele bsq for the digital elevation file _slp bsq for the DEM slope file _asp bsq for the DEM as pect sky bsq for the sky view file _ilu bsq for the solar illumination file in rugged terrain and cla bsq for the classification map of SPECL The visibility index map is named _visindex bsq the aerosol optical thickness is _aot bsq the cloud building shadow map is named _fshd bsq and the atmospheric water vapor map wv bsq Thermal band imagery In case of thermal band imagery the surface temperature and emissivity are stored in separate files Surface temperature is appended to the reflectance cube i e is included as the last channel in the _atm bsq file e g mage1_atm bsq The surface temperature calculation is based on an assumption for the emissivity A constant emissivity or a surface cover
166. etection of shadow regions because they receive a much larger diffuse radiation component making them less sensitive to partial shadow effects Instead visible channels serve to define a potential cloud mask The surface reflectance is first computed with the assumption of full solar illumination i e the global flux on the ground consists of the direct Eqiy and diffuse Eq component If DN denotes the digital number of a pixel Lp the path radiance and 7 the atmospheric transmittance ground to sensor the surface reflectance can be obtained as _ n d co i c i DNi x y Lpi e 10 98 pi x y TlEdira Edif Here d is the Earth Sun distance at the image acquisition time in astronomical units cy and c1 are the radiometric calibration coefficients offset and slope to convert the digital number into the corresponding at sensor radiance L i e L co c1 DN and is the channel index The proposed de shadowing algorithm consists of a sequence of eight processing steps as sketched in Fig 10 19 It starts with the atmospheric correction The next step is the masking of water bodies and cloud areas with simple spectral criteria as detailed below Water pixels have to be excluded as far as possible to avoid their assignment as shadow pixels Step 3 calculates the covariance matrix C p where p is the surface reflectance vector comprising only the non water and non cloud pixels For each pixel this vector holds the ref
167. fective bandwidth of each channel CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 143 specified as AA 0 068 0 099 0 071 and 0 114 um for the blue green red and NIR band respectively reference 46 from http www digitalglobe com The calibration is different for compressed 8 bit data and the original 11 bit data ATCOR contains a template for an 8 bit cal file quickb_8bit_std cal and an 11 bit file quickb_16bit_std cal However it is recommended to use only the 11 bit data for ATCOR The IMD metadata file contains the absolute calibra tion factor absCalFactor for each multispectral channel in the unit Wm sr Depending on processing date the effectiveBandwidth AA um unit is also included The nominal offset is co 0 in each channel and the ATCOR gain c has to be specified in the unit Wem sr yum which requires the following conversion equation for Quickbird absCalFactor 0 1 a AA Therefore the template calibration file has to be updated i e copied to a new file name and edited according to the absCalFactor of the scene IMD file and the above spectral bandwidth values AA 9 6 6 IRS 1C 1D Liss The metadata file contains the geographic coordinates as well as the solar elevation and azimuth angles It also includes the radiometric calibration coefficients the bias B Lmin and gain G Lmax in the ATCOR radiance unit mWem sr yum The radiometri
168. for a mid latitude summer atmosphere rural aerosol visibility 15 km ground elevation 500 m above sea level For convenience a log and an ini file are created for the documentation of the processing parameters e g datal image_toarad ini In addition the corresponding sensor calibration file will be created Example sensor casi96 scalef 1000 then file casi96_scalef1000 cal will be created on the atcor4 sensor casi96 directory with the radiometric calibration coefficient c 0 001 for each band Chapter 9 Implementation Reference and Sensor Specifics This chapter discusses miscellaneous topics associated with the current implementation of ATCOR First the user is acquainted with the structure and handling of the atmospheric database Second the supported input output file types are given The next item discusses the preference parameters e g the definition of thresholds employed for the masking of cloud and water areas and options for interpolating certain spectral regions Then the parameters of the inn file are described which is employed for the interactive and batch processing Last but not least a section on problems and tips is included 9 1 The Monochromatic atmospheric database This section presents the technical details of the atmospheric database To be capable of handling typical hyperspectral sensors with arbitrary spectral bands in the so lar and thermal spectral re
169. g batch processing The angle 8r can actually be calculated from the imagery as demonstrated by the following ex ample When processing the scene with ATCOR the map of local solar zenith angles is stored in CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 21 0 8 0 8 CT O E Function G 0 4 exponent b 1 top to bottom curves Ar 45 degr fAr 55 degr Br 55 degr 40 50 60 70 80 90 local illumination angle degree Figure 2 5 Geometric function G for three thresholds of Pr The reflectance is not modified for 8 lt Br Br 45 55 65 and reduced for 8 gt Br until the minimum value g 0 2 is reached a separate file ilu If the output file after atmospheric topographic correction contains bright overcorrected areas this file should be linked to the ilu file using any available standard image processing software The ilu file contains the illumination map 3 scaled as byte data ilu 100 x cos Bi arccos ilu 100 2 13 2 14 Let us assume an example A pixel in a dark area of the ilu image has the value ilu 32 i e 8 71 The overcorrected reflectance value be pz 80 and this value shall be reduced to 40 a value typical for the flat terrain neighborhood Then the threshold angle has to be specified such that cos3 cosGp 0 5 with exponent b 1 in equation 2 12 in this case Gp 50 So if the desired reflectance reduction factor is G then the requir
170. gions e g 920 960 nm The approach is based on the differential absorption method CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 19 and employs bands in absorption regions and window regions to measure the absorption depth see chapter 10 4 3 Otherwise if a sensor does not possess spectral bands in water vapor regions e g Landsat TM or SPOT an estimate of the water vapor column based on the season summer winter is usually sufficient Typical ranges of water vapor columns are sea level to space tropical conditions wv 3 5 cm or g cm midlatitude summer wv 2 3 cm dry summer spring fall wv 1 1 5 cm dry desert or winter wv 0 3 0 8 cm 2 2 BRDF correction The reflectance of many surface covers depends on the viewing and solar illumination geometry This behavior is described by the bidirectional reflectance distribution function BRDF It can clearly be observed in scenes where the view and or sun angles vary over a large angular range Since most sensors of the satellite version of ATCOR have a small field of view these effects will only play a role in rugged terrain and for the wide FOV sensors such as IRS 1D WiFS or MERIS For flat terrain scenes across track brightness gradients that appear after atmospheric correction are caused by BRDF effects because the sensor s view angle varies over a large range In extreme cases when scanning in the solar principal plane the brightness is particularly high in the hot
171. gions a large database of atmospheric LUTs was compiled with the MODTRAN 5 radiative transfer code in 2010 The database is called monochromatic because of its high spectral resolution compare figure 9 1 After resampling with the spectral response functions of any sensor a typical size of the sensor specific database is 10 MB Chapter 9 1 4 con tains a description of the resampling program RESLUT 9 1 1 Visible Near Infrared region In the solar spectral region 0 34 2 54 um MODTRAN was run with different wavenumber spacings to achieve a wavelength grid spacing of approximately 0 4 nm except for the 1400 nm and 1800 nm regions This required the use of MODTRAN s p1_2008 database i e 0 1 em in the 2 1 2 5 wm region In addition different RT algorithms were used in atmospheric window regions the scaled DISORT algorithm with 8 streams SD 8 was employed in absorption regions the more accurate SD 8 with the correlated k algorithm was selected 27 Since the wavenumber grid is not equidistant in wavelength the LUTs were resampled with an equidistant 0 4 nm grid of Gaussian filter functions of FWHM 0 4 nm to speed up subsequent calculations So the new LUT database should be sufficient for instruments with bandwidths gt 2 nm covering the solar spectral region from 340 to 2540 nm The standard database is calculated for nadir viewing instruments An off nadir database with 121 CHAPTER 9 IMPLEMENTATION REF
172. h regions are active the average of the water vapor maps of both regions is taken Option 2 employs a linear regression of bands which yields better results for the water vapor map if the data is noisy or not accurately calibrated If iwv_model 2 and channels in the 940 nm and 1130 nm regions are specified then only the 940 nm region is evaluated with a linear regression of bands If the regression is intended for the 1130 nm region then the 940 nm channels all channels in line 29 have to be specified as 0 CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 136 line 34 0 icirrus flag for cirrus removal 0 disabled 1 enabled 1 forced The value icirrus 1 enforces cirrus detection and removal i e termination criteria are ignored line 35 0 irradO flag for solar flux on ground 0 disabled 1 enabled For irrad0 2 the surface reflected leaving radiance is calculated additionally For a flat terrain ASCII spectra of the direct diffuse and global flux on the ground are provided in the folder of the input scene see chapter 10 1 3 In case of a flat terrain the global i e direct plus diffuse flux image is calculated For a rugged terrain the images of the direct and diffuse fluxes are calculated Note as the flux files have to use a float encoding 32bits pixel the file size is twice or four times the size of the input scene for a 16bit pixel and 8bit pixel input scene respectively Notice concerning visibility itera
173. has 9 reflective and 5 thermal bands ATCOR calculates surface reflectance and a surface brightness temperature from band 13 ASTER has four gain settings high H normal N and low1 L1 low2 L2 for the reflective bands Table 9 3 contains the cl values for the different reflective bands and gain settings in the ATCOR radiance unit mWem sr yum It was taken from the ASTER user s guide 18 The thermal band 13 has a calibration gain of cl 5 693E 4 CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 144 band high gain normal gain low gain 1 low gain 2 1 0 0676 0 1688 0 225 N A 2 0 0708 0 1415 0 189 N A 3 0 0423 0 0862 0 115 N A 4 0 01087 0 02174 0 0290 0 0290 5 0 00348 0 00696 0 00925 0 0409 6 0 00313 0 00625 0 00830 0 0390 T 0 00299 0 00597 0 00795 0 0332 8 0 00209 0 00417 0 00556 0 0245 9 0 00159 0 00318 0 00424 0 0265 Table 9 3 Radiometric coefficients c1 for ASTER 9 6 9 DMC Disaster Monitoring Constellation DMC is a constellation of several orbiting satellites with an optical payload intended for rapid disaster monitoring All DMC sensors have three spectral bands green red NIR with a spatial resolution of 32 m and a swath of 600 km The metadata file dim and htm formats pertaining to each scene contains the solar geometry and the radiometric calibration coefficients The bias and gain specified in the metadata are defined as L bias DN gain 9 8 usi
174. has to be rescaled to the physical range from 0 to 1 where 0 indicates no direct illumination full shadow and 1 means full direct illumination The histogram of is used to rescale the image data Fig 10 20 shows a schematic sketch of such a histogram with a smaller peak at 2 representing the shadow pixels and the main peak at mar representing the majority of the fully illuminated areas The statistical assumption is used that full direct solar illumination is already obtained for pixels with z y maz Then the values are linearly mapped from the unscaled Prim max interval onto the physically scaled 0 1 interval where the scaled shadow function is named e a ee A 10 102 Omar P min 1 if gt maz 10 103 The smallest value of the scaled shadow function is P 0 which means no direct illumination However to avoid overcorrection and to cope with scenes containing merely partial shadow areas it is advisable to set O at a small positive value This value of P i e the minimum fractional direct illumination deepest shadow in a scene typically ranging between 0 05 and 0 10 is scene dependent see the detailed discussion below In principle the de shadowing could now be performed with the physically scaled function P which represents the fraction of the direct illumination for each pixel in the p vector i e the complete scene without cloud and water pixels However since the matched filter is n
175. he run button the messages are printed to the prompt or the console if available Error and status messages may also be found in the log file during and after processing 000 X ATCOR Tiled Processing Select Input File Name src_idl atcor atcor_23 demo_data tm_rugged tm_blforest bsd DeFine Name of Output Cube Varc_idl atcor atcor_23 deno_data tn_rusged tn_blforest_atn bsq Number of Tiles Dimension A Y Dimension 2 ATCOR Method w ATCOR 2 flat gt ATCOR 3 rugged Help Run Done Figure 5 36 ATCOR Tiled Processing CHAPTER 5 DESCRIPTION OF MODULES 78 5 4 Menu Topographic The Topographic menu contains programs for the calculation of slope aspect images from a dig ital elevation model the skyview factor and topographic shadow Furthermore it supports the smoothing of DEMs and its related layers 0 00 X Satellite AT COR File New Sensor Atm Correction Topographic Filter Simulation Tools Help c DLR 2011 Slope Aspect Version 8 0 2 Skyview Factor Shadow Mask DEM Smoothing Licensed for Daniel Figure 5 37 Topographic modules 5 4 1 Slope Aspect Slope and aspect are to be calculated from the DEM before ATCOR is run for rugged terrain This menu function as depicted in Fig 5 38 allows to calculate both layers in one step Inputs Input DEM file The standard DEM file used for atmospheric correction This DEM should be in meters Output file n
176. ia 1992 The atmospheric longwave radiation Rat emitted from the atmosphere toward the ground can be written as Ratm a O T 7 12 where is the air emissivity and T is the air temperature at screen height 2 m above ground sometimes 50 m above ground are recommended For cloud free conditions Brutsaert s 1975 equation can be used to predict the effective air emissivity 1 24 alae 7 13 CHAPTER 7 VALUE ADDED PRODUCTS 112 Here Pwv is the water vapor partial pressure millibars hPa and Ta is the air temperature K Figure 7 1 shows py as a function of air temperature for relative humidities of 20 100 The partial pressure is computed as Pwy RH es 100 7 14 where RH is the relative humidity in per cent and e is the water vapor partial pressure in saturated air Murray 1967 a To 273 16 es Ta eso cap o e The constants are a 17 26939 b 35 86 and eso es 273 16K 6 1078 hPa An alternative to equation 7 13 is the following approximation Idso and Jackson 1969 which does not explicitly include the water vapor and holds for average humidity conditions compare Figure 7 2 7 15 a 1 0 261 exp 7 77 x 1074 273 TJ 7 16 un o o 30 Water Yapor Partial Pressure hPa D 5 19 15 20 25 30 35 Air Ternperature C Figure 7 1 Water vapor partial pressure as a function of air temperature and humidity Relative humidities are 20 to 100 w
177. iation are removed from the list of shadow pixels e a 7 x 7 pixel convolution filter is being applied to the shadow mask e a transition region shadow sunlit is introduced e already dark pixels DN lt DN mean o in the NIR are not reduced in brightness during BRDF correction The threshold is evaluated for a NIR channel but the non reduction of brightness reflectance is applied to all channels 9 4 Job control parameters of the inn file former ini If the file name of the input image is example_image bsq then a file example_image inn is created during the interactive ATCOR session When all image processing parameters have been defined by the user this inn file is written to the directory of the corresponding image When the image is re loaded during a later session or when a batch job is submitted all input parameters are read from this file It is suggested not to edit this file because it might create inconsistent input data The file might contain empty lines when input is not required for a specific case The IDL routine for writing this file is available on request for users who want to run ATCOR batch jobs without employing the interactive GUI panels The contents of an inn file are line 1 20 08 1989 Date dd mm year i e 20 08 1989 occupies the first 10 columns line 2 10 0 scale factor reflectance line 3 5 0 pixel size m line 4 landsat4_5 Landsat 4 5 TM sub dire
178. iation components Figure 10 4 Solar illumination geometry and radiation components If 0 0 s n denote solar zenith angle terrain slope solar azimuth and topographic azimuth respectively the illumination angle 8 can be obtained from the DEM slope and aspect angles and the solar geometry cosB x y cosO cosO x y sinOgsinO 1 y cos s bn x y 10 17 The illumination image cosB x y is calculated within ATCOR and stored as a separate map The CHAPTER 10 THEORETICAL BACKGROUND 153 diffuse solar flux on an inclined plane is calculated with Hay s model Hay and McKay 1985 also see Richter 1998 for the enhancement with the binary factor b E x y 2 Ea 2 bT 2 cosB z y cosOs 1 brs 2 Vsty z y 10 18 The sky view factor can be computed from local information as Vsgy x y cos On x y 2 based on the local DEM slope angle On ATCOR uses the horizon algorithm that provides a more accurate value of the sky view factor by considering the terrain neighborhood of each pixel Dozier et al 1981 Vsky and Vierrain are related by Vary 2s y 1 Wierrat Es y 10 19 10 1 2 Integrated Radiometric Correction IRC The IRC method was published by Kobayashi and Sanga Ngoie 40 41 to provide a combined atmospheric and topographic correction The algorithm is briefly outlined here more details can be found in the original papers The first step is the orthorectification of the scene using a
179. ich can be set high to reduce the percentage of wrongly classified water pixels because these are likely shadow areas and water pixels will not be de shadowed However if an external water map exists either from file image_hcw bsq or image_water_map bsq if image bsq is the file name of the scene and this file indicates more than 15 of water pixels then this information is regarded as reliable and this water map is excluded from de shadowing If the average scene elevation is above 1 2 km the gradient criterion cannot be applied and the NIR SWIR1 water surface reflectance thresholds apply as defined in the pref erence parameter file chapter 9 3 These surface reflectance thresholds can be defined as positive values as the gradient criterion in not valid anyway or as negative An increase decrease of the thresholds will increase decrease the number of water pixels e Rugged terrain the slope and illumination maps show strong horizontal and vertical stripes Strong artifacts in the DEM files will immediately be visible in the atmospherically topo graphically corrected surface reflectance image This problem frequently occurs for resampled DEMs e g the original DEM resolution is 30 m which is resampled to a 5 m pixel size Ar tifacts will be enhanced especially if the stepsize of the original DEM height resolution is coded as integer Float data would have smoother transitions A simple way to get better results is
180. ients see chapter 2 4 Surface reflectance spectra retrieved from narrow band hyperspectral imagery often contain spikes and dips in spectral absorption regions of atmospheric gases e g oxygen absorption around 760 nm water vapor absorption around 940 nm These effects are most likely caused by a spectral mis calibration In this case an appropiate shift of the center wavelengths of the channels will remove the spikes This is performed by an optimization procedure that minimizes the deviation between the surface reflectance spectrum and the corresponding smoothed spectrum The merit function to be minimized is x7 5 D toi 6 parecia 2 16 i l where ye 9 is the surface reflectance in channel i calculated for a spectral shift 6 pi th is the smoothed low pass filtered reflectance and n is the number of bands in each spectrometer of a hyperspectral instrument So the spectral shift is calculated independently for each spectrometer In the currently implemented version the channel bandwidth is not changed and the laboratory values are assumed valid More details of the method are described in 26 A spectral re calibration should precede any re calibration of the radiometric calibration coefficients see section 5 7 8 for details about this routine 20 70 60 E 15 T 50 3 8 40 5 10 E E g 30 e 20 10 0 O he 0 4 0 6 0 8 1 0 1 2 1 4 0 4 0 6 0 8 1 0 1 2 1 4 Wavelength m Wavelength m Figure 2 6 Wavelength shift
181. ile but defaults always to the ATCOR installation in order to allow to select an ATCOR system file from within the installation such as the cal files the solar reference files the sensor definition files The function then allows to adjust and save the respective text contents of the selected file CHAPTER 5 DESCRIPTION OF MODULES 54 000 IX Calibration File Plot File Font_Size Display Output Help src_idl atcor atcor23 cal ali ali22dec2004 cal Calibration Canstonts c0 Offost mW com ar am c1 Ga in mW cm ar pm DN 4 6 Band Number Wovalength Figure 5 7 Plotting a calibration file O O X src_idl atcor atcor_23 cal ali ali_22dec20 File Help mcm2 sr micron LO 00 y M0 NN Figure 5 8 Displaying a calibration file same file as in Fig 5 7 5 1 8 Edit Preferences The default settings of ATCOR may be edited through this panel The updated preferences are then written to the ASCII file as displayed at the top of the panel The preferences persist for the user who started the ATCOR application the next time the system is started and also for batch processing NOTE Preferences are set when one of the ATCOR modules has been opened So one should select one of the options from within the menu Atm Correction before editing the preferences CHAPTER 5 DESCRIPTION OF MODULES 55 File users richt_r idl rese atcor3 preference_parameters dat Water Yapor Option for Wat
182. ill be calculated anyhow from the image in the image processing 5 3 9 Inflight radiometric calibration module This chapter presents the GUI for the inflight calibration which may also be invoked from one of the four possible ATCOR main panels The purpose is the calculation of the radiometric calibration coefficients for spectral bands in the solar region based on measured ground reflectance spectra The user should be familiar with chapter 2 4 before using this function Note a ground reflectance spectrum from a field spectrometer has to be resampled with the chan nel filter curves of the selected sensor The field spectrometer file format should be converted into a simple ASCII file containing the wavelength nm or um in the first column and reflectance in CHAPTER 5 DESCRIPTION OF MODULES 67 the second column The resampling can then be done with the sequence Filter Resample a Spectrum from the ATCOR main panel The result is an ASCII file with 2 columns the first contains the channel center wavelength the nm and um unit is allowed the second contains the resampled reflectance value either in the 0 1 or 0 100 range If targetl is the name of the target ATCOR provides three ASCII files with information on target background properties and the derived calibration file These are the output of the c1 option of ATCOR s calibration module e File targetl adj contains the original target DN the ad
183. illumination gradients shall be removed without any further atmospheric correction the algorithm can also be applied to radiance DN data In this case the brightness gradient may be caused by a combina tion of surface BRDF and atmospheric BRDF left right asymmetry in path radiance The algorithm is intended for large field of view sensors minimum FOV 209 It computes the col umn means with a certain angular sampling interval 1 or 3 The input image may be geocoded or not If it is not geocoded the total field of view FOV corresponds to the number n of across track image pixels per line If geocoded the scan angle for each pixel must be provided in a separate file _sca It contains the scan angle in degree scaled with a factor of 100 and coded with 16 bits per pixel This definition is taken from the airborne ATCOR PARGE interface Schlapfer and Richter 2002 Scan angles on the right hand side with respect to flight heading are defined as negative those on the left side as positive e g a value of 2930 represents a scan angle of 29 3 on the right side CHAPTER 10 THEORETICAL BACKGROUND 174 The nadir region is defined here as the 3 scan angle range Usually a 3 angular sampling interval from 3 to FOV 2 on the left side and 3 to FOV 2 on the right side is adequate except for geometries close to the hot spot geometry In the latter case a 1 sampling interval can be selected If bnadir denotes the average
184. image bsq denotes the file name of the input image then the following products are available e image_atm_emi3 bsq 3 or 4 emissivity classes obtained from an on the fly in memory pre classification vegetation soil sand water The pre classification requires daytime data acquisition and spectral bands in the solar region This file has one channel with the emissivity values for the specified thermal band or in case of ANEM the pixel dependent values assign the maximum emissivity of all available thermal bands CHAPTER 4 WORKFLOW label definition 0 geocoded background 1 shadow 2 thin cirrus water 3 medium cirrus water 4 thick cirrus water 5 land 6 saturated 7 snow 8 thin cirrus land 9 medium cirrus land 10 thick cirrus land 11 thin medium haze land 12 medium thick haze land 13 thin medium haze water 14 medium thick haze water 15 cloud land 16 cloud water 17 water Table 4 4 Class label definition of hcw file e image_atm_emiss bsq contains the spectral emissivity map for all thermal channels 44 e image_atm_emiss_lp3 bsq is the same emissivity map but filtered with a 3 channel low pass filter to smooth spectral noise features requires at least 10 thermal bands e image_atm_isac_emiss bsq emissivity cube for the ISAC algorithm perature perature e image_at_sensor_tbb bsq at sensor brightness temperature cube e image_at_surface_tbb bsq at
185. infrared region around 1 6 and 2 2 um are available as well A fully automatic shadow removal algorithm has been implemented However the method involves some scene dependent thresholds that might be optimized during an interactive session In addition if shadow areas are concentrated in a certain part of the scene say in the lower right quarter the performance of the algorithm improves by working on the subset only The de shadowing method employs masks for cloud and water These areas are identified with spectral criteria and thresholds Default values are included in a file in the ATCOR path called preferences preference_parameters dat As an example it includes a threshold for the reflectance of water in the NIR region p 5 So a reduction of this threshold will reduce the number of pixels in the water mask A difficult problem is the distinction of water and shadow areas If water bodies are erroneously included in the shadow mask the resulting surface reflectance values will be too high Details about the processing panels can be found in section 5 3 10 Chapter 3 Basic Concepts in the Thermal Region Fig 3 1 left presents an overview of the atmospheric transmittance in the 2 5 14 um region The main absorbers are water vapor and CO which totally absorb in some parts of the spectrum In the thermal region 8 14 um the atmospheric transmittance is mainly influenced by the water vapor column ozone around 9 6 um and
186. input data is 8 bit data then a scale factor of 4 is recommended i e a surface reflectance of 20 56 will be coded as 82 If the input file name is mage bsq then the default output file name for the atmospherically cor rected image is mage_atm bsq The user may modify the output name but it is recommended to keep the _atm bsq qualifier to facilitate the use of subsequent programs Then the sensor view geometry has to be specified as well as the sensor and the calibration file The atmospheric file contains the look up table LUT results of the radiative transfer calculations separately for the CHAPTER 4 WORKFLOW 32 solar and thermal region These LUTs are calculated for a nadir view but for tilt sensors the files phasefct bin in the atcor bin directory contain the path radiance as a function of the scattering angle and the appropriate file is automatically included INPUT IMAGE FILE Yexport data data atcor2 3 demo_data tm_flat tm_essen1000 bsq Date dd mm year 20 08 1989 OUTPUT IMAGE FILE Vexport data data atcor2 3 demo_data tm_flat tm_essen1000_atm bsq I OVERWRITE Scale Factor 4 0 Help Satelite Cog benmetm Band selection Selected SENSOR Landsat 4 5 TM Select H Z Pixel size m 30 0 CALIBRATION FILE export data data atcor2 3 cal landsat4_5 tm_standard cal ATMOSPHERIC FILE aamsrura ATH FILE for thermal band s midlat_summer Adjacency range km 1 00 Help Zones a
187. ion of DN to at sensor radiance co offset for conversion to at sensor radiance d relative sun earth distance average d 1 Eo solar irradiance top of atmosphere NOT at aircraft altitude o solar zenith angle Inputs e input file name e calibration file name x cal e solar radiation file e0_solar_ spc e output file name CHAPTER 5 DESCRIPTION OF MODULES 88 e scale factor see below e date of the year given exactly as day month eg 26 7 for July 26th used for sun earth distance calculation e solar zenith angle use Tools Solar Zenith and Azimuth for its calculation Output A cube containing the scaled apparent reflectance in is stored The data format is driven by the scaling factor as follows e scale lt 10 byte e scale gt 10 integer e scale gt 500 unsigned integer e scale lt 1 floating point effective value unit wavelength reference and FWHM are taken from the file e0_solar_ spc OOP 3 ATCOR Apparent Reflectance Calculation setect Input File Name data hyperion Bern_02 Hyper ion_sub67 bq setect Calibration File src_idl atcor atcor 23 sensor hyper iont67 hyperion_167 cal setect Solar Reference File E0 src_id1 atoor atcor_23 sensor hyperionl67 0_solar_hyperion167 spc Tetine Name of Output Cube data hyperion Bern_02 Myper ion_sub_rhoapp bsq Scale factor x Refl E Date d m evos Solar Zenith deg E Help Done Figure
188. ion of thin cirrus requires specific narrow bands around 1 38 um or 1 88 um compare chapter 10 5 4 As a first approximation haze is an additive component to the radiance signal at the sensor It can be estimated and removed as described below Cloud areas have to be masked to exclude them from haze areas and to enable a successful haze removal The treatment of cloud shadow regions is discussed in chapter 10 5 5 The haze removal algorithm runs fully automatic It is a combination of the improved methods 57 90 and consists of five major steps 1 Masking of clear and hazy areas with the tasseled cap haze transformation 15 TC x BLUE x3 RED 10 82 where BLUE RED 21 and za are the blue band red band and weighting coefficients respectively The clear area pixels are taken as those pixels where TC is less than the mean value of TC 2 Calculation of the regression between the blue and red band for clear areas clear line slope angle a see figure 10 14 If no blue band exists but a green spectral band then the green band is used as a substitute 3 Haze areas are orthogonal to the clear line i e a haze optimized transform HOT can be defined as Zhang et al 2002 HOT BLUE x sina RED x cosa 10 83 4 Calculation of the histogram of HOT for the haze areas CHAPTER 10 THEORETICAL BACKGROUND 178 5 For bands below 800 nm the histograms are calculated for each HOT level j The haze signal A to be
189. is an arbitrarily defined contrast threshold Another often used concept is the optical thickness of the atmosphere 9 which is the product of the extinction coefficient and the path length x e g from sea level to space in a vertical path 4 8m 2 2 The optical thickness is a pure number In most cases it is evaluated for the wavelength 550 nm Generally there is no unique relationship between the horizontal visibility and the vertical total optical thickness of the atmosphere However with the MODTRAN radiative transfer code a certain relationship has been defined between these two quantities for clear sky conditions as shown in Fig 2 1 left for a path from sea level to space The optical thickness can be defined separately for the different atmospheric constituents molecules aerosols so there is an optical thickness due 14 CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 15 to molecular Rayleigh and aerosol scattering and due to molecular absorption e g water water ozone etc The total optical thickness is the sum of the thicknesses of all individual contributors molecular scattering 6 aerosol 6 molecular absorption 2 3 The MODTRANO visibility parameter scales the aerosol content in the boundary layer 0 2 km altitude For visibilities greater than 100 km the total optical thickness asymptotically approaches a value of about 0 17 which at 550 nm is the sum of the molecular thickness 6 0
190. ith a 10 increment bottom to top curves respectively eq 7 14 The calculation of the heat fluxes G H and LE on the right hand side of equation 7 7 requires different models for vegetated and man made surfaces For vegetated or partially vegetated surfaces we employ a simple parametrization with the SAVI and scaled NDVI indices Choudury 1994 Carlson et al 1995 G 0 4 Rn SAV Im SAVI SAVIm 7 17 where SAV Im 0 814 represents full vegetation cover The sensible heat flux is computed as H B T Ta 7 18 B 286 0 0109 0 051 NDVI 7 19 CHAPTER 7 VALUE ADDED PRODUCTS 113 0 95 color Idso amp Jackson Brutsaert 0 90 RH 90 E RH 70 5 5 0 85 RH 50 o 2 E 0 80 Z 0 75 0 70 0 65 O 5 10 15 20 25 30 35 Air Temperature C Figure 7 2 Air emissivity after Brutsaert eq 7 13 and Idso Jackson eq 7 16 n 1 067 0 372 NDVI 7 20 x P850 P650 NDVI 7 21 0 75 P850 P650 ey Equation 7 18 corresponds to equation la of Carlson et al 1995 because G is neglected there and so Rn G represents the energy left for evapotranspiration The factor 286 in equation 7 19 converts the unit em day into Wm NDVI is the scaled NDVI ranging between 0 and 1 and truncated at 1 if necessary Equation 7 21 corresponds to equation 3 of Carlson et al 1995 with NDV Ip 0 bare soil and NDVI 0 75 full vegetation cover The latent heat flux LE is c
191. ixel does not occur in the selected reference channel However a common reference channel is needed in this method to obtain a consistent pixel independent spectrum of unscaled path radiance i and transmittance TW 10 2 Masks for haze cloud water snow A useful first step before executing an atmospheric correction is the calculation of a pixel map for haze cloud water snow etc Such a pre classification has a long history in atmospheric correction methods 37 38 55 56 57 39 44 It is also employed as part of NASA s automatic processing chain for MODIS 45 using the classes land water snow ice cloud shadow thin cirrus sun glint etc A similar approach is taken here The calculation is done on the fly and if the scene is named scene bsq then the corresponding map is named scene_out_hcw bsq There is also the possibility to provide this information from an external source if a file scene_hcw bsq exists in the same folder as the scene bsq then this information is taken and the internal ATCOR calculations for this map are skipped In this case the coding of the surface types has to agree with the ATCOR class label definition of course see Table 10 1 This file is written if the corresponding flag is set to 1 see chapter 9 3 and figure 5 9 in chapter 4 label class 0 geocoded background 1 shadow 2 thin cirrus over water 3 medium cirrus over water 4 thick cirrus
192. jacency corrected DNY and the ground reflectance data for each band e File targetl rdn radiance versus digital number contains the band center wavelength target radiance L and corresponding digital number This file can be used as input to a regression program cal regress that allows the calculation of the calibration coefficients with a least squares fit in case of multiple targets more than two e File targetl cal contains three columns band center wavelength offset or bias set to zero co 0 and c according to equation 2 18 Remark a bright target should be used here because for a dark target any error in the ground reflectance data will have a large impact on the accuracy of cy Figure 5 22 Radiometric calibration target specification panel At the top line of the menu of figure 5 22 the mode of calibration is specified one or two targets can be employed by selecting the button cl or c0 amp cl respectively If a calibration is intended CHAPTER 5 DESCRIPTION OF MODULES 68 for n gt 2 targets each target has to be specified separately in the cl mode which creates a file target i rdn i 1 2 n with the name target 1 specified by the user These files contain the radiance and corresponding digital number spectrum as a 2 column ASCII table and they are employed in the batch module cal_regress see chapter 5 to calculate a least squares regression for the calibration c
193. km is applied to smooth sensor noise and small scale variations of the spectral correlation coefficient for the DDV reference pixels The resulting visibility index and AOT CHAPTER 10 THEORETICAL BACKGROUND 190 maps are stored as separate files The visibility calculation based on the reference pixels has to account for the adjacency effect because reference areas are embedded in non reference areas see the sketch below Since the weighting fraction of reference to non reference area within the adjacency range is not known for each pixel the visibility calculation is performed with an average adjacency weighting factor of 0 5 q Figure 10 23 Weighting of q function for reference pixels Ladj cor Co DN 0 59 DN DNav ctear 10 108 L VIS Ly Tpref Eg T Ladj cor 10 109 Next the visibility is converted into the nearest visibility index vi range 0 54 compare Fig 10 9 to store the visibility index map as byte data Spatial interpolation is performed to fill the gaps for non reference pixels or the average vi value can be taken to close the gaps A moving average window of 500 m x 500 m is employed to reduce the influence of noise In addition the haze over land haze over water and cloud pixels are coded with 55 56 and 57 in the visibility index map respectively The cloud building shadow map is stored separately fshd bsq file containing the fraction of direct solar irradiance per pixel scaled with the fa
194. l atcor atcor_23 atn_database Solar irradiance file fl Ysrczidl atcor atcor_23 atn_database e0_solar_Fonten2011_04nn dat Solar irradiance file f2 Ysrc_idl atcor atcor_23 sun_irradiancese0_solar_kurucz1997_04nm dat High resolution database 2 Ysrc_idl atcor atcor_23 atn_database_kurucz1997 Convert Database 1 irradiance f1 into Database 2 irradiance f2 Lal a QUIT 1 Figure 5 59 Convert monochromanic database to new solar reference function 5 7 11 Convert atm for another Irradiance Spectrum The conversion as described in module 5 7 10 can be applied to a sensor specific atmospheric library of a self defined sensor using this function In the panel as of Fig 5 60 the sensor has first to be entered and the new solar function e0_solar dat is to be selected before the conversion may be applied CHAPTER 5 DESCRIPTION OF MODULES Figure 5 60 Convert atmlib to new solar reference function 98 CHAPTER 5 DESCRIPTION OF MODULES 99 5 8 Menu Help Finally the Help menu allows browsing of the ATCOR user manual provides a linke to the web resources and displays license information 000 XI Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Help Il Licensed for Daniel Version 8 0 2 c DLR 2011 Browse Manual Web Resources About Your License Figure 5 61 The help menu 5 8 1 Help Options The options of the help menu are listet below Br
195. l region so the cirrus contribution is easily traced The scatterplot is computed in terms of the apparent TOA or at sensor reflectance of 1 38 Versus pred Where the apparent reflectance is defined as TL gt _ 10 91 P E cos0 where L is the recorded radiance signal the extraterrestrial solar irradiance for the selected band and 0 is the solar zenith angle Following 19 the method can be described by the following set of equations Te A PO PO pe A AO 200 T 10 92 Here pe is the reflectance of the cirrus cloud Te the two way transmittance direct plus diffuse through the cloud p the reflectance of the virtual surface land or water surface including all effects of molecular and aerosol scattering below the cirrus and se is the cloud base reflectance of upward radiation Eq 10 92 can be simplified because of se p lt lt 1 yielding PA pelA TA PA 10 93 With the assumption that the cirrus reflectance p A is linearly related to the cirrus reflectance at 1 38 um we obtain p A pe 1 38um y 10 94 where y is an empirical parameter derived from the scene scatterplot of p1 38 versus prea land or p124 water It depends on the scene content cirrus cloud height and solar and viewing angles Fig 10 17 shows an example of such a scatterplot The red line is the left side boundary of data points that are not influenced by ground surface reflection i e cirrus contaminated pixels a
196. lation of atmospheric LUTs adapted to the spectral channel properties This interface is similar to the one for the airborne ATCOR and the corresponding part of the airborne ATCOR user manual is repeated here The hyperspectral tool is an optional add on to the satellite ATCOR Although it is mainly intended for hyperspectral sensors it can of course also be employed for user defined multispectral instruments The first step is the definition of a new sensor subdirectory compare Figure 4 13 and Fig 5 11 It serves as the start folder of further subdirectories for the user defined hyperspectral sensors e g chris_m1 A few sensor description files have to be provided by the user as explained in the next section 4 6 1 Definition of a new sensor A few steps have to be taken to include a new satellite sensor in ATCOR They are supported by the respective panels On the level of the ATCOR installation directory the following steps are to be taken e A new sensor subdirectory in the atcor sensor folder has to be created This is easiest done using the routine Define Sensor Parameters as described in chapter 5 2 1 Please make sure this name does not agree with an existing multispectral sensor name in the atcor cal folder This process may also be done manually by copying and adapting an existing user defined sensor or one of the samples provided or use the function above to make the changes CHAPTER 4 WOR
197. layers of the air volume between ground and sensor T is the atmospheric ground to sensor transmittance e is the surface emissivity ranging between 0 and 1 Lgp T is Planck s blackbody radiance of a surface at temperature T and F is the thermal downwelling flux of the atmosphere see Fig 3 2 So the total signal consists of path radiance emitted surface radiance and reflected atmospheric radiation The adjacency radiation i e scattered radiation from the neighborhood of a pixel can be neglected because the scattering efficiency decreases strongly with wavelength For most natural surfaces the emissivity in the 8 12 um spectral region ranges between 0 95 and 0 99 Therefore the reflected downwelling atmospheric flux contributes only a small fraction to the signal Neglecting this component for the simplified discussion of this chapter we can write L L DN L Lpp T path Co C1 path 3 2 TE TE In the thermal region the aerosol type plays a negligible role because of the long wavelength and atmospheric water vapor is the dominating parameter So the water vapor and to a smaller de gree the visibility determine the values of Lpath and 7 In case of coregistered bands in the solar and thermal spectrum the water vapor and visibility calculation may be performed with the solar channels In addition if the surface emissivity is known the temperature T can be computed from eq 3 2 using Planck s law For simplicity a constan
198. lb First last Thermal IR Band to Sensor Total FOV deg 530000 per pa Help Define the Sensor Figure 5 12 Definition of a new sensor Inputs sensor Zatcor_23 sensor hyperion167 sensor_hyperion167 dat loaded Done 58 Wavelength File A wavelength reference file Format ASCII 3 columns no header column 1 band number column2 center wavelength column 3 band width Unit nm Type of Filter Function The type defines the basic shape of each of the created response curves Options are Butterworth Order 1 slow drop off Butterworth Order 2 close to Gauss Butterworth Order 3 between Gauss Rect Butterworth Order 4 close to Rectangular Gaussian Rectangular Triangular Decreasing Binning from Rectangular to Triangular Outputs A numbered series of band_x rsp files are created at the same location as the wavelength refer ence file is situated The files contain wavelength reference and the relative response in two ASCII formatted columns CHAPTER 5 DESCRIPTION OF MODULES 59 AAA Generate Spectral Filter Functions Columns 1 3 are band number center wavelength bandwidth micron or nm Wavelength File Yarc_idl atcor atcor_23 sensor huperon137 sensor_huperion197 uvl Select Type of Filter Function wit Butterworth order 1 slow drop off wv 3 Butterworth order 3 between Gauss rectangular w 4 Butterworth order 4 close to rectangular v5
199. le name cal with three columns center wavelength co C1 where name is specified by the user Note If several calibration targets are employed care should be taken to select targets without spectral intersections since calibration values at intersection bands are not reliable If intersections of spectra cannot be avoided a larger number of spectra should be used if possible to increase the reliability of the calibration CHAPTER 5 DESCRIPTION OF MODULES 97 5 7 10 Convert High Res Database New Solar Irradiance The standard solar irradiance curve used with ATCOR is the Fontenla database 2011 However there s some uncertainty about the solar irradiance and people may want to use ATCOR with a different solar reference function This module CONVERT_DB3 converts the complete high resolution atmospheric database from the current definition to a new irradiance function Normally the standard database is converted this function does not apply to the thermal IR but also the specific CHRIS database may be selected In the panel see Fig 5 59 the two databases may be selected on the basis of the directory f1 and the new reference function e0_solarx dat N CONVERT_DB3 Convert bp7 Files in atm_database for Another Solar Irradiance File 4 Standard high resolution atmospheric database 340 2547 nm wv CHRIS Proba database smaller spectral coverage 380 1080 nm High resolution database 1 Yarc_id
200. le Vertical Opens a window for a vertical profile through the image of the first band only Profile Spectrum Opens a window for a spectrum of the image for images with 4 and more bands only Export Allows to export the currently displayed image to one of the given image data formats The displayed image may be exported as a scaled 8bit 24bit image to the available standard image formats Note when clicking in the zoom window the current image value and location is displayed or a small plot of the spectrum at this pixel location is created same as the function Profile Spectrum of above The menu in the such loaded window allows to save the spectrum to an ASCII table to adapt the graph s properties and font size configure the display and to output the graph to an appropriate graphics format CHAPTER 5 DESCRIPTION OF MODULES Figure 5 4 Display of ENVI imagery 51 CHAPTER 5 DESCRIPTION OF MODULES 52 5 1 2 Show Textfile Use this function if you need to edit a plain text file which comes together with the data to be processed The file is opened in a simple editor and may be changed and saved The function comes handy if an ENVI header needs to be checked or updated e g Selecting the Save or the Save As function in the submenu will allow to overwrite the file or to create a new one O O O Xx sre_idl atcor atcor_23 demo_data tm_rugged tm_blforest hdr File Save Help Save As
201. le afterwards 3 Using the same sensor as above run the atmospheric correction with the smile correction option switched ON after putting the new file smile_poly_ord4 dat into the sensor definition directory 4 Apply the spectral polishing routine see Section 5 5 2 and 5 5 3 and 5 Run the Spectral Smile Interpolation module see Section 5 5 4 on the atmospherically cor rected image 4 8 Haze cloud water map Although the surface reflectance cube and temperature emissivity for thermal channels is the main result of the atmospheric correction some additional products are often requested One of these products is a map of the haze cloud water and land pixels of a scene This map not only delivers a basic scene classification but it may also contain information about potential processing problems For example if turbid water pixels are not included in the water mask the haze mask may also be not appropriate and consequently results of the haze removal over land might be of poor quality Such a pre classification as part of the atmospheric correction has a long history 37 38 55 56 57 39 44 It is also employed as part of NASA s automatic processing chain for MODIS 45 using the classes land water snow ice cloud shadow thin cirrus sun glint etc Therefore the calculated haze cloud water map is a useful optional output of ATCOR It is enabled by setting the parameter hcw 1 in the preference_parameters d
202. le check on the derived DEM solar illumination file is also performed at the start of ATCOR see the discussion below A TIFF input elevation file is converted into the corresponding ENVI elevation file skyview_batch input filename pixelsize 10 0 dem_unit 0 unders unders azi_inc azi_inc ele_inc ele_inc filename is the full file name including the path filename should have the last four char acters as _ele and the extension bsq to indicate a digital elevation file and to enable an automatic processing e g example_DEM25m ele bsq pixelsize is specified in meters dem_unit is the integer code for the DEM height unit 0 represents m 1 means dm 2 means cm The option dem_unit 0 is default and can be omitted The keyword unders specifies the undersampling factor in pixels to reduce the execution time for very large files The default angular azimuth resolution is aziinc 10 degrees and the default elevation incre ment is ele inc 30 degrees However the recommended resolution is 10 degrees for azimuth and 5 degrees for elevation In case of large files an undersampling factor gt 1 can be specified to reduce the execution time Usually an undersampling factor of 3 is sufficient A high an gular resolution is more important than a low undersampling factor The output file replaces the ending ele with the ending _sky e g example DEM25m_sky bsq A TIFF input elevation fil
203. lectance values in the 3 selected channels around 0 85 1 6 2 2 um The matched filter is a vector tuned to a certain target reflectance spectrum p to be detected 1 _ C p P a pt DTO o P ey surface reflectance Y exclude cloud amp water matched filter vector _ E unscaled shadow function Y scaled shadow function ud threshold core shadow areas gt TA PEE J expand shadow mask cen A gt de shadowing with P GD gt Figure 10 19 Flow chart of processing steps during de shadowing CHAPTER 10 THEORETICAL BACKGROUND 185 Here J is the scene average spectrum without the water cloud pixels Selecting p 0 for a shadow target yields a special simplified form of the matched filter where the sh index symbolizes shadow Cp eS sh p C p 10 100 The shadow matched filter vector is then applied to the non water non cloud part of the scene and yields the still un normalized values that are a relative measure of the fractional direct illumination also called unscaled shadow function here P x y Vaalele y P 10 101 The matched filter calculates a minimum RMS shadow target abundance for the entire non water non cloud scene Therefore the values of are positive and negative numbers The arbi trary image depending range of
204. lint Removal sccccocesecsserserevesees Y Yes No Haze or Sun Glint Removal ssserscccrccssessosereces Yes QNo Shadow Removal Clouds Buildings ssscccccesesesseecs Y Yes No Shadow Removal Clouds Buildings sesseeeeersreeeeees Q Yes No Value Added Products sssseeees eereereerseereeress Yes ONo Value Added Products ooommmoo reereserery seeeeeees Y Yes No Cirrus Removal ssercccccercccsesessecssccvevessves O Yes No Solar Flux at Ground occccooroccnorccconcccncrcanannnos Y Yes No Solar Flux at Ground sssccccccscecescesvevserseesees Y Yes No Figure 5 27 Image processing options Right panel appears if a cirrus band exists Options that are not allowed for a specific sensor will appear insensitive If the haze removal option is selected in combination with Variable Visibility the visibility index proportional to total op tical thickness map is coded with the values 0 182 The value visindex 0 corresponds to visibility 190 km each integer step of 1 corresponds to an AOT increase of 0 006 The array serves as a fast method of addressing the radiative transfer quantities transmittance path radiance etc in case of a spatially varying visibility i e in combination with the DDV algorithm IF the Haze or Sunglint Removal button is selected the next panel will ask for haze removal over land option 1 haze or sunglint removal over water option 2 or haze removal over land and water option 3
205. ll major commercially available small to medium FOV sensors with a sensor specific atmospheric database of look up tables LUTs con taining the results of pre calculated radiative transfer calculations New sensors will be added on demand The current list of supported sensors is available at this web address A simple interface has been added to provide the possibility to include user defined instruments It is mainly intended for hyperspectral sensors where the center wavelength of channels is not stable and a re calculation 11 CHAPTER 1 INTRODUCTION 12 of atmospheric LUTs is required e g Hyperion Chris Proba An integral part of all ATCOR versions is a large database containing the results of radiative transfer calculations based on the Modtran 5 code Berk et al 1998 2008 While ATCOR uses the AFRL MODTRAN code to calculate the database of atmospheric look up tables LUT the correctness of the LUT s is the responsibility of ATCOR Historical note For historic reasons the satellite codes are called ATCOR 2 flat terrain two geometric degrees of freedom DOF 56 and ATCOR 3 three DOF s mountainous terrain 59 They support all commercially available small to medium FOV satellite sensors with a sensor specific atmospheric database The scan angle dependence of the atmospheric correction functions within a scene is neglected here The airborne version is called ATCOR 4 to indicate the four geometric DOF s x y z
206. loat image would require twice the disk space of the input image The default output data type is signed 16 bit integer for all integer and float input data employing the scale factor s 100 The scale factor is always included in the output ENVI header file Note Any positive value of the scale factor s is accompanied with a truncation of surface reflectance values at 0 in the output cube So a negative reflectance e g caused by a wrong choice of vis ibility or inaccurate radiometric calibration will be reset to zero in the output image In the SPECTRA module no truncation is applied If a user wants the output reflectance cube without zero truncation the scale factor s should be specified with a negative value e g s 1 will provide a float output surface reflectance retaining negative reflectance values s 100 will provide a 16 bit integer output file The byte scale 10 lt s lt 1 and output data range 0 255 cannot be used to represent negative values The negative scale factor should only be used for test purposes since the results do not make a physical sense and some further processing options or modules are excluded in this case e g the value added calculation of surface energy balance components the automatic spectral classification SPECL or the BRDF corrections Note The TIFF format is also allowed for the input file However some restrictions apply in this case compare chapter 6 2 e All bands in one TIFF
207. m region the following assumptions are being made for extrapolation Extrapolation for the 0 30 0 40 um region P0 3 0 4um 0 8 Po0 45 0 50um if blue a band 0 45 0 50 um exists e P0 3 0 4um 9 8 Po 52 0 58um green band no blue band available Extrapolation for the 0 40 0 45 um region P0 1 0 45um 0 9 P0 45 0 50um if a blue band 0 45 0 50 um exists e P0 4 0 52um 0 9 Po0 52 0 58um green band no blue band available The reflectance reduction factors in the blue part of the spectrum account for the decrease of surface reflection for most land covers soils vegetation The extrapolation to longer wavelengths is computed as e Ifa 1 6 ym band exists P2 0 2 5um 0 5 P1 6um if p850 P650 gt 3 vegetation CHAPTER 7 VALUE ADDED PRODUCTS 110 P2 0 2 5um P1 6um else e If no bands at 1 6 um and 2 2 um are available the contribution for these regions is estimated as P15 18um 0 50 Po s5um if psso poso gt 3 vegetation p2 0 2 5um 0 25 Po s5um if peso peso gt 3 P1 5 1 8um P0 85um else P2 0 2 5u4m P0 85um gt else At least three bands in the green red and near infrared are required to derive the albedo product Wavelength gap regions are supplemented with interpolation The contribution of the 2 5 3 0 wm spectral region can be neglected since the atmosphere is almost completely opaque and absorbs all solar radiation The output flx file contains the channel
208. m_database To sensor specific atmospheric library fexport data data atcor2 3 atm_lib chris_modeth Select aerosol types solar region F rural turban maritime 1 desert The name h99000 symbolizes a satellite altitude The string ww10 in the filename indicates a water vapor column 1 0 g cm2 or cm w20 indicates a water vapor column of 2 0 g cm2 or cm etc Show Selected Files atm_database h99000_wv04_rura bp 2atm_database h99000_ww10_rura bp atm_database h99000_wv20_rura bp 2atm_database h99000_ww29_rura bp 2 Reflective Region w Thermal Region Selected SENSOR CHRIS_MODES Select ATM files ets IRN Cancel OK Figure 9 4 GUI panels of the satellite version of program RESLUT The default output image data type is byte if the input is byte data Then a scale factor s 4 is employed i e the per cent reflectance value of each pixel is multiplied with s 4 and rounded to byte As an example a surface reflectance value of 20 2 will be coded as 81 However the user can modify the scale factor on ATCOR s main panel A value of s 10 to s 100 causes the output file to be coded as signed 16 bit integer i e with two bytes per pixel The specification s 1 0 produces a float output image i e with 4 bytes per pixel Attention The float output could be used for testing on small images For large files and an input data type of 2 bytes per pixel the output f
209. mponent as compared to wave lengths longer than 0 8 um The distinction of water bodies from cloud shadow areas may be difficult or impossible if it is based merely on spectral reflectance shape and amplitude information Water bodies should be excluded as far as possible to improve the performance of the de shadowing algorithm Currently water and cloud pixels are masked with the spectral criteria p 0 85um lt 5 and p 1 6um lt 1 water 10 105 p 0 48um gt 30 and p 1 6um gt 30 cloud 10 106 If no channel in the blue region is available a channel in the green 0 5 0 6 um or red part of the spectrum 0 6 0 68 um could be used as a substitute Both criteria do not uniquely define the CHAPTER 10 THEORETICAL BACKGROUND 188 corresponding class The water criteria allow some margin for turbid water in the NIR region The more restrictive criterion p 0 85 um lt 3 would perform better for clear water bodies However it would fail for moderately turbid or muddy waters Other common water classification criteria such as average reflectance over all bands p lt 3 or p 0 4 0 6um lt 6 may also fail So one has to compromise and tolerate a certain amount of misclassification for a fully automatic algorithm The scaled shadow map 9 x y is written to an output file The histogram of the unscaled shadow function Fig 10 20 typically has a main peak at maz a smaller secondary peak at 2 due to shadow pixels and
210. ms for the simulation of at sensor radiance scenes based on surface reflectance or emissivity and temperature images see chapter 5 6 The Tools menu contains a collection of useful routines such as the calculation of the solar CHAPTER 4 WORKFLOW 31 zenith and azimuth angles spectral classification nadir normalization for wide field of view im agery adding of a synthetic blue channel for multispectral sensors without a blue band e g SPOT spectral calibration conversion of the monochromatic atmospheric database from one to another solar irradiance spectrum BIL to BSQ conversion and more see chapter 5 7 Finally the Help menu allows browsing of the ATCOR user manual provides a link to web resources and displays license and credits information see chapter 5 8 4 2 First steps with ATCOR The 4tm Correction button of Fig 4 5 displays the choices ATCOR2 multispectral sensors flat terrain and ATCOR8 multispectral sensors rugged terrain If the add on for user defined mainly hyperspectral sensors is included then the buttons ATCOR2 hyperspectral sensors flat terrain and ATCORS hyperspectral sensors rugged terrain will also appear compare Fig 4 5 The last button starts the ATCOR processing in the image tiling mode i e the image is divided into sub images in x and y direction as specified by the user This mode is intended for large scenes compare section 5 3 12 and the inn file
211. muth was 179 almost exactly pointing into the solar azimuth The left image shows HyMap band 30 at 868 nm after atmospheric correction The right image is the result after nadir normalization with a 1 sampling interval In this example the column means were calculated globally i e surface cover independent The algorithm also contains an option to compute the column means separately for 4 surface covers It can currently only be selected if the input imagery is reflectance data and not geocoded The processing time is much larger than for the global cover independent method The four surface classes are CHAPTER 10 THEORETICAL BACKGROUND 175 e bright vegetation ratio vegetation index NIR RED gt 10 e medium dark vegetation 6 lt ratio vegetation index lt 10 e dry vegetation or mixed vegetation soil 3 lt vegetation index lt 6 e soil vegetation index lt 3 The reflectance of off nadir water pixels criterion near infrared reflectance lt 5 is not modified 2 Empirical BRDF correction in rugged terrain For many surface covers the reflectance increases with increasing solar zenith and or viewing angle 47 Scenes in mountainous regions often exihibit a large variation of terrain slopes and thus bidirectional brightness variations for a certain surface cover e g meadow or forest This behavior cannot adequately be eliminated with the Lambertian assumption of equation 10 15 This equation leads to overc
212. n file CHAPTER 6 BATCH PROCESSING REFERENCE 103 e g data2 project1 imagel ini must be available that contains all processing parameters This file will be generated during the interactive session It may be also be created by the user e g employing the program write_atcor3_inn file pro that is available on request The default output file name without the output keyword specification is the input name with an atm bsq appended e g data2 project1 imagel_atm bsq The keyword output can be used to specify the full output name or only the output path the latter option is recommended In that case all output files are written to the specified output directory and the reflectance output file name is the name of the input file with atm bsq appended Example output data4 project1 then the output reflectance file will be data4 project1 imagel_atm bsq The corresponding tile program atcor2_tile in this example is called to split the image into 3 sub images in x direction and 2 in y direction compare chapter 5 3 12 The optional keyword vis can be used to overwrite the visibility value in the inn file For a constant visibility per scene npref 0 in the inv file the input vis value is the start value that will be iterated as described in chapter 10 4 1 In case of a variable scene visibility npref 1 the vis parameter is ignored if the scene contains enough
213. n be included in the cirrus removal The classes are currently defined with the following criteria Water class If the surface elevation of a pixel is lower than 1 2 km above sea level then a water pixel has to fulfill the criteria p blue lt 0 20 and p blue gt p green 0 03 Pp NIR lt p green and p 1 6um lt Twater SWIR1 10 38 where Tivater sw1r1 is the water threshold reflectance in the SWIR1 band around 1 6 jum as de fined in the preference parameter file see chapter 9 3 Basically the gradient of the apparent water reflectance has to be negative If the pixel elevation is higher than 1 2 km the criterion of a negative gradient for the apparent re flectance does not properly work as the path radiance in the visible especially in the blue becomes small and the following criterion based on surface reflectance instead of apparent reflectance is used p NIR lt Twater NTR and p SWIR1 lt Twater SWIR1 10 39 where Twater NIR is the water reflectance threshold for the NIR band around 850 nm Equation 10 39 is also applied if any threshold Twater NIR OF Twater sWIR1 iS set to a negative value In this case the elevation criterion pixel below 1 2 km is overruled Saturated pixels These pixels fulfill the criterion DN blue gt T saturation 10 40 where DN blue is the digital number in a blue band around 470 nm and the threshold Tsaturation is defined in the preference paramete
214. n Fig 10 2 The radiometric calibration assigns to each digital number DN the corresponding at sensor radi ance L L k co k c1 k DN k 10 6 where k indicates the channel number and cy c are the calibration coefficients offset and slope For sensors with adjustable gain settings the equation is L k co k e1 k DN k 9 k 10 7 where g k is the gain setting in channel k The atmospheric correction has to be performed itera tively since the surface reflectance and large scale 0 5 1 km neighborhood background reflectance are not known So three steps are employed in the ground reflectance calculation CHAPTER 10 THEORETICAL BACKGROUND 149 L cy 0DN Figure 10 2 Schematic sketch of solar radiation components in flat terrain Step 1 The influence of the neighborhood adjacency effect is neglected and the surface reflectance is obtained from a r d co DN Lp To Eg pr 0 15 10 8 p where the spectral band index is omitted for clarity The factor d takes into account the sun to earth distance d is in astronomical units since the LUTs with path radiance and global flux are calculated for d 1 in ATCOR Step 2 The second step calculates the average reflectance in a large neigborhood of each pixel range R 0 5 1 km law P y2 gt Pij 10 9 ij l where N corresponds to the number of pixels for the selected range R of the adjacency effect 58 65
215. n as the pressure level or ground elevation is specified Other constituents vary slowly in time e g the CO2 concentration ATCOR calcu lations were performed for a concentration of 380 ppmv Ozone may also vary in space and time Since ozone usually has only a small influence ATCOR employs a fixed value of 330 DU Dobson units corresponding to the former unit 0 33 atm cm for a ground at sea level representing average conditions The three most important atmospheric parameters that vary in space and time are the aerosol type the visibility or optical thickness and the water vapor We will mainly work with the term visibility or meteorological range because the radiative transfer calculations were performed with the Modtran 5 code Berk et al 1998 2008 and visibility is an intuitive input parameter in MODTRAN although the aerosol optical thickness can be used as well ATCOR employs a database of LUTs calculated with Modtran 5 Aerosol type The aerosol type includes the absorption and scattering properties of the particles and the wave length dependence of the optical properties ATCOR supports four basic aerosol types rural urban maritime and desert The aerosol type can be calculated from the image data provided that the scene contains vegetated areas Alternatively the user can make a decision usually based on the geographic location As an example in areas close to the sea the maritime aerosol would be a logical choic
216. n bsq OUTPUT INAGE FILE Vdata atcor42 demo_data dais99 bar1_topo dais_bar1_atn_polish bsq J OVERWRITE AA ARAS PAS RICE O Output file already exists change name or press OVERWRITE 1 QUIT Figure 5 46 Radiometric spectral polishing Input to the spectral polishing program is the reflectance cube calculated with ATCOR It em ploys the vegetation index 0 lt NDVI lt 0 33 NDVI pnrr PRED PNIR PRED to mask soil pixels A soil spectrum is a slowly varying function of wavelength therefore a spectral smoothing will only remove spikes without disturbing the spectral shape Then the average re flectance spectrum over all soil pixels is calculated and smoothed with a 5 channel filter except for the atmospheric water vapor regions where a linear interpolation is performed The ratio of the filtered to the original soil spectrum is the spectral polishing function applied to all image pixels If zxx_atm bsq is the atmospherically corrected input image then zxx_atm_polish bsq is the pol ished output reflectance cube and the spectral polishing function is stored in xxr_atm_polish dat an ASCII file with two columns containing the center wavelength of each channel and the polishing factor Figure 5 46 shows the GUI panel 5 5 4 Spectral Smile Interpolation For sensors affected by spectral smile the surface reflectance cube is calculated accounting for the smile shift function in the column across t
217. n case of two targets a bright and a dark one should be selected to get a reliable calibration Using the indices 1 and 2 for the two targets we have to solve the equations L co DN L co c DNX 2 19 This can be performed with the co amp c option of ATCOR s calibration module see chapter 5 The result is L Lo E a OR 2 20 DN DN3 ee Cl CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 24 m r z B Me sf 7 SAS A z m q 6 fp 2 J 1d 2 fe p oft DAS 4 TAS a s fi gt targets least squores fit O 500 1000 1500 2000 DN digital number Figure 2 7 Radiometric calibration with multiple targets using linear regression co Li cx DN 2 21 Equation 2 20 shows that DN must be different from DN to get a valid solution i e the two targets must have different surface reflectances in each band If the denominator of eq 2 20 is zero ATCOR will put in a 1 and continue In that case the calibration is not valid for this band The requirement of a dark and a bright target in all channels cannot always be met Calibration with n gt 2 targets In cases where n gt 2 targets are available the calibration coefficients can be calculated with a least squares fit applied to a linear regression equation see figure 2 7 This is done by the cal_regress program of ATCOR It employs the rdn files obtained during the single target calibration the cl option of ATCOR
218. nd 0 no 1 yes 2 hcw quality file line 12 water vapor threshold to switch off the cirrus algorithm unit cm line 13 define saturation with factor b DN saturated gt b DN max b 0 9 to 1 line 14 include non linear influence of vegetation in water vapor calculation yes no line 15 start stop wavelengths for interpolation in the 940 nm region line 16 start stop wavelengths for interpolation in the 1130 nm region line 17 start stop wavelengths for interpolation in the 1400 nm region line 18 start stop wavelengths for interpolation in the 1900 nm region line 19 haze sun glint over water apparent NIR reflectance T clear Ta haze line 20 reduce over under correction in cast shadow 0 no 1 yes Note on the non linear influence of vegetation in water vapor calculations This option applies to the APDA regression algorithm and only if the 940 nm region is selected for the water vapor retrieval The retrieval is based on a linear interpolation across the absorption region and errors can occur due to the non linear behavior of the reflectance of vegetated surfaces in this region A simple empirical correction to the water vapor map W is applied using the NDVI calculated with the apparent reflectances in the red NIR channels W new W old 0 1 VDVI 0 7 cm 9 2 The correction is only performed for pixels with N DVI gt 0 25 and values NDVI gt 0 7 are reset to 0 7 Note on cut off limit for surface reflectance The d
219. nd a METADATA DIM The first file is intended for a quick overview the second file contains the complete set of specifications The absolute calibration gains for each band can be taken from either file and should be put into the corresponding cal file as they are In the METADATA DIM file the calibration gains are named PHYSICAL_GAIN The SPOT unit is 1 Wm sr umt but it is automatically converted into the ATCOR radiance unit The standard offset values are zero Occasionally however for SPOT 4 5 data a slightly negative offset has to be introduced for band 4 1 6 um in cases when the scene water reflectance is too high it should be close to zero CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 140 The geometry of data acquisition is described in the the METADATA DIM file The solar geometry is specified with the solar elevation and azimuth angle The sensor tilt geometry is defined by the incidence angle 6 at the earth s surface or the corresponding sensor tilt view angle Oy at the orbit altitude h see Fig 9 5 Both angles are specified in the METADATA DIM but the tilt angle is input to ATCOR The tilt view angle is not included in old versions of METADATA DIM but was added later For a given incidence angle the corresponding tilt view angle can be calculated as dy arcsin E sin 61 180 7 9 7 E h where Rg 6371 km is the earth radius and h 832 km is the SPOT orbit altitude Example incidence angles of
220. nd water vapor is necessary because these influence the values of path radiance transmittance and global flux e If the visibility is assumed too low optical thickness too high the path radiance becomes high and this may cause a physically unreasonable negative surface reflectance Therefore dark surfaces of low reflectance and correspondingly low radiance cy c1 DN are especially sensitive in this respect They can be used to estimate the visibility or at least a lower bound If the reflectance of dark areas is known the visibility can actually be calculated The interested reader may move to chapter 10 4 2 but this is not necessary to understand the remaining part of the chapter e If the main atmospheric parameters aerosol type or scattering behavior visibility or optical thickness and water vapor column and the reflectance of two reference surfaces are measured CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 18 the quantities Lpatn T p and Ey are known So an inflight calibration can be performed to determine or update the knowledge of the two unknown calibration coefficients co k c1 k for each spectral band k see section 2 4 Selection of atmospheric parameters The optical properties of some air constituents are accurately known e g the molecular or Rayleigh scattering caused by nitrogen and oxygen molecules Since the mixing ratio of nitrogen and oxygen is constant the contribution can be calculated as soo
221. nel Plot Sensor Response one may select the respective sensor response from within the available response functions in the ATCOR installation or elsewhere When selecting a response file the related bands are loaded automatically and the total number of bands is displayed The band range for display can be adjusted manually afterwards Normalization of the curves can be such that the area below the curves is constant same weight of the functions or the maximum is at 1 for all curves The displayed graph may be adjusted in appearance and size and finally being exported to a standard graphics file for further use IX Sensor Response Viewer 000 Choose any rsp file to diplay the related series of response curves select Sensors Response Vsrc_idl atcor atcor_23 sensor aster14_hs asterOL rsp Channels Bands from gt 0 a to E Normalization of Response 2 to Arear Default w to Maximum Done 005 X ATCOR Sensor Response Plot File Font_Size Display Output Response from asterd1 rsp eraa 0 5 1 0 15 2 0 Wavelength Figure 5 6 Plotting the explicit sensor response functions 5 1 6 Plot Calibration File When selecting this function the dialog defaults to the atcor installation for the selection of a cal file to be displayed Both gain and offset are then plotted in the same graph to get an overview of their relative values 5 1 7 Show System File This is the same function as Show Textf
222. ng the radiance unit Wm 2sr ym Since ATCOR uses the radiance unit mW em 2sr umt and the equation L c cDN 9 9 the calibration coefficients have to be calculated as co 0 1 bias 9 10 ci 0 1 gain 9 11 Note analysis of some DMC data from 2007 indicates that the specified bias in the NIR band is too high and better results are obtained if bias NIR 0 is employed 9 6 10 RapidEye The RapidEye constellation consists of 5 identical instruments in different orbits enabling a high temporal revisit time for any area The sensor has 5 multispectral bands covering the blue to NIR region with the specialty of a red edge band at 710 nm bandwidth 40 nm In addition the instruments can be tilted in the across track direction The nadir spatial resolution is 6 5 m The xml metafile contains information on the solar elevation angle illuminationElevationAngle solar azimuth illuminationAzimuthAngle and the view geometry i e the acrossTrackInci denceAngle and the view azimuth azimuthAngle ATCOR requires the sensor tilt angle 9y which is close to the across track incidence angle 9 on the ground The exact calculation can be done with eq 9 7 using the RapidEye orbit height 630 km CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 145 9 6 11 GeoEye 1 GeoEye 1 provides optical data with four multispectral channels in the 480 840 nm region with a spatial resolution of about 1 7 m
223. nith angle The diffuse flux spectrum Ef is evaluated for a surface reflectance of p 0 and the global flux for p 0 15 i e Ey Edir Eaif 0 1 s 0 15 where s is the spherical albedo The spectral band index is omitted for brevity For a flat terrain these fluxes are provided in the directory of the input file e g scene bsq e the direct spectral flux on the ground scene_edir dat e the diffuse spectral flux on the ground scene_edif dat for surface reflectance p 0 e the global spectral flux on the ground scene_eglo dat for a typical average surface reflectance p 0 15 These spectra will already give a realistic description for a flat terrain but they lack the dependence on the spectral reflectance variations in the scene Therefore an image of the global flux is also provided that accounts for the spatial reflectance and visibility water vapor patterns VIS named scene_eglobal bsq Edir VIS z y T Fais p 0 VIS x y I s x y p x y E x y 10 25 Here p indicates a spatial averaging with a filter size corresponding to the specified adjacency range The unit of the global flux is mWem um and it is stored as float data 32 bits pixel Therefore its file size will be twice or four times the size of the input scene if the scene is encoded as 16bit pixel and 8bits pixel respectively For a rugged terrain images of the direct and diffuse fluxes will be calculate
224. nomenclature and directional reflectance and emissivity Ap plied Optics Vol 9 1474 1475 1970 Parlow E Net radiation of urban areas Proc 17th EARSeL Symposium on Future Trends in Remote Sensing Lyngby Denmark 17 19 June 1997 pp 221 226 Balkema Rotterdam 1998 Riano D Chuvieco E Salas J and Aguado I Assessment of different topographic corrections in Landsat TM data for mapping vegetation types IEEE Trans Geoscience and Remote Sensing Vol 41 1056 1061 2003 Richter R Derivation of temperature and emittance from airborne multispectral thermal infrared scanner data Infrared Phys Technol Vol 35 817 826 1994 Richter R A spatially adaptive fast atmospheric correction algorithm Int J Remote Sensing Vol 17 1201 1214 1996 Richter R Atmospheric correction of satellite data with haze removal including a haze clear transition region Computers amp Geosciences Vol 22 675 681 1996 Richter R On the in flight absolute calibration of high spatial resolution spaceborne sensors using small ground targets Int J Remote Sensing Vol 18 2827 2833 1997 References 197 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 Richter R Correction of satellite imagery over mountainous terrain Applied Optics Vol 37 4004 4015 1998 Richter R Bandpass
225. nsor Gaussian Filter Files band01 rsp band02 rsp band96 rsp New Sensor RESLUT Atm LUTs atm ATCOR G Figure 5 11 Sensor definition files the three files on the left have to be provided created by the user a CHAPTER 5 DESCRIPTION OF MODULES 57 5 2 1 Define Sensor Parameters This panel is the first step if a new sensor is to be defined The panel as displayed in Fig 5 12 allows the below options Sensor Selection Select Select any of the already defined sensors from within ATCOR note a sensor is selected by on sensor file within a sensor directory New A new sensor is created within the ATCOR installation which results in a new directory in the sensor directory of the installation Rename The current sensor is renamed both directory and sensorx dat file Delete Allows to delete any sensor directory and all of its contents Inputs Sensor Type This is to be selected first smile sensor and thermal sensors require additional inputs as displayed in the panel Sensor Total FOV deg edge to edge FOV in across track direction in degrees Number of Across Track Pixels Nominal number of pixels in the unrectified data First last Reflective Band Band numbers starting at one none 0 First last Mid IR Band Band numbers starting at one none 0 First last Thermal IR Band Band numbers starting at one none 0 Th
226. nt of a specific sensor definition ie the sensor definition is editable and adjustable The ATCOR2 variant is recommended to speed up processing time and for fast checks as hyperspectral image processing may be very time consuming in rugged terrain 5 3 5 ATCOR3 User defined Sensors This routine is to be taken if highest accuracy is required in terrain for imaging spectroscopy in struments The functionality is analoguous as described for the other panels CHAPTER 5 DESCRIPTION OF MODULES Figure 5 18 Panel for DEM files _CanceL continue Figure 5 19 Panel to make a decision in case of a DEM with steps 63 CHAPTER 5 DESCRIPTION OF MODULES 64 760 740 Elevation m N N o 700 20 40 60 80 100 Pixel Number Figure 5 20 Influence of DEM artifacts on the solar illumination image Top illumination with low pass filtered DEM middle illumination based on original DEM bottom 100 m transsect using original DEM pixel size is 6 m CHAPTER 5 DESCRIPTION OF MODULES 65 5 3 6 SPECTRA module The SPECTRA module see figure 5 21 serves to extract spectra of different targets of the scene as a function of the visibility It is started from within one of the four possible ATCOR main panels These spectra can be compared to field spectra or library spectra to estimate the visibility Scene derived spectra also may indicate calibration errors in certain channels In that case a copy of the sensor calibration
227. o account to calculate deriva tive used to reconstruct the value of the center band Neighbour Derivative all spectral bands except for the center itself are taken into account to calculate derivative used to reconstruct the value of the center band Lowpass Filter Only the smoothing is performed no derivatives are calculated Savitzky Golay Filter to perform a numerical polinomial fit of 4th degree through the selected total window size Output A cube containing the spectrally filtered copy of the original image data cube is generated compare Paper Earsel SIG IS Workshop Edinburgh 2011 AAA 3 ATCOR Derivative Polishing setect Input File Name data hyperion Bern_02 Hyper ion_sub167 bsq Selec Sensor Spectral Response Ysrc_idl atcor atcor_23 sensor hyperion167 band001 rsp Number of polishing bands on each side 3 Smoothing Factor 0 no smoothing ay Polishing Filter Type y Neighbour Derivatives w Lowpass Filter w Savitzky Golay Define Polished Output Data Cube data huperion Bern_02 Hyper ion_sub167_poliish bsq Help Run Polishing Tone y ZA Figure 5 45 Statistical spectral polishing 5 5 3 Spectral Polishing Radiometric Variation A module that was originally developed for the airborne version of ATCOR is the spectral polishing The algorithm is only intended for hyperspectral imagery CHAPTER 5 DESCRIPTION OF MODULES 84 INPUT FILE sat Yaata7 atcor42 deno_data dais88 barl_topo dais_bar _at
228. oefficients Next the target box size and the corresponding ground reflectance file have to be specified The button for the file name of target 2 is insensitive because the single target mode was selected here Then the file name for the calibration results should be specified The default name is test cal However it is recommended to include the name of the ground target here Now the target s can be clicked in the zoom window that pops up automatically Target 1 has to be clicked with mouse button 1 mb1 left target 2 with mouse button 2 mb2 center The zoom window is moved in the main window by pressing mb1 for target 1 and mb2 for target 2 Alternatively the target coordinates x y column line can be specified In addition to the file xxx cal the files zxx rdn radiance versus digital number and xx1 adj original and adjacency corrected DN s are automatically created Select display bands file 0 ira del Visibility bn 35 0 Ata file perura Ped 4 Grom 5 Blue el miW cow or micrometer pogo a y Message I box center cocedinates x y 169 211 Create Zoon Window Contrast stretchirg Gaussian w Histo Ee MIN foo MA A w f Patun j Figure 5 23 Radiometric CALIBRATION module The appearance of the inflight calibration module is similar to the SPECTRA module In the left part the image is loaded A zoom window can be created and two contrast
229. ol 54 161 167 1995 Choudhury B J Synergism of multispectral satellite observation for estimating regional land surface evaporation Remote Sensing of Environment Vol 49 264 274 1994 Choudhury B J Ahmed N U Idso S B Reginato R J and Daughtry C S T Rela tions between evaporation coefficients and vegetation indices studied by model simulations Remote Sensing of Environment Vol 50 1 17 1994 Coll C Caselles V Rubio E Sospreda F and Valor E Temperature and emissivity separation from calibrated data of the Digital Airborne Imaging Spectrometer Remote Sens Environm Vol 76 250 259 2001 193 References 194 13 14 15 16 17 18 19 23 24 27 Coll C Richter R Sobrino J A Nerry F Caselles V Jimenez J C Labed Nachbrand J Rubio E Soria G and Valor E A comparison of methods for surface temperature and emissivity estimation In Digital Airborne Spectrometer Experiment ESA SP 499 p 217 223 Nordwijk Netherlands 2001 Corripio J G Vectorial algebra algorithms for calculating terrain parameters from DEMs and the position of the sun for solar radiation modelling in mountainous terrain Int J of Geographical Information Science Vol 17 1 23 2003 Crist E P and Cicone R C A physically based transformation of Thematic Mapper data the Tasseled Cap IEEE Trans
230. ol properties The water vapor retrieval over land is performed with the APDA atmospheric precorrected differential absorption algorithm 75 In its simplest form the technique uses three channels one in the atmospheric water vapor absorption region around 940 or 1130 nm the measurement channel the others in the neighboring window regions reference channels The depth of the absorption feature is a measure of the water vapor column content see figure 10 10 In case of three bands the standard method calculates the water vapor dependent APDA ratio as Lolo u Laplu 10 73 wi L1 p1 Lip w3 L3 p3 Lap where the index 1 and 3 indicates window channels e g in the 850 890 nm region and 1010 1050 nm region respectively Index 2 indicates a channel in the absorption region e g 910 950 nm L and Lp are the total at sensor radiance and path radiance respectively The symbol u indicates the water vapor column The weight factors are determined from w A A2 A3 A1 and w3 2 A1 As A1 10 74 Rappa p u Reflectance 0 90 0 95 1 00 Wavelength um Figure 10 10 Reference and measurement channels for the water vapor method The at sensor radiance is converted into an at sensor reflectance The problem is the estimation of the surface reflectance pa in the absorption band eq 10 73 The technique tries to estimate the reflectance pa with a linear interpolation of the surface reflec
231. ol type retrieval irrespective of the atm name in the inn file line 14 temfile atmospheric LUT file name thermal region empty if no thermal band CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 132 line 15 1 0 adjacency range km line 16 35 0 visibility km line 17 0 7 mean ground elevation km asl not used in case of rugged terrain where elevation file applies line 18 33 0 178 0 solar zenith solar azimuth angle degr line 19 10 0 150 0 off nadir sensor tilt angle sensor view azimuth angle degr For nadir looking sensors the tilt angle is zero and the view azimuth is an arbitrary value line 20 0 0 1 0 0 0 1 npref iwaterwv ihaze iwat_shd ksolflux ishadow icl_shadow seven parameters controlling the processing options npref 0 constant visibility npref 1 variable visibility based on dark reference areas in the scene npref 1 variable visibility for each sub image during batch job with tiling iwaterwv 0 no water vapor correction or no water vapor bands available iwaterwv 1 water vapor correction using bands in the 940 nm region iwaterwv 2 water vapor correction using bands in the 1130 nm region iwaterwv 3 940 and 1130 nm bands are employed e Haze removal is enabled by setting the parameter haze gt 0 no haze removal is specified with ihaze 0 Separate parameter values define haze removal over land haze sunglint removal over water
232. omething else e g snow or a specular reflection from a surface Only saturated pixels with a very high NDSI gt 0 7 are assigned to the snow class medium cloud probability coded 60 same as for low probability but with p blue gt 0 25 and p red gt 0 18 10 54 This is similar to the standard cloud assignment in the _out_hcw bsq file where p blue gt Te Te 0 25 or 25 and p red gt 0 15 The 25 reflectance threshold in the blue or green band is the default value in the preference parameter file e high cloud probability coded 90 same as for medium probability but with p blue gt 0 35 and p red gt 0 25 10 55 If a thermal band exists the relationships 10 43 must also be fulfilled Water probability CHAPTER 10 THEORETICAL BACKGROUND 164 The criteria for the water class are described in the previous section The following water probability rules are employed e low water probability coded 30 water pixels fulfilling the above criteria eq s 10 38 10 39 This is the water assignment in the _out_hcw bsq file e medium water probability coded 60 same as for low probability but with a more stringent NIR reflectance threshold If no SWIR1 band around 1 6 um exists the criterion is p NIR lt 0 04 10 56 Note the default threshold Tuater n1r is 0 05 or 5 in the reflectance percent unit defined in the preference parameter file yielding more low prob
233. omparison of the various solar reference functions is given Information on the IDL version of ATCOR can be found on the internet http www rese ch What is new in the 2012 version e In mountainous terrain the height resolution of the radiative transfer functions is improved from 100 m to 20 m e The skyview calculation now supports a user defined angular increment for azimuth elevation and a pixel undersampling factor The default azimuth elevation increment is 30 10 to be compatible with the previous version The new recommendated values are 10 and 5 for azimuth and elevation respectively For a large scene an undersampling factor of 3 is recommended to avoid an excessive calculation time CHAPTER 1 INTRODUCTION 13 The water vapor retrieval can be performed with channels in the 820 nm atmospheric absorp tion region This is a useful enhancement for VIS NIR spectrometers covering the spectrum from 400 900 nm The previous version only supported a retrieval if channels exist in the 940 nm or 1130 nm region Water vapor retrieval with band regression the previous version supported the calculation of the water vapor map for ground elevations up to 2 5 km now extrapolation is provided up to elevations of 3 5 km An external water vapor map is accepted for instruments that have the capability i e neces sary spectral bands to derive the water vapor map If scenel bsq is the name of the input scene and the name of the
234. omputed as the residual LE R G H 7 22 A different heat flux model is employed for urban areas with man made surfaces asphalt concrete roofs etc These are defined here with the reflectance criteria peso 0 10 and psso 0 10 and P650 gt P850 0 7 7 23 representing low vegetation indices with NDVI lt 0 176 This simple spectral definition is not unambiguous it might also apply to soils For urban areas the latent heat is usually very small and the fluxes G and H dominate Therefore the terms G LE and H are approximated by the following three equations Parlow 1998 G 0 4 Rn 7 24 LE 0 15 Rn G 7 25 H R G LE 7 26 For low vegetation indices SAVI lt 0 1 the ground heat flux G from equation 7 17 i e the vegetation model agrees well with G from equation 7 24 i e the urban model However major CHAPTER 7 VALUE ADDED PRODUCTS 114 differences exist for the LE and H terms see table 7 1 Parameters for this table are E 800 Ry 600 Ratm Rsurface 100 Wm Ts 30 C and Ta 20 C The veg and urb indicate the heat fluxes derived from the vegetation and urban model respectively For the urban surfaces asphalt concrete the G veg H veg and LE veg values are given in brackets for comparison but the corresponding urban heat fluxes are valid because the urban criterion equations 7 23 p650 gt 0 10 p850 gt 0 10 and p650 gt Paso 0 7 applies The l
235. on as opposed to the physically based approach of ATCOR Nevertheless the latter approach also cannot avoid problems in faintly illuminated areas Therefore it is supplemented by an empirical method with three adjustable parameters 3r b and g as explained below This approach was tested on different rugged ter rain scenes with vegetated and arid landscapes and usually yields satisfactory results It reduces overcorrected reflectance values starting at a threshold local solar zenith angle Grp greater than the scene s solar zenith angle Os Equation 10 80 defines the implemented basic geometric correction CHAPTER 10 THEORETICAL BACKGROUND 176 function which depends on the local solar incidence angle solar illumination and the threshold angle Gr The exponent b 1 3 1 2 3 4 or 1 is the second parameter and can be selected by the user Some guidelines on the choice of b are are discussed below The third adjustable parameter is the lower bound g of the correction function see Figure 10 13 G cos cosBr gt g 10 80 The threshold illumination angle Gr should have some margin to the solar zenith angle to retain the original natural variation of pixels with illumination angles close to the solar zenith angle The threshold angle can be specified by the user and the following empirical rules are recommended e Br b 20 if O lt 45 e If 45 lt 0 lt 20 then Gr 6 15 e If 0 gt 55 then Gr 6 10
236. ontal coordinates corresponding to the georeferenced pixel positions Z vertical coordinate containing the elevation information from the DEM DN x y digital number of georeferenced pixel Lp z 0 P path radiance dependent on elevation and viewing geometry lO ground to sensor view angle transmittance direct plus diffuse components Ts z Sun to ground beam direct transmittance B x y angle between the solar ray and the surface normal illumination angle b x y binary factor b 1 if pixel receives direct solar beam otherwise b 0 E extraterrestrial solar irradiance earth sun distance d 1 astronomical unit Ej z y z diffuse solar flux on an inclined plane see equation 10 18 Ela global flux direct plus diffuse solar flux on a horizontal surf at elevation z CHAPTER 10 THEORETICAL BACKGROUND 151 i y Sky and terrain view factors trigonometric approach horizon line approach d 1 24 Adleceasy Netghbourkoo Figure 10 3 Radiation components in rugged terrain sky view factor Left schematic sketch of radiation components in rugged terrain 1 path radiance 2 pixel reflected radiance 3 adjacency radiance 4 reflected terrain radiance Right sky and terrain view factor Elz radiation incident upon adjacent slopes i 0 1 initial value of average terrain reflectance Doe enim 2s y locally varying average terrain reflectance calculated iteratively i 1 2 3 Vel 2Y terr
237. optical thickness atmospheric transmittance The atmospheric direct or beam transmittance for a vertical path through the atmosphere can be calculated as T e 2 4 Fig 2 1 right shows an example of the atmospheric transmittance from 0 4 to 2 5 um The spectral regions with relatively high transmittance are called atmospheric window regions In absorbing regions the name of the molecule responsible for the attenuation of radiation is included Apparent reflectance CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 16 The apparent reflectance describes the combined earth atmosphere behavior with respect to the reflected solar radiation md L 2 E cos 25 p apparent where d is the earth sun distance in astronomical units L co cl DN is the at sensor radiance co C1 DN are the radiometric calibration offset gain and digital number respectively E and 0 are the extraterrestrial solar irradiance and solar zenith angle respectively For imagery of satellite sensors the apparent reflectance is also named top of atmosphere TOA reflectance 2 1 Radiation components We start with a discussion of the radiation components in the solar region i e the wavelength spectrum from 0 35 2 5 wm Figure 2 2 shows a schematic sketch of the total radiation signal at the sensor It consists of three components 1 path radiance L1 i e photons scattered into the sensor s instantaneous field of view with out
238. or to edit plain text ASCII files Plotting the explicit sensor response functions ee 35 LIST OF FIGURES 5 7 5 8 5 9 5 10 pala 5 12 5 13 5 14 5 15 5 16 5 17 5 18 5 19 5 20 5 21 5 22 5 23 5 24 5 25 5 26 Dat 5 28 5 29 5 30 5 31 5 32 5 33 5 34 5 35 5 36 5 37 5 38 5 39 5 40 5 41 5 42 5 43 5 44 5 45 5 46 5 47 5 48 5 49 5 90 5 51 5 52 5 93 Plotting a calibration fle 2 24 2445844 be ASG Ree a we ee EG Displaying a calibration file same file as in Fig 5 7 o Panel to edit the ATCOR preferences e The New Sensor Meta 2 6 2844 4 be Ewe Oe ER eR Re we Re SE Sensor definition files the three files on the left have to be provided created by the WO ada aaa oe ea he ida eR Oe ee oe be AER BGs SS Definition of anew sensor s ccor ccs e kpe doe Siapa a Ra Daoa p a Spectral Filter Creation 4 6 54 4884 4 uta aa aaa os Black body function calculation panel ooa aa ee Panels of RESLUT for resampling the atmospheric LUTS The Atm Correction Menu 2 aa a aa AOU panel oir aca Daria a aa oes a he BR a Paneltor DEM Des aid S ace ee Be a Ade eee ee A e E a Panel to make a decision in case of a DEM with steps Influence of DEM artifacts on the solar illumination image SPECTRA Module vomi ce Sh 84 a a op E ee et e a E Radiometric calibration target specification panel sooo oa o
239. orrected reflectance values in faintly illuminated areas having small values of cos Several approaches have been pursuit to solve this problem e an empirical coefficient C is calculated based on a regression of brightness values and the local ilumination angle derived from the DEM The coefficient depends on scene content and wavelength 85 54 e the sun canopy sensor SCS geometry is employed in forested terrain instead of the solely terrain based geometry 25 e the SCS method is coupled with the C correction 83 These approaches produced good results on sample scenes with uniform cover types presented in the above papers When applying the methods to a wider range of areas some of the practical problems are e mountainous scenes often contain a number of different covers e g deciduous forest conif erous forest mixed forest shrubs meadow rocks etc e the computation of the C coefficients for different surface covers would require a pre classi fication e the correlation obtained for the C coefficients is often less than 0 7 yielding unreliable results with this method These remarks are supported by reference 54 These authors applied different correction ap proaches to a TM scene containing different cover types and noted that there is no optimum method for all cover types A drawback of the Minnaert and empirical C methods is that they do do not distinguish between the direct and diffuse solar illuminati
240. ort data data atcor2 3 atm_database_th Number of files to be converted 24 File 1 of 24 File 10 of 24 File 20 of 24 File 24 of 24 DONE time 40 sec All bp files converted on export data data atcor2 3 atm_database_thu2003_RS Output directory also contains reference irradiance e0_solar_thu2003_RSL_ku2005_04nm dat sS C gt QUIT Figure 9 3 User interface to convert database from one to another solar irradiance irradiance source 9 2 Supported I O file types The input image to ATCOR must have the band sequential BSQ ENVI format or the TIFF format Some restrictions apply to the TIFF format as detailed below Several data types exist for the encoding The following data types of an input image are supported e byte or unsigned 8 bit integer ENVI data type 1 e signed 16 bit integer ENVI data type 2 e unsigned 16 bit integer ENVI data type 12 e signed 32 bit long integer ENVI data type 3 e float 32 bit ENVI data type 4 In case of the float input data type some restrictions exist for the image processing options The haze and cloud shadow removal cannot be selected because they rely on integer based histogram manipulations The user should re scale the float data to one of the supported integer data types in this case CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 126 Resampling from high spectral resolution database export data data atcor2 3 at
241. ot a perfect shadow transformation it is much better to restrict its application to the potential most likely shadow areas This is an important processing step to reduce the number of mis classifications or false alarms If omitted it will cause strange shadow pixels scattered all over the image An example can be found in the central part of Fig 10 21 where the standard shadow map contains a lot of artifact shadow areas Therefore the proposed method tries to find the core shadow areas in a scene and subsequently expands the core regions to obtain the final mask that includes a smooth shadow clear transition The physically scaled shadow function is then applied only to the pixels in the final mask The histogram of the unscaled shadow function can be employed to separate regions of low values of Y from the moderate to high values compare Fig 10 20 A threshold Py can be set in the vicinity of the local histogram minimum and the core shadow mask is defined by those pixels with z y lt Pr The details of the choice of Py are discussed below As always with thresholding some arbitrariness is involved in the final selection CHAPTER 10 THEORETICAL BACKGROUND 186 histogram 1 0 0 8 0 5 0 2 0 0 Ls 1 0 0 5 0 0 0 5 1 0 unscaled shadow function Figure 10 20 Normalized histogram of unscaled shadow function Once the core shadow mask has been defined it is expanded to include the surrounding shad
242. over water ALOS AVNIR2 o 2 000052 02s 180 10 17Scatterplot of apparent reflectance of cirrus 1 38 um band versus red band 182 10 18Sketch of a cloud shadow geometry soosoo e ee 183 10 19Flow chart of processing steps during de shadowing 00 184 10 20Normalized histogram of unscaled shadow function 00 4 186 10 21Cloud shadow maps of a HyMap scene a 187 10 22De shadowing of a Landsat 7 ETM scene e 189 10 23 Weighting of q function for reference pixels 190 10 24Examples of template reflectance spectra employed by the SPECL code 191 List of Tables 4 1 4 2 4 3 4 4 Fel 9 1 9 2 9 3 10 1 10 2 Al A 2 A3 AA A 5 A6 A T Example of a sensor definition file no thermal bands Sensor definition file instrument with thermal bands Sensor definition file smile sensor without thermal bands Class label definition of hew file os c esios soc eais Dh kke Heat fluxes for the vegetation and urban model 0 0 00000 Elevation and tilt angles for Ikonos o o e e Elevation and tilt angles for Quickbird o e Radiometric coefficients cl for ASTER o e 0000220004 Class labels in the hew file 2 2 e Visibility iterations on negative reflectance pixels red NIR bands
243. ow clear transition zone of 100 m width De shadowing with the scaled shadow function is then exclusively applied to the pixels in this final mask This means the direct solar flux Eqir term in eq 10 98 has to be multiplied with z y m d eo i c i DNi a y Lp lidia y Ej In equations 10 98 10 104 the aerosol optical thickness or visibility required for the atmospheric terms path radiance transmittance direct and diffuse flux can be derived from the image pro vided the necessary bands in the visible and shortwave infrared region exist and the scene contains dark reference areas 39 Otherwise the user has to specify an estimated visibility The second important atmospheric parameter is the water vapour column For instruments with bands in the atmospheric water vapour regions this information can be derived from the image data 75 otherwise an estimate has to be provided by the user In summary three channels around 0 85 1 6 and 2 2 um are used to define a matched filter vector with three elements per pixel For each image pixel the surface reflectance in these three channels and the scene average reflectance of these channels are calculated to obtain the unscaled shadow function and finally the scaled shadow function The same shadow function is employed to de shadow the imagery not only in the initial three channels but for all channels of the sensor eq 10 104 10 104 pi x y Details of
244. owse Manual Opens this manual in the default PDF display application of your machine Web Resources Opens the html document atcor3_webresources htm in the systems default ap plications for viewing HTML documents About Provides basic information about the copyright and also displays the build number of the software please provide the build number for debugging purposes in case the ATCOR support is contacted Your License Provides information about the licensed features in your license key and also dis plays which license key is currently active Chapter 6 Batch Processing Reference For most ATCOR modules a convenient graphical user interface is available but batch jobs can also be submitted A detailed discussion of the interactive panel driven modules is given in chapter D 6 1 Using the batch mode ATCOR can process scenes in the batch mode For large scenes the tiling option is also available in batch mode which splits a big scene into a number of smaller sub scenes processes the sub scenes and finally merges them into one file A prerequite for the tiling is that enough IDL memory is available to keep one image channel and the sub scene channel in memory Also note that a full IDL developer license is required in order to make use of the batch commands on the IDL prompt The batch mode can be accessed after the processing parameters have been specified in the inter active graphical user interface GUI panel i e after the SP
245. pectral satellite sensors in the atmospheric window regions espe cially over land because land scenes are spatially inhomogeneous and this type of cloud is partially transparent On the other hand water vapor dominates in the lower troposphere and usually 90 or more of the atmospheric water vapor column is located in the 0 5 km altitude layer Therefore CHAPTER 10 THEORETICAL BACKGROUND 181 if a narrow spectral band is selected in a spectral region of very strong water vapor absorption e g around 1 38 um or 1 88 um the ground reflected signal will be totally absorbed but the scattered cirrus signal will be received at a satellite sensor or a sensor in a high altitude aircraft e g 20 km AVIRIS scenes So a narrow channel at 1 38 um is able to detect cirrus clouds and if a correlation of the cirrus signal at this wavelength and other wavelengths in the VNIR and SWIR region can be found then the cirrus contribution can be removed from the radiance signal to obtain a cirrus corrected scene The basic ideas of cirrus correction were presented in several papers 19 20 22 71 The algorithm differs for water and land pixels For water a scatterplot of the 1 38 um versus the 1 24 pum channel is used for land the band correlation is determined from a scatterplot of the 1 38 um versus a red channel around 0 66 wm To obtain a high sensitivity only vegetation pixels are taken because they have a low reflectance in the red spectra
246. pends on other factors such as the overall CPU load in case of multi user machines or the traffic on net worked machines Accordingly the estimate for the remaining time is not always continuously decreasing but may increase sometimes CHAPTER 5 DESCRIPTION OF MODULES 75 Figure 5 32 Value added panel for a flat terrain Figure 5 33 Value added panel for a rugged terrain CHAPTER 5 DESCRIPTION OF MODULES 76 Figure 5 34 LAI FPAR panel Figure 5 35 Job status window CHAPTER 5 DESCRIPTION OF MODULES TT 5 3 12 Start ATCOR Process Tiled from inn This is a way to start a tiled process of ATCOR from within the ATCOR GUI instead of the standard batch based process atcor_tile The process requires that an x inn file has been created before by going through the atcor GUI or by editing a respective ASCII file manually or by using the routine write_atcor_inn_file pro provided in the directory docu of the ATCOR installation The below parameters are to be entered for processing Input file name name of image data cube to be processed The file must be accompanied by a valid x inn file for processing Name of output cube to be created Number of tiles in X and Y dimensions the total number of tiles to process is then X x Y tiles ATCOR method selection between flat processing and rugged processing for the latter the DEM has to be prepared and ready After starting the process using t
247. ption features to detect and remove possible wavelength calibration errors see chapter 2 3 For this purpose a certain number of target spectra have to be selected in the SPECTRA module Input to the spectral calibration module are the DN spectra of selected fields saved as ASCII files in the SPECTRA module by pressing the button Save last spectrum The files should be numbered consecutively starting with a name such as location_target1 without extension The next target has to be named location_target2 etc For each target field three files will generated for example e location_target1 dat contains surface reflectance spectrum e location targetl txt contains target coordinates and processing parameters visibility water vapor column etc e location target1_dn1 dat contains the DN spectrum For a given location scene up to 9 targets can be extracted and used for the spectral calibration The geometry scene visibility and average water vapor content of the targets enter as parameters to the spectral calibration see Fig 5 57 The water vapor content has to be averaged from the values found in the location target txt files The first target DN file has to be selected by the user the remaining target files are automatically found provided the nomenclature with a consecutive numbering is applied The result of the spectral calibration are files with the spectral shifts per spectrometer and
248. puts ENVI file s smoothed or filtered by the given factor and method Smooth uses the standard smoothing i e lowpass filter in the spatial domain CHAPTER 5 DESCRIPTION OF MODULES 81 Figure 5 40 Example of a DEM left with the corresponding sky view image right DEM File may have 16 or 32 bit integer ar Float data Input IEN FILE auto os data otecr2 3 dero data ta_rugged tn biferest ele Quit Output Shadow File Vauto_os dsta7 atcor2 3 deno date tn ruaged ta blForest zent9_azil47_chd 4 OVERMRITE Solar zenith angle degree 491 Solar azimuth angle degree Oenerth H east etc DEM resolution x y pixel size meters eo encia m ee RUN ax _ Figure 5 41 Panel of SHADOW Median uses a median filter for data correction e g to remove noise or outliers from the DEM ATTENTION The _ilu file is not smoothed automatically by this routine If the ilu has already been calculated before it should be either removed or be smoothed directly OOP Selec Input DEM File Name Ysrc_idl atcor atcor_23 deno_data tn rusged tn_blforest_ele bsq Dimensions 500 500 Diameter of DEM Filter Pixels E Output Name of Filtered DEM Ysrc_id1 atcor atcor_23 deno data tn rusced tn blforest_sn3_ele bsq Help Smooth Median Done y ZA Figure 5 42 Panel of DEM smoothing CHAPTER 5 DESCRIPTION OF MODULES 82 5 5 Menu Filter The Filter m
249. r 0 1 is required to convert into the ATCOR radiance unit i e co 0 1 B and c 0 1 G For the thermal band two files e g xxx_nn61 tif and xxx_nn62 tif are included per scene the 61 indicates the low gain the 62 indicates the high gain data Either one can be selected for ATCOR but an update of the corresponding thermal bias and gain from the meta file is required in the radiometric calibration cal file Notice the standard negative offset values often lead to negative surface reflectances for dark tar gets therefore in many cases the magnitude of the negative offset has to be decreased typically by a factor 2 ETM bands 1 4 Note concerning Landsat 4 5 TM A difficult topic there is no standard header format with metadata Different formats existed in the past depending on Landsat processing station and year of processing The radiometric calibration is varying as a function of the day after launch compare references 82 33 A file cal_gain_table_kamstrup_hansen dat is available on the directory atcor cal landsat4_5 contain ing the gain cl values for 1985 2005 for bands 1 4 calculated with the regression equation of 33 Other publications deviate 10 20 from the Kamstrup Hansen cl values However for reprocessed NLABS LPGS Landsat 4 or Landsat 5 TM data equations 9 4 to 9 6 are also valid 9 6 2 SPOT The metadata is specified in two files a VOL_LIST PDF a
250. r an 8 bit pixel data encoding we obtain T 0 9 255 230 and pixels with a grey value greater than 230 are flagged as truly or potentially saturated Although the saturation nominally starts at DNmaz some sensors already show a non linear behavior around 0 9 DNmaz so the factor 0 9 is a precaution to be on the safe side This saturation check is performed for two channels in the visible region blue band around 470 500 nm and a green band around 550 nm For multispectral sensors the blue band is usually the critical one concerning saturation For hyperspectral sensors with many blue bands the one closest to 450 nm is taken Although the haze cloud water file contains saturated pixels based on two visible bands the percentage of saturated pixels for all bands will be given in the corresponding log file However the check of the blue and green channel normally captures all saturated pixels Note cloud or building shadow pixels are not included here they are stored separately file image_fshd bsq 4 9 Processing of multiband thermal data Several options have been implemented to process multiband thermal data see chapter 10 1 4 for details Apart from the final products surface temperature and emissivity intermediate products are available such as surface radiance at sensor blackbody temperature and surface blackbody temperature The intermediate products might be useful to trace back spectral or radiometric problems If
251. r can have a similar spectral reflectance behavior as haze so the clear water threshold is scene dependent In addition the upper threshold defining haze or sun elint might be scene dependent However the default values usually provide good results and a solid basis for a possible iteration of these two parameters Figure 10 16 presents an example of haze removal over water with the two default values of Ti clear 0 04 and To haze 0 12 The de hazing over water is successful to a large ex tent however some artifacts appear close to the land border image center where haze pixels over water are classified as land or cloud This is due to a simple spectral classification of the land water mask an external water map would lead to better results Figure 10 16 Haze removal over water ALOS AVNIR2 true color image northern Germany 16 April 2007 Left part of original scene right after haze removal 10 5 4 Cirrus removal On the first glance images contaminated by cirrus appear similar to hazy scenes discussed in the previous section However haze usually occurs in the lower troposphere 0 3 km while cirrus clouds exist in the upper troposphere and lower stratosphere 8 16 km The effect of boundary layer haze can be observed in the visible region but seldom in longer wavelength channels gt 850 nm However cirrus also affects the NIR and SWIR spectral regions Thin cirrus clouds are difficult to detect with broad band multis
252. r emissivity can also be included in the processing Usually isolated point like measurements of air temperature are available from meteorological sta tions These have to be interpolated to generate a spatial map coregistered to the image prior to applying the ATCOR model Data in the file containing the air temperature must have the Celsius unit data of the air emissivity file must range between 0 and 1 Future improvements to the ATCOR model will include an air temperature map derived from the image triangle or trapezoidal method employing the thermal band surface temperature and NDVI Carlson et al 1995 Moran et al 1994 In case of mountainous terrain the air temperature Ta z0 and water vapor partial pressure Pwv 20 at a reference elevation z have to be specified The height dependence of air temperature is then obtained with linear extrapolation employing a user specified adiabatic temperature gradient OT Oz Tale Talo E 20 2 7 28 where 07 0z is typically in the range 0 65 0 9 Celsius 100 m The water vapor partial pressure is extrapolated exponentially according to Pw Z Pwo Z0 107 70 2s 7 29 CHAPTER 7 VALUE ADDED PRODUCTS 115 where z is the water vapor scale height default 6 3 km The list of all output channels of the value added flx bsq file is 1 2 10 11 Soil adjusted vegetation index SAVI scaled with factor 1000 Leaf area index LAI scale
253. r file If a blue band does not exist a green band around 550 m is used as a substitute If a green band also does not exist a red band around 650 nm is used Tsaturation b encoding default b 0 9 e g 0 9 255 230 for 8 bit sensors with encoding 255 The value b 0 9 is used as default instead of the obvious b 1 0 because saturation or nonlinear effects often occur at these lower radiance levels As saturation usually occurs in the blue to red part of the spectrum channels in this region are checked and assigned to the class saturated false color coded red in the _out_hcw bsq file However the _atm log file contains the percentage of saturated pixels for each channel Cloud over land CHAPTER 10 THEORETICAL BACKGROUND 161 Pixels must satisfy the conditions p blue gt T and p red gt 0 15 and p NIR p red lt 2 and p NIR gt 0 8 p red and p NIR p SWIR1 gt 1 and NDSI lt 0 7 or DN blue gt Tsaturation 10 41 where p blue is the apparent reflectance in a blue band Te is the cloud threshold as defined in the preference parameter file and DN blue is the corresponding digital number If no blue band is available a green band around 550 nm is taken as a substitute If no green band exists a red band around 650 nm is taken NDSI is the normalized difference snow index p green p SWIR1 NDSI a p green p SWIR1 Note that saturated pixels in visible bands are
254. r imagery over the land J Geophys Res Vol 102 D14 17 173 17 186 1997 Ackerman S A Strabala K I Menzel W P Frey R A Moeller C C and Gumley L E Discriminating clear sky from clouds with MODIS J Geophys Res Vol 103 D24 32 141 32 157 1998 Krause K Radiance conversion of QuickBird data Technical note RS_TN_radiometric_radiance_4002 http www digitalglobe com Digital Globe 1900 Pike Road Longmont CO 80501 USA 2003 Kriebel K T Measured spectral bidirectional reflection properties of four vegetated sur faces Applied Optics Vol 17 253 259 1978 Mouroulis P Green R O and Chrien T G Design of pushbroom imaging spectrometers for optimum recovery of spectroscopic and spatial information Applied Optics Vol 39 2210 2220 2000 Moran M S Clarke T R Inoue Y and Vidal A Estimating crop water deficit using the relation between surface air temperature and spectral vegetation index Remote Sensing of Environment Vol 49 246 263 1994 Murray F W On the computation of saturation vapor pressure J Applied Meteorology Vol 6 203 204 1967 Stamnes K Tsay S C Wiscombe W J and Jayaweera K Numerically stable algorithm for discrete ordinate method radiative transfer in multiple scattering and emitting layered media Applied Optics Vol 27 2502 2509 1988 Nicodemus F E Reflectance
255. r water pixels using the apparent reflectance in the NIR band Pixels are labeled as clear if p NIR lt Ti clear clear pixels 10 87 The default value is T clear 0 04 i e 4 The value is one of the editable preference parameters see chapter 9 3 Thin haze over water is defined as T clear lt p NIR lt 0 06 thin haze 10 88 Medium haze over water is defined as 0 06 lt p NIR lt To haze medium haze 10 89 The default value is T2 haze 0 12 i e 12 This value is also one of the editable preference parameters The third step is a linear regression between haze pixels in the NIR band and each other reflective band The regression is iterated with only those pixels deviating less than half a standard deviation from the average If a and 3 denote offset and slope of the regression line respectively the de hazed pixel for each channel j can be calculated as DN corrected j DN original j aj 8 DNnir DN clear j 10 90 where DN clear j is the average of all clear water pixels in channel j The same technique is also employed to remove sun glint The main problem is the specification of the clear water threshold CHAPTER 10 THEORETICAL BACKGROUND 180 If the threshold is too low clear water pixels are included in the haze mask if it is set too high haze or sun glint pixels will be included in the clear pixel class There is no unique solution because sandy bottoms over shallow wate
256. rack direction of the detector array The smile shift is specified as a 4th order polynomial function i e the file smile_poly_ord4 dat in the corresponding sensor folder see chapter 4 7 Due to the smile shift the wavelength values of a spectral channel vary slightly in across track direction The smile interpolation function allows the specification of a common center wavelength for each channel Then for each channel all pixel reflectances are interpolated to this new reference wavelength Since the smile shift between adjacent bands does not vary significantly a linear interpolation can be applied If A denotes the center wavelength of band i and column j and p 7 the surface reflectance of a column j pixel then the new interpolated reflectance is Ares i Aj 2 e 1 pj i 1 AG 1 3 G 1 new Pj i Aref 4 pj i F 5 1 where Ares 1 is the user defined reference center wavelength for band i There are three options for the reference wavelength grid 1 use wavelength corresponding to the center of the detector array 2 use average wavelength over all detector columns per band 3 use nominal wavelength specified in the ENVI header of the reflectance cube CHAPTER 5 DESCRIPTION OF MODULES 85 This tool is available in the interactive mode main menu then Filter then Spectral Smile Interpolation Image Cube and in the batch mode smile_interp3_batch see chapter
257. ral_calibration_results txt where the wavelength shift is listed for each spectrometer and each target The final shift is taken as the average of all target wavelength shifts In addition a new wavelength file sensor_new wul is created containing the channel center wavelengths and the FWHMs bandwidth as full width at half maximum A copy of the original radiometric calibration file e g xxx cal is provided for convenience e g xxx_new cal which contains the original radiometric calibration coefficients and the updated channel center wavelengths In case of originally non Gaussian filter curves the output FWHM values of sensor new wul rep resent the equivalent Gaussian FWHM values even though the spectral re calibration is based on the original non Gaussian filter curves The corresponding sensor with the new spectral calibra tion has to be added to the list of existing sensors see chapter 4 5 to process imagery with the updated spectral calibration In case of non Gaussion filter curves the original channel response files band rsp should be copied to the new sensor directory applying the appropriate wave length shifts For sensors with Gaussian filter curves the gauss_rsp module see chapter 5 can be applied to the sensor_new wul file to generate the corresponding band rsp files Note that a change of the spectral calibration usually requires a radiometric re calibration CHAPTER 5 DESC
258. rate water from the surface LE is usually obtained as the residual to balance the net radiation with the dissipation terms Net radiation is expressed as the sum of three radiation components Rn Reolar Ratm Reur face 7 8 where Rsolar is the absorbed shortwave solar radiation 0 3 3 wm or 0 3 2 5 wm Ratm is the longwave radiation 3 14 um emitted from the atmosphere toward the surface and Reur face is the longwave radiation emitted from the surface into the atmosphere Downwelling radiation is CHAPTER 7 VALUE ADDED PRODUCTS 111 counted with a positive sign the upwelling thermal surface radiation has a negative sign The absorbed solar radiation can be calculated as 2 5um Besar 1 By A dd 7 9 0 3um where p A is the ground reflectance 1 p A is the absorbed fraction of radiation and Ey A is the global radiation direct and diffuse solar flux on the ground The numerical calculation of equation 7 9 is based on the same assumptions regarding the extrapolation of bands and interpolation of gap regions as discussed in chapter 7 1 dealing with the surface albedo If the satellite imagery contains no thermal band s from which a map of ground temperature can be derived then Rsolar is the only surface energy component that can be evaluated In case of flat terrain with spatially varying visibility conditions or rugged terrain imagery a map of the global radiation is included as an additional value added channel
259. re 5 31 The function G for soil sand is applied with a wavelength independent exponent b After testing a large number of vegetated mountainous scenes two vegetation modes were finally selected because of their good performance 1 b 0 75 for channels with A lt 720 nm and b 0 33 for A gt 720 nm weak correction 2 b 0 75 A lt 720 nm and b 1 A gt 720 nm strong correction In most of the tested cases the first mode was appropriate A simple criterion vegetation index P850nm P660nm gt 3 is used to distinguish soil sand and vegetation The right part of Figure 10 13 shows the effect of shifting the threshold illumination angle Br For larger values of Gr the decline of function G starts later with a larger gradient and the lower bound g is met at slightly higher values of In most cases g 0 2 to 0 25 is adequate in extreme cases of overcorrection g 0 1 should be applied 10 5 2 Haze removal over land In many cases of satellite imagery the scene contains haze and cloud areas The optical thickness of cloud areas is so high that the ground surfaces cannot be seen whereas in hazy regions some information from the ground is still recognizable In ATCOR the scene is partitioned into clear hazy and cloud regions Here we will treat the low altitude boundary layer 0 3 km haze as opposed to high altitude cirrus Thin boundary layer haze can be detected with broad band mul tispectral instruments while a detect
260. re clustered around this line and its slope represents the correlation coefficient y the blue line represents the first of several iterations Papers on the cirrus algorithm often restrict eq 10 94 to the wavelength interval 0 4 lt A lt 1 um but we will extend this relationship into the SWIR region Substituting eq 10 94 into eq 10 93 yields Te A PA A pel1 38um 10 95 Neglecting the cirrus transmittance Te i e setting Te 1 we obtain the cirrus path radiance corrected apparent reflectance image index cc PeclA A pel1 38um y 10 96 CHAPTER 10 THEORETICAL BACKGROUND 182 ACIR band 0 01 A A E A 0 00 0 05 0 10 0 15 0 20 p RED band Figure 10 17 Scatterplot of apparent reflectance of cirrus 1 38 um band versus red band As the cirrus is almost on top of the atmosphere we have p 1 38u4m pi 1 38u4m and the appar ent cirrus reflectance can be calculated with eq 10 91 Cirrus removal is conducted as the first step during atmospheric correction followed by the aerosol and water vapor retrievals If the average water vapor column W of a scene is less than some threshold default W 0 6 cm then the cirrus removal algorithm is switched off to avoid a misinterpretation of bright surfaces as cirrus in the 1 38 wm channel Normally atmospheric water vapor completely absorbs surface features in the 1 38 wm channel but the channel might become partly transparent to surface f
261. red band vegetation water and NIR band water It is assumed that the lowest reflectance in the red band is 0 01 1 percent and 0 0 in the NIR band Therefore the obtained visibility value usually can be considered as a lower bound The higher visibility value of AEROSOL TYPE and VISIB ESTIMATE is recommended as a start visibility for the SPECTRA module CHAPTER 4 WORKFLOW 33 The Inflight Calibration is described in chapter 5 3 9 This module is usually not required for first use of the software The WATER VAPOR button can be used to test the appropriate band combinations for the retrieval of a water vapor map without a calculation of the surface reflectance cube Th button IMAGE PROCESSING starts the atmospheric correction process from the entered pa rameters A series of sequential panels are displayed after this button is pressed Fig 4 7 shows the panel with the image processing options Some options may not be accessible they are blocked if the required spectral bands are missing In case of a mountainous terrain the ATCOR3 button has to be selected Fig 4 1 Atm Correction This panel is similar to Fig 4 6 but an additional panel for the specification of the DEM files will appear Fig 4 8 The user has to provide the DEM file matched to the size of the input image The slope and aspect files can be calculated from the corresponding module under Topographic Fig 4 1 These two files may need
262. rement to enable an accurate interpolation ATCOR RESLUT gt tral Database gt 10MB 0 4 nm MODTRANS gt spec Figure 9 1 Monochromatic atmospheric database The database comprises calculations performed for a satellite altitude of 650 km but for consistency with the airborne ATCOR the symbolic height 99 000 is used in the file names MODTRAN s mid latitude summer atmosphere was used for the air pressure and temperature height profiles at six water vapor columns W 0 4 1 0 2 0 2 9 4 0 and 5 0 cm or or g cm sea level to space values and different aerosol types These represent dry to humid atmospheric conditions 59 6 They are needed for the water vapor retrieval to create interpolated extrapolated values for the range W 0 3 5 5 cm In spectral regions where water vapor absorbs the accuracy of the sur face reflectance retrieval depends on the number of water vapor grid points and the interpolation method full range of W or sub interval pertaining to a pixel 66 The CO2 mixing ratio of the atmosphere is set at 400 ppmv the ozone column is fixed at 330 DU Dobson Units equivalent to the former 0 33 atm cm for a ground at sea level The aerosol types rural urban maritime and desert have been provided in the database These files have to be resampled with the sensor specific channel filter curves The file names for the solar region include the altitude the aerosol type and th
263. rent reflectance calculation PARAMETERS calfile Atcor calibration file to be used for conversion of the cube eOsolar File containing the solar irradiance for the sensor atcor file outfile name of output KEYWORDS scale scale for processing same convention as for ATCOR scale 1 0 is floating point out put zen solar zenith angle default 0 degrees in degrees date date as two elemt array day month at_smiledetect incube dbfile respfile resol outfile featureflags vis zen ele chlist results spline spline zeroborder 0 1 2 range overwrite Smile detection routine PARAMETERS incube input data cube dbfile raw database file to be used for convolution no height interpolation respfile response file e g band001 rsp resol internal resolution for the calculation outfile name of output file for smile coefficients CHAPTER 6 BATCH PROCESSING REFERENCE 107 KEYWORDS featureflags bytarr n feat 12 feature regions featureflags i 1 if feature is set else 0 vis visibility km zen solar zenith angle deg zenith at 0 deg ele average ground elevation km chlist list of bands which are used for smile detection and for interpolation of the results numbering starting at 0 results write idl save dump of all results in a file named sav together with the regular output spline 1 spline channel interpolation 0 linear channel interpolation of smile
264. resampling effects on the retrieval of radiance and surface reflectance Applied Optics Vol 39 5001 5005 2000 Richter R and Coll C Bandpass resampling effects for the retrieval of surface emissivity Applied Optics Vol 41 3523 3529 2002 Richter R and Schl pfer D Geo atmospheric processing of airborne imaging spectrometry data Part 2 atmospheric topographic correction Int J Remote Sensing Vol 23 2631 2649 2002 Richter R and M ller A De shadowing of satellite airborne imagery Int J Remote Sensing Vol 26 3137 3148 2005 Richter R Schl pfer D and M ller A An automatic atmospheric correction algorithm for visible NIR imagery Int J Remote Sensing Vol 27 2077 2085 2006 Richter R Bachmann M Dorigo W Mueller A Influence of the adjacency effect on ground reflectance measurements IEEE Geoscience Remote Sensing Letters Vol 3 565 569 2006 Richter R and Schl pfer D Considerations on water vapor and surface reflectance re trievals for a spaceborne imaging spectrometer IEEE Trans Geoscience Remote Sensing Vol 46 1958 1966 2008 Richter R Kellenberger T and Kaufmann H Comparison of topographic correction methods Remote Sensing Vol 1 184 196 2009 Richter R Atmospheric topographic correction for satellite imagery ATCOR 2 3 User Guide DLR IB 565 01 11 Wessling Germany 2011
265. rom the green red and NIR bands which ought to be present The routine simply asks for an input multispectral image to be processed NOTE the input image should have at least 3 bands and the first three bands are assumed to be the triple GREEN RED NIR 5 7 7 Spectral Smile Detection This routine uses sharp atmospheric absorption features and Fraunhofer lines for inflight smile i e spectral across track non uniformity detection The calculation is done by correlation analysis of a number of spectral bands in the vicinity of selected absorption features The outputs may be used for smile aware atmospheric correction Initially the smile characterization for each spectrometer channel is derived from laboratory mea surements From such data the wavelength shift with respect to the center pixel of the detector array can be parametrized using a 4th order polynomial fit However in case of instrument changes during the mission a spectral re calibration might be necessary from the image data or from on board calibration facilities using well defined absorption features Onboard spectral calibration devices such as interference or rare earth filters would be well suited for this purpose However such devices are often not available in sensor systems Therefore atmospheric gas absorption fea tures or solar Fraunhofer lines have to be taken as a reference from the imagery itself Its major steps are 1 A calibrated image is averaged in along
266. rradi ance database see Figure 9 3 It enables an update of the monochromatic atmospheric database without the need to repeat the time comsuming MODTRAN 5 computations involving the cor related k algorithm in some spectral regions The user can also provide additional solar irradiance files to the sun_irradiance folder provided the spectral range increment and irradiance unit agree with the template spectra Attention ATCOR will always work with files in the active folder atm_database therefore the old atm_database has to be renamed or deleted and the folder with the new database has to be renamed as atm_database before applying the sensor specific resampling program RESLUT Since each atm_database folder contains its corresponding solar irradiance spectrum a unique identification is always possible Previously generated channel resampled atm files do not have CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 124 CONVERT ATCOR folder file sun_irradiance e0_solar_fi e0_solar_f2 e0_solar_f3 atm_database e0_solar_f1 atm_database_f2 Figure 9 2 Solar irradiance database a reference to their solar irradiance file but they are based on the e0_solar_kurucz1997_06nm dat irradiance Beginning with the 2011 release the directory of the atm files contains an ASCII file named irradiance_source txt identyfying the underlying solar irradiance file
267. s Outputs An ASCII file which may be used as smile description file in the respective sensor directory Note that this file should be named smile_poly_ord4 dat in order to be automatically recognized by ATCOR As a side output an IDL save dump sav is written in parallel which contains all used parameters and the effectively calculated smile results CHAPTER 5 DESCRIPTION OF MODULES 94 AAA 1 ATCOR Inflight Smile Detection setect Input File Names data hyperion Bern_02 Hyperion_subl67 bs9 Selecr Atmospheric Database File Ysrc_idl atcor atcor_23 atn_database h33000_ww20_rura bp7 Select Sensor Spectral Response sro_idl atcor atcor_23 sensor hyper ion67 bandOOL rsp efine Dutput Smile Coefficients data hnuper ion Bern_02 snile_test dat Smile Detection Resolution nm Search Range nm E Band Range from tos 16 Visibility km 40 0000 Solar Zenith deg 0 0000 Mean Ground Elevation km b 100000 Feature Wavelengths nm J 430 4 520 F 760 F 820 F 940 P 1130 W 1268 11470 42004 I 2055 12317 1 2420 Spectral Interpolation Types Linear w Spline Extrapolation Type to detector limits w extrapolate trend repeat values w to zero at borders Help Run Plot Smile Save Report Done SS 24 Figure 5 56 Spectral smile detection 5 7 8 Spectral Calibration Atm Absorption Features The program SPECTRAL _CAL is only intended for hyperspectral sensors and employs atmo spheric absor
268. s or errors 5 3 2 ATCOR2 multispectral sensors flat terrain The panel as decribed above and in figure 5 17 will appear when ATCOR2 is selected 5 3 3 ATCOR3 multispectral sensors rugged terrain In case of the rugged terrain version of ATCOR the panel for the DEM files has to be specified in addition Figure 5 18 It pops up after the input file has been specified A quick quality check is performed on the DEM files The solar illumination file is calculated and if its standard deviation is large the panel of Figure 5 19 pops up with a warning In this case the DEM elevation file and the derived files of DEM slope aspect etc probably have a lot of large steps The DEM resolution is often not appropriate for high spatial resolution imagery and integer coded DEM s might have to be resampled and stored as float data Appropiate action for resampling or low pass filtering is recommended in these cases see the tips in chapter 9 5 Figure 5 20 shows an example in terms of the DEM illumination The top image is obtained after low pass filtering the original elevation file the central image is the illumination based on the original DEM and the bottom shows a 100 pixel transsect of the original elevation data revealing the steps The original DEM had a resolution of 30 m was coded as 16 bit integer and initially resampled to the 6 m pixel size of the image with integer arithmetic After reprocessing the elevation file the other DEM derived file
269. s SAVI LAI FPAR and albedo coded as 16 bit integer with the following scale factors e SAVI range 0 1000 scale factor 1000 e g scaled SAVI 500 corresponds to SAVI 0 5 e LAI range 0 10 000 scale factor 1000 e g scaled LAI 5000 corresponds to LAI 5 0 e FPAR range 0 1000 scale factor 1000 e g scaled FPAR 500 corresponds to FPAR 0 5 e Albedo range 0 1000 scale factor 10 e g scaled albedo 500 corresponds to albedo 50 The next section presents a simplified treatment of the radiation and heat fluxes in the energy balance 7 2 Surface energy balance Surface energy balance is an essential part of climatology The energy balance equation applicable to most land surfaces can be written as Asrar 1989 R G H 4 LE 7 7 where R is the net radiant energy absorbed by the surface The net energy is dissipated by conduction into the ground G convection to the atmosphere H and available as latent heat of evaporation LE The amount of energy employed in photosynthesis in case of vegetated surfaces is usually small compared to the other terms Therefore it is neglected here The terms on the right hand side of equation 7 7 are called heat fluxes The soil or ground heat flux G typically ranges from 10 to 50 of net radiation Convection to the atmosphere is called sensible heat flux H It may warm or cool the surface depending on whether the air is warmer or cooler than the surface The energy available to evapo
270. s employed for the atmospheric correction We start with the basic equations in the solar and thermal spectral region for clear sky conditions standard case then move on to non standard conditions comprising bidirectional reflectance BRDF effects hazy scenes and a treatment of shadow areas caused by clouds or buildings Stan dard atmospheric conditions include the option of a constant visibility aerosol optical thickness and water vapor content per scene as well as the retrieval of a visibility and water vapor map if the required spectral bands are available for the specific sensor Water vapor correction on a pixel by pixel basis is usually necessary for hyperspectral imagery The section on the non standard conditions contains a short discussion on empirical correction methods for bidirectional effects It continues with the description of a statistical haze removal method The third section presents a technique to compensate shadow effects i e cloud or building shadow areas are masked and de shadowed Then an overview is presented of all major processing steps involved in the atmospheric correction After atmospheric correction the surface reflectance cube can be used for classification A simple automatic method is included here based on template reflectance spectra of different surface covers Finally the accuracy of the atmospheric correction is discussed 10 1 Basics on radiative transfer This chapter presents the basic concepts
271. s for an AVIRIS scene Figure 2 6 shows a comparison of the results of the spectral re calibration for a soil and a vegetation target retrieved from an AVIRIS scene 16 Sept 2000 Los Angeles area The flight altitude was 20 km above sea level asl heading west ground elevation 0 1 km asl the solar zenith and azimuth angles were 41 2 and 135 8 Only part of the spectrum is shown for a better visual comparison CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 23 of the results based on the original spectral calibration thin line and the new calibration thick line The spectral shift values calculated for the 4 individual spectrometers of AVIRIS are 0 1 1 11 0 88 and 0 21 nm respectively 2 4 Inflight radiometric calibration Inflight radiometric calibration experiments are performed to check the validity of the laboratory calibration For spaceborne instruments processes like aging of optical components or outgassing during the initial few weeks or months after launch often necessitate an updated calibration This approach is also employed for airborne sensors because the aircraft environment is different from the laboratory and this may have an impact on the sensor performance The following presenta tion only discusses the radiometric calibration and assumes that the spectral calibration does not change i e the center wavelength and spectral response curve of each channel are valid as obtained in the laboratory or it was already
272. s recommened The exponential option exaggerates the de shadowing results but it can be used to enhance the trends for a quick visual inspection When pressing the button Run Interactive De Shadowing a reduced size image of the original scene the shadow mask and the de shadowed image will pop up with the histogram of PhiU con taining the threshold for core areas point 1 of Fig 5 25 the range of re scaling of PhiU point 2 and the current value for PhiS point 3 Now these three parameters can be modified and the results will again be shown as the corresponding quicklook image and histogram When good results have been obtained the parameters can be saved and the final image processing can take place Figure 5 26 presents an example with two iterations Results for iteration 1 contain too many shadow pixels black areas in the central section of the image therefore the threshold 0 15 was decreased to 0 38 parameter 1 of Fig 5 25 After de shadowing most areas in the shadow mask were overcorrected in iteration 1 therefore the maximum range 0 40 was decreased to 0 19 parameter 2 of Fig 5 25 see diagonal line from lower left to upper right in histogram of Fig 5 26 The shadow mask for iteration 2 is appropriate and no overcorrection effects can be observed Note when leaving the panel of Fig 5 25 it is possible to edit the cloud shadow map before continuing the scene processing using any available image processing
273. s should also be reprocessed CHAPTER 5 DESCRIPTION OF MODULES 62 INPUT IMAGE FILE Yexport data data atcor2 3 demo_data tm_flat tm_essen1000 bsq Date dd mm year 20 08 1989 OUTPUT IMAGE FILE Vexport data data atcor2 3 demo_data tm_flat tm_essen1000_atm bsq Ct OVERWRITE Scale Factor 4 0 Help Band selection Selected SENSOR Landsat 4 5 TM Select MN Z Pixel size m 30 0 CALIBRATION FILE export data data atcor2 3 cal landsat4_5 tm_standard cal ATMOSPHERIC FILE aamsrura ATH FILE for thermal band s midlat_summer Adjacency range km 1 00 Help Zones 1 19 1 Solar zenith degree 43 0 Ground elevation km 0 1 Visibility km SPECTRA AEROSOL TYPE VISIB ESTIMATE INFLIGHT CALIBRATION Help WATER VAPOR IMAGE PROCESSING Output file already exists change name or press OVERWRITE MESSAGES QUIT Figure 5 17 ATCOR panel The pixel size of the DEM files must be the same as the image pixel size specified on the main panel see figure 5 17 The physical units of pixel size m and adjacency range km are also used to calculate the equivalent number of pixels needed to cover the adjacency range 5 3 4 ATCOR2 User defined Sensors The menus for user_defined usually hyperspectral sensors share the same functionalities as de scribed above for multispectral systems in both flat and rugged terrain options The major dif ference is the requireme
274. s temperature T5650 for each pixel in each channel is computed based on the at sensor radiance L converted into brightness temperature In the current implementation only channels in the spectral region 8 13 um are employed for the maximum brightness temperature search because the spectral regions A lt 8 um and gt 13 um are strongly affected by atmospheric water vapor absorption Next a reference channel is defined where most pixels with maximum brightness temperature occur Only those blackbody pixels are retained which have the maximum brightness temperature in this reference channel most hits method For these selected blackbody pixels the scatterplot of measured at sensor radiance L versus blackbody radiance corresponding to Lgg T32150 is computed for each channel This means the surface radiance of eq 10 35 is approximated as Leurface Lap Tinas The final step is a least squares regression of the scatterplot data L versus Lsurface yielding the intercept path radiance Lp and slope transmittance T of eq 10 35 Care has to be taken to apply the regression only to the points near the top edge of all cluster points but allow some margin so the fitting line is allowed to sink by an amount of the sensor noise equivalent spectral radiance NESR The quality of the regression is significantly increased by allowing only those pixels in the scatterplot that had their maximum temperatures in the reference channel Two commen
275. saa osa sa e a e ee Simulation modules Mea Apparent Reflectance Calculation e Thetools menu o Se ddoe doeu a a essa a ER a Calculation of sun angles aa e O a a e a a SPECL spectral classification of reflectance cube Example of classification with SPECI occo acs s e a ae a Re a a 84 LIST OF FIGURES 9 5 04 Nadir normalization se os sa 64 4 a di a a ES 91 5 55 Topographic correction only no atmospheric correction 92 5 56 Spectral smile detection mea near 94 5 57 SPECTRAL CAL spectral calibration 0 eee ee 95 5 58 CAL_REGRESS radiometric calibration with more than one target 96 5 59 Convert monochromanic database to new solar reference function 97 5 60 Convert atmlib to new solar reference function o a e 98 5 61 The help MEM s ccs 204662 ps oe ER ra A a wh a amp 99 Ti Water vapor partial pressure s sor iee eop a a A ee ae 112 Tie o A ee t 113 8 1 Weight factors of hyperspectral bands e 117 8 2 Sensor simulation in the solar region o e e e 118 8 3 Graphical user interface of program HS2MS e e 119 8 4 TOA radiances for three albedos o 120 9 1 Monochromatic atmospheric database ee 122 9 2 Solar irradiance database gt o ss 66242556 5 be bee EEE EAR ERE ED 124 9 3 User interface
276. software Then the edited CHAPTER 5 DESCRIPTION OF MODULES 70 Figure 5 25 Panel to define the parameters for interactive de shadowing map is employed for the processing This provides some flexibility because it is difficult to calculate a satisfactory shadow map in all cases CHAPTER 5 DESCRIPTION OF MODULES 71 nction in od frequercy MIGON tu H scoled shodow function scaled gt y 3 Figure 5 26 Quicklook of de shadowing results Top left histogram of PhiU threshold 0 15 range 0 40 iteration 1 Top right histogram of PhiU threshold 0 38 range 0 19 iteration 2 Center results for iteration 1 left to right original shadow mask de shadowed image Bottom results for iteration 2 CHAPTER 5 DESCRIPTION OF MODULES 72 5 3 11 Panels for Image Processing When pressing the button IMAGE PROCESSING in one of the main panel figure 5 17 some additional panels will pop up First the processing options are to be selected see figure 5 27 Blocked Options Are Not Available For The Selected Sensor Might also apply for a reduced set of bands Either Haze or Cirrus Removal not both Blocked Options Are Not Available For The Selected Sensor OS Variable Visibility aerosol optical thickness Y Yes No Variable Visibility aerosol optical thickness Yes QO No Variable Water Vapor s eooocorercrrrrrrsrroronirarsrns Yes No Y tes OM Haze or Sun G
277. spot angular region where retroreflection occurs see Figure 2 3 left image left part The opposite scan angles with respect to the central nadir region show lower brightness values A simple method called nadir normalization or across track illumination correction calculates the brightness as a function of scan angle and multiplies each pixel with the reciprocal function Figure 2 3 Nadir normalization of an image with hot spot geometry Left reflectance image without BRDF correction Right after empirical BRDF correction The BRDF effect can be especially strong in rugged terrain with slopes facing the sun and others oriented away from the sun In areas with steep slopes the local solar zenith angle may vary from 0 to 90 representing geometries with maximum solar irradiance to zero direct irradiance CHAPTER 2 BASIC CONCEPTS IN THE SOLAR REGION 20 i e shadow The angle P is the angle between the surface normal of a DEM pixel and the solar zenith angle of the scene In mountainous terrain there is no simple method to eliminate BRDF effects The usual assumption of an isotropic Lambertian reflectance behavior often causes an overcorrection of faintly illuminated areas where local solar zenith angles G range from 60 90 These areas appear very bright see Figure 2 4 left part ET Figure 2 4 BRDF correction in rugged terrain imagery Left image without BRDF correction Center after BRDF correction with threshol
278. stretching options Gaussian and histogram equalization are available In the right part two windows are provided to display the c1 spectrum and the box averaged target DN spectrum The ymin ymax widgets allow the user to scale the graphical display The parameters visibility and adjacency range can be varied and their influence on the calibration curve can be studied CHAPTER 5 DESCRIPTION OF MODULES 69 5 3 10 Shadow removal panels The interactive session of the de shadowing method enables the setting of three parameters that influence the results compare Figures 5 24 5 25 1 a threshold r for the unscaled shadow function PhiU to define the core size of the shadow regions see chapter 10 5 5 for details 2 the maximum range maz for re scaling the unscaled shadow function PhiU into the 0 1 interval of the scaled shadow function 3 the last parameter sets the minimum value of the scaled shadow function PhiS typically in the range PhiS 0 02 0 10 i e the darkest shadow pixels of the scene are treated as being illuminated with a fraction PhiS of the direct solar irradiance histogram e 0 5 0 0 0 5 1 0 unsealed shadow function Figure 5 24 Normalized histogram of unscaled shadow function The first two parameters are most important The last parameter is not very critical and a default value in the range PhiS 0 04 0 08 covers most cases of interest The linear type of re scaling of PhiU i
279. surface brightness temperature cube image_atm_emiss_lp5 bsq is the same emissivity map but filtered with a 5 channel low pas filter to smooth spectral noise features requires at least 30 thermal bands image_at_sensor_channel_tmaz bsq map of channel numbers with maximum at sensor tem image_at_surface_channel_tmaz bsq map of channel numbers with maximum surface tem The last channel of image_atm bsq contains the surface temperature map evaluated with the ap propriate emissivity the preceding thermal channels in this file contain the surface radiance In case of the ISAC algorithm an additional file tmage_isac_lpath_trans dat contains the spectral path radiance and transmittance estimates for the scene Fig 4 16 shows an example of these spectra derived from a SEBASS scene Fig 4 17 presents the at sensor at surface radiance and brightness temperatures The at sensor products clearly show the atmospheric absorption features which are removed in the at surface CHAPTER 4 WORKFLOW 45 unscaled ISAC path radiance unsealed ISAC Transmittance transmittance 7 amp 9 10 11 12 13 14 7 8 9 10 11 12 13 14 Wavelength ym Wavelength jam Figure 4 16 Path radiance and transmittace of a SEBASS scene derived from the ISAC method quantities apart from small residual effects The bottom graphic presents the corresponding surface emissivity spectrum CHAPTER 4 WORKFLOW 46 L Wom sr xm 11 0 10 5 10 0 F 2
280. t 1 parametric orthorectification Int J Remote Sensing Vol 23 2609 2630 2002 Schlapfer D PARGE User Guide Version 3 0 ReSe Applications Schlpfer Wil Switzer land 2010 Schowengerdt R A Remote Sensing Models and Methods for Image Processing 3rd Edition Elsevier Academic Press 2007 Sirguey P Simple correction of multiple reflection effects in rugged terrain Int J Remote Sensing Vol 30 1075 1081 2009 Slater P N Remote Sensing Optics and Optical Systems Addison Wesley London 1980 Slater P N Radiometric considerations in remote sensing Proc IEEE Vol 73 997 1011 1985 Slater P N et al Reflectance and radiance based methods for the in flight absolute cali bration of multispectral sensors Remote Sensing of Environment Vol 22 11 37 1987 Soenen S A Peddle D R and Coburn C A SCS C a modified sun canopy sensor topographic correction in forested terrain JEEE Trans Geoscience and Remote Sensing Vol 43 2148 2159 2005 Sutherland R A Broadband and spectral emissivities 2 18 wm of some natural soils and vegetation Journal of Atmospheric and Oceanic Technology Vol 3 199 202 1986 Teillet P M Guindon B and Goodenough D G On the slope aspect corrextion of mul tispectral scanner data Canadian J Remote Sensing Vol 8 84 106 1982 Wiegand C L Gerbermann A H Gallo
281. t emissivity 1 0 or e 0 98 is often used and the corresponding temperature is called brightness temperature The kinetic surface temperature differs from the brightness temperature if the surface emissivity does not match the assumed emissivity With the assumption 1 0 the kinetic temperature is always higher than the brightness temperature As a rule of thumb an emissivity error of 0 01 one per cent yields a surface temperature error of 0 5K For rugged terrain imagery no slope aspect correction is performed for thermal bands only the elevation dependence of the atmospheric parameters is taken into account Chapter 4 Workflow This chapter familiarizes the user with ATCOR 2 3 s workflow and with the program s basic functionality using the graphical user interface A detailed description of all modules and user interface panels is given in the subsequent chapter 5 ATCOR may also be used in batch mode for most of its functions A description of the batch mode can be found in chapter 6 4 1 Menus Overview To start ATCOR double click the file atcor sav It will be opened through IDL or the IDL virtual machine and the graphical user interface of Fig 4 1 will pop up Alternatively type atcor on the IDL command line after having added the atcor directory to the IDL search path A large number of processing modules is available from this level as described in chapter 5 Most of them can be used without reading a detailed manu
282. tance or emissivity and temperature images 000 X Satellite ATCOR File New Sensor Atm Correction Topographic Filter Simulation Tools Help Licensed for Daniel Versi TOA At Sensor Radiance Cube input reflectance At Sensor Apparent Reflectance Resample Image Cube n channels gt m lt n channels Figure 5 48 Simulation modules menu 5 6 1 TOA At Sensor Radiance Cube This routine calculates an At Sensor Radiance Cube from an reflectance image cube All parame ters used for the processing are generated from the x inn file of the input cube If the function is called the cube is opened and the x inn file is read which results in an at sensor cube _toarad bsq Note that this routine does not consider adjacency effects and is a simple forward propagation based on the given parameters and the given standard model No specific panel is displayed The routine asks for the input reflectance image All other infor mation is taken from the inn file Please make sure that the reflectance image spectral definition corresponds exactly to the chosen atmospheric library and sensor definition as of the inn file 5 6 2 At Sensor Apparent Reflectance This routine calculates an at sensor apparent reflectance from a calibrated at sensor radiance image cube This routines alculates for each image band the following output Papp DN x c1 co T d Eo cos 00 5 2 where DN stored data values c gain for convers
283. tance values in the window channels ch 1 3 that are not or only slightly influenced by the water vapor content Therefore the reflectance p2 is calculated as p2 wipi W3P3 10 75 Then equation 10 73 can be written as _ peta u Ege u Ta u Ega u ee p2T2 u 0 Eyal 0 m u 0 Eyal 0 fe CHAPTER 10 THEORETICAL BACKGROUND 172 where Ej2 u is the global flux on the ground for the measurement channel index 2 ATCOR employs 4 to 5 water vapor columns u 0 4 1 0 2 0 2 9 4 0 cm sea level to space geometry to calculate an exponential fit function Rappa u exp a Byu 10 77 which can be solved for the water vapor column u see Fig 10 11 where the diamonds in the figure mark the calculated water vapor grid points u 0 4 1 0 2 0 2 9 cm ee EUA 10 78 APDA Ratio water vapor column ern Figure 10 11 APDA ratio with an exponential fit function for the water vapor Equations 10 73 10 75 to 10 78 are iterated starting with u 1 0 cm calculating Rappa Up dating u L p u p1 p3 and repeating the cycle A minimum of two channels one reference one measurement channel is required The advanced APDA method can take into account multiple absorption channels in the 910 960 nm and 1110 1150 nm regions Two water vapor retrieval algorithms are available in ATCOR compare chapter 9 4 parameter iwv_model 1 2 1 The water vapor maps with the smallest standard deviation in the 940 nm and 1
284. ted surface radiance reaching the sensor the third term is the atmospheric radiance reflected at the surface and attenuated by the surface to sensor path The spectral band index elevation and angular dependence is omitted for brevity The Lgg T term is Planck s blackbody radiance B A T weighted with the spectral channel response function R A F BO TIRAN bsa 10 31 A2 T RAJAA M1 For a discrete temperature interval T T T2 and increment e g Ti 270K T2 330K incre ment 1 K equation 10 31 is solved numerically Temperature and radiance are then approximated by an exponential fit function function with channel dependent coefficients a1 a2 Lgg exp l a a1 a2 10 32 1 T 10 33 a aoln LBB For convenience an offset ag is introduced with default ag 0 The offset term can be used to adjust a temperature bias in a scene Example if scene temperatures are too low by 3K they can be raised by setting ay 3 T 10 34 ao l 9 a alnLBB Remark The computer implementation of the channel resampled radiance equations is coded to minimize CHAPTER 10 THEORETICAL BACKGROUND 157 spectral resampling effects 60 61 Temperature emissivity separation For a sensor with n thermal channels there are n equations of 10 30 with n 1 unknowns namely the n surface emissivities plus a surface temperature So the system of equations 10 30 is always underdetermined
285. ten encountered problems and tips to come around are listed here e Distinction of haze and cloud when can the haze removal algorithm be applied Ground surface information under haze areas can still be recognized in the 600 900 nm region but the brightness contrast is low The haze removal is only applied for channels in the 400 800 nm region However for cloud areas no ground information can be observed If in doubt whether a certain area should be assessed as haze or cloud covered take a look at the scene in the NIR around 850 nm channel if surface features can be seen in this area the haze algorithm might be applied If not the area is cloud covered and a haze removal run will not be successful The cloud mask is not appropiate The cloud mask might contain too few pixels then decrease the cloud reflectance threshold in the atcor preferences preference_parameters dat file The default threshold is 25 re CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 137 flectance in the blue green spectrum The opposite case can also happen if the cloud mask comprises too many pixels the threshold has to be raised e The haze mask is not appropiate This problem may be connected with the setting of the cloud threshold If the cloud threshold is too high cloud pixels are counted as haze and the results of the haze removal are bad because the haze algorithm is applied to clouds e The water mask is not appropiate This might le
286. th angle 8 and a weighting coefficient w The wavelength depending weighting w is based on a typical value of the ratio of the direct to diffuse solar flux on the ground f 1 1 cos05 cosB w if A lt 1 lum 5 3 f cosOs cos if A gt Lipm 5 4 Os is the solar zenith angle of the scene For A gt 1 1m the diffuse flux is neglected i e w 1 The factor f is 1 for a flat terrain and some bounds were employed to prevent overcorrection So for each pixel the new digital number is calculated as DN new DN x f 5 5 The method was compared with the standard Minnaert correction eq 5 6 and was superior in most cases Figure 5 55 shows the GUI panel C089 log cos s cosp A 5 6 cos k f CHAPTER 5 DESCRIPTION OF MODULES 92 INPUT IMAGE FILE data7 atcor2 3 deno data ta_rugged tn_blforest bsq OUTPUT IMAGE FILE Ydata7 atcor2 3 deno_data ta rueged tn_blforest_topo bsq OVERWRITE Band selection Spatial subimage Selected SENSOR Landsat 4 5 TM Solar zenith degree 49 0 Solar azimuth deg fias RUN Topographic Correction with Display of Images RUN Topographic Correction without Display of Images Figure 5 55 Topographic correction only no atmospheric correction 5 7 6 Add a Blue Spectral Channel This routine MAKEBLUE adds a blue spectral channel for multispectral imagery not containing a blue band The band is calculated from empirical relationships f
287. the reference pixels line 22 0 65 0 0 25 ibrdf beta_thr thr_g parameters for BRDF correction in rugged terrain For a flat terrain these parameters are not used ibrdf 0 no empirical BRDF correction or flat terrain ibrdf 1 correction with cosine of local solar zenith angle eq 10 80 with b 1 ibrdf 2 correction with sqrt cos of local solar zenith angle eq 10 80 with b 1 2 ibrdf 11 correction with cosine of local solar zenith angle eq 10 80 with b 1 for soil sand Vegetation eq 10 80 with exponent b 3 4 and b 1 3 for A lt 720 nm and A gt 720 nm respectively i e option a in the BRDF panel see Figure 5 31 weak correction ibrdf 12 correction with cosine of local solar zenith angle eq 10 80 with b 1 for soil sand Vegetation eq 10 80 with exponent b 3 4 and b 1 for A lt 720 nm and A gt 720 nm respectively i e option b in the BRDF panel see Figure 5 31 strong correction ibrdf 21 correction with sqrt cos of local solar zenith angle eq 10 80 with b 1 2 for soil sand Vegetation eq 10 80 with exponent b 3 4 and b 1 3 for A lt 720 nm CHAPTER 9 IMPLEMENTATION REFERENCE AND SENSOR SPECIFICS 134 and A gt 720 nm respectively i e option a in the BRDF panel see Figure 5 31 weak correction This is the recommended standard yielding good results in most cases e ibrdf 22 correction with sqrt cos of local solar zenith angle eq 10 80 with
288. tially varying water vapor column contents 10 4 1 Constant visibility aerosol and atmospheric water vapor This is the easiest case for atmospheric correction Still it can often be applied if homogeneous atmospheric conditions exist These might be encountered for small area scenes i e high spatial resolution imagery If the sensor has no channels in atmospheric water vapor regions results of atmospheric correction are not sensitive with respect to the selected water vapor content and a climatological value e g midlatitude summer US standard or tropical water vapor profile is usually sufficient For hyperspectral instruments the processing has to include the image derived pixel by pixel water vapor map The program performs a check whether the specified visibility leads to negative reflectance pixels for dark surfaces in the red band 660 nm vegetation and NIR band 850 nm water If this is the case the visibility is iteratively increased up to VIS 80 km to reduce the percentage of negative reflectance pixels below 1 of the scene pixels During an interactive ATCOR session the user is notified and can continue with the recommended visibility update or with the initial visibility During batch mode operation the program continues with the updated visibility and the CHAPTER 10 THEORETICAL BACKGROUND 166 initial value is disregarded A corresponding notice is given in the atm log output file The final iterated visibili
289. tion is image_envi_atm bsq and of course image_atm tif The image_envi bsq is deleted after processing but image_envi_atm bsq is kept because it is required for a run of the spectral classification SPECL module The user has to delete this file manually if the SPECL module is not employed The setting tiff2envi 0 is automatically replaced with tiff2envi 1 if the inn file specifies the de hazing or de shadowing option because these require additional disk files In case of tiling and a TIFF input file the program automatically switches to tiff2envi 1 because the inter mediate tile images must have the ENVI bsq format atcor3_batch input filename output file vis vis tiff2envi tiff2envi or atcor3_tile input filename ntx 3 nty 2 output file vis vis tiff2envi tiff2envi The 3 in atcor3_batch means the code for rugged terrain i e a DEM is employed as well as other DEM related files e g slope aspect skyview Otherwise the same explanations hold as for the flat terrain ATCOR The corresponding tile program atcor3_tile in this example is called to split the image into 3 sub images in x direction and 2 in y direction compare chapter 5 3 12 The keywords output and vis are described in atcor2_batch above Note optional keywords for atcor2_batch atcor3_batch atcor2_tile atcor3_tile There are four keywords concerning spectral interpolation to overwrite the interpolation set
290. tions ATCOR will automatically iterate the initial visibility parameter visib set in the inn file if the number of negative reflectance pixels is larger than 1 of the scene for the red band around 650 nm vegetation is checked here or the NIR band around 850 nm water is checked here The specified visibility is always kept if the visibility is set to a negative value i e visib 20 means the program performs the calculation with visib 20 km and does not iterate even if a large number of negative reflectance pixels occurs If the parameter npref is set to 1 the program computes the visibility map based on dark reference pixels in the scene and npref 1 overwrites the initial value of the visib parameter With npref 1 the program still iterates the average visibility of the visibility map by checking for water pixels in the NIR band unless the specified visib is negative A constant scene visibility is employed for npref 0 In case of scene tiling and npref 0 or npref 1 the iterated visibility obtained for sub scene 1 is also applied to all other sub scenes to avoid brightness steps for the merged sub scenes caused by potentially different visibilities Attention If scene tiling has to be performed and the aerosol map is requested for each sub image then specify npref 1 but this could cause different average visibilities in the sub scenes and potentially brightness steps at the sub scene borders 9 5 Problems and Hints Some of
291. to be started before running ATCOR The program accepts float values of the solar zenith and azimuth angles The output file name of the DEM shadow map includes the zenith and azimuth angles rounded to integer values The DEM shadow map is a binary file where shadow pixels are coded with 0 and sunlit pixels with 1 It includes self shadowing and cast shadow effects Self shadowing consists of pixels oriented away from the sun with slopes steeper than the solar elevation angle The cast shadow calculation CHAPTER 5 DESCRIPTION OF MODULES 80 Figure 5 39 Panel of SKYVIEW is based on a ray tracing algorithm and includes shadow regions caused by higher surrounding mountains Figure 5 41 shows the GUI panel 5 4 4 DEM Smoothing Smooth a DEM or any other single band image in order to remove artifacts in the atmospherically corrected imagery All related DEM layers are automatically smoothed as well e g slope aspect skyview Alternatively this task coud be done with any image processing software Inputs Input DEM file Name Usually a DEM _x ele bsq is selected here but any other single band ENVI image or the _ilu bsq file is also accepted The routine searches automatically for related files i e x_sky _slp and or _asp and smoothes them with the same parameters Diameter of DEM Filter Size of filter box in pixels diameter Output Name Name of Elevation file output auxiliary layer names will be derived from that Out
292. track direction leading to a signature image of the size of the detector array 2 The surface reflectance is calculated atmospheric correction and smoothed CHAPTER 5 DESCRIPTION OF MODULES 93 3 The spectral bands within the spectral matching range are selected 4 Spectral shifts with intervals between 0 01 0 05 nm are calculated and applied to the selected spectral band response functions 5 An appropriate pre calculated fine spectral resolution atmospheric LUT is selected which serves for the calculation of at sensor radiance values for the series of spectrally shifted re sponse functions using the surface reflectance spectrum from step 2 6 The derived spectral signatures are correlated to the observed column averaged signal in the image such that the best fitting spectral shift AA A can be found for each image column j i e the A with the highest Pearson s correlation coefficient is selected This is equivalent to minimizing the merit function Ap 5nm A So Er k Ler Aj O 5 7 Aj A 5nm where L j k is the average at sensor radiance of the image for column j and channel k and Lr Ax Aj k is the corresponding reference radiance for a wavelength shift A within a 5 nm interval around Az 7 A 4th order polynomial is fitted through the calculated spectral points and the respective polynomial parameters of eq 4 1 are stored 8 The polynomial parameters are interpolated and optionally extrapol
293. tral and radiometric sensor calibration and on the accuracy and appropriate spatial resolution of a digital elevation model DEM in rugged terrain In addition many surfaces have a bidirectional reflectance behavior i e the reflectance depends on the illumination and viewing geometry The usual assumption of an isotropic or Lambertian reflectance law is appropriate for small field of view FOV lt 30 scan angle lt 15 sensors if viewing does not take place in the solar principal plane However for large FOV sensors and or data recording close to the principal plane the anisotropic reflectance behavior of natural surfaces causes brightness gradients in the image These effects can be removed with an empirical method that normalizes the data to nadir reflectance values In addition for rugged terrain areas illumi nated under low local solar elevation angles these effects also play a role and can be taken care of with an empirical method included in the ATCOR package The ATCOR software was developed to cover about 80 of the typical cases with a reasonable amount of coding It is difficult if not impossible to achieve satisfactory results for all possible cases Special features of ATCOR are the consideration of topographic effects and the capability to process thermal band imagery There are two ATCOR models available one for satellite imagery the other one for airborne imagery 68 69 The satellite version of ATCOR supports a
294. ts first because of the involved assumptions the obtained intercept is not the physical path radiance and the slope not the physical atmospheric transmittance Both quantities may be negative in some channels therefore they are referred to as unscaled path radiance I and unscaled transmittance T They might be rescaled to proper atmospheric path radiance and transmittance spectra e g using a radiative transfer code Second the ISAC method requires an adequate spread in surface temperatures in the scene and surface temperatures higher than the atmospheric radiation temperature So results for night time imagery will likely be degraded The compensated unscaled surface radiance spectrum is calculated as L A Lp A u L A 10 36 arfaa T Y A and the unscaled ISAC surface emissivity can be obtained with ede 00 Ll Ejes 10 37 where Tep is the brightness temperature image in the reference channel The compensated surface radiance spectrum DD can be converted into the equivalent compensated brightness temperature spectrum where most of the atmospheric absorption features are removed Both the compensated surface radiance and compensated brightness temperature are spectrally consistent with the data and represent the best estimate for the spectral shape CHAPTER 10 THEORETICAL BACKGROUND 159 The emissivity spectrum jsac A may exceed the value 1 in certain channels if the maximum brightness temperature of a p
295. ty depends slightly on the initial value because a coarse set of visibility grid points is used to restrict the number of iterations n lt 8 see Table 10 2 Example 1 start VIS 15 km potential visibility iterations 19 24 34 44 54 64 74 km and the next step of 84 km is reset to VIS 80 km If the number of negative reflectance pixels red NIR bands is already less than 1 of the number of scene pixels for VIS 34 km the program will terminate the visibility loop after three iterations Example 2 start VIS 23 km potential visibility iterations 28 38 48 58 68 78 km and the next step of 88 km is reset to VIS 80 km If the criterion threshold of 1 negative reflectance pixels is already fulfilled for VIS 28 km the program will terminate the visibility loop after the first iteration If the criterion is fulfilled for the start visibility no iteration is executed The upper visibility threshold of 80 km is a trade off although higher visibilities are possible they are not very likely and even if a situation with a higher visibility say VIS 120 km is encountered results of a calculation with vis 80 km do not differ much from the results with VIS 120 km So the iteration capability is most important for low visibility start values visibility km vis increment km 5 15 3 15 20 4 20 28 5 28 60 10 60 80 10 but max VIS 80 km Table 10 2 Visibility iterations on negative reflectanc
296. type in Table 4 3 is reserved for arbitrary user specified channel filter functions the band rsp files A calibration file e g chris_m1 cal has to be provided in the new sensor sub directory The RESLUT resample atmospheric LUTs program has to be run to generate the atmo spheric LUTs for the new sensor employing the monochromatic atmospheric database in atcor atm_database These resampled atm files will automatically be placed in a sub directory of atcor atm_lib with the name of the selected sensor RESLUT will also create the resampled spectrum of the extraterrestrial solar irradiance in the appropriate CHAPTER 4 WORKFLOW 39 atcor bin cal sensor chris_m1 chris_m3 atm_database atm_database_chris atm_lib standard multispectral sensors chris_m1 chris_m3 spec_lib Figure 4 13 Directory structure of ATCOR with hyperspectral add on sensor chris_m1 folder see chapter 9 1 4 e g e0_solar_chris_m1 spc Remember that the sensors might have to be specified as scene dependent if the center wavelengths and or bandwidths change so you might need sensor subdirectories such as chris _m1_11april2003 or chris_m1_scene3_29june2004 Wavelength File I Columns 1 3 are band number center wavelength bandwidth micron or nm vi Butterworth order 2 close to Gauss v2 v3 va 3 Gauss vs v6 wis v
297. ulated as pi p3 2 6 2 3 Finally water pixels are masked with the criterion p4 lt 7 where the large threshold of 7 is employed to account for potential turbid water bodies and the blue band reflectance is calculated as p 1 2 x po 6 3 specl2_batch input filename sensor xx or specl2_tile input filename sensor rxz ntr ntx nty nty The spectral classification based on template reflectance spectra is also available in the batch mode and with the tiling option The xx is a keyword for the sensor type e g xx landsat7 for Landsat 7 ETM The complete list of sensor keywords is shown when typing specl2_batch on the IDL command line without the sensor specification The ntx nty keywords have the meaning explained for the ATCOR tile programs above smile_interp3_batch input filename fpoly fpname option number silent silent Purpose The atmospheric correction accounts for the column dependent smile shift as spec ified in the smile_poly ord4 dat of the corresponding sensor folder but the image columns of each band belong to slightly different wavelengths This function interpolates the pixel reflectance values for each band to a specified reference wavelength Three options exist for the reference wavelength grid 1 use wavelength corresponding to the center of the detector array 2 use average wavelength over all detector columns per band 3 use nominal wavelength specified in the
298. um exist the following relationships are used DN blue lt Tsaturation and p blue gt 0 22 NDSI gt 0 6 or p green gt 0 22 NDSI gt 0 25 p SWIR2 p green lt 0 5 10 47 CHAPTER 10 THEORETICAL BACKGROUND 162 Again if the blue or green band is saturated and NDSI gt 0 7 then the snow class is assigned Cirrus over land The apparent reflectance is calculated in the cirrus band 1 38 wm and additionally in a band around 1 24um If the latter does not exist a NIR channel is used Thin cirrus over land is calculated with 1 0 lt p cirrus lt 1 5 10 48 employing the percent reflectance unit Medium thickness cirrus is calculated as 1 5 lt p cirrus lt 2 5 10 49 and the thick cirrus class consists of pixels with p cirrus gt 2 5 10 50 and the pixels have to belong to the land class Cirrus over water The same spectral criteria eq s 10 48 10 49 10 50 hold but the pixels have to belong to the water class Haze over land see chapter 10 5 2 The the mean of the tasseled cap transformation TC is calculated Clear pixels are those with TC lt mean TC and p blue lt T cloud over land threshold and p NIR gt Twater NIR wa ter reflectance threshold defined in preference parameters dat Next the mean and standard deviation o of the HOT transformation are calculated Pixels are assigned to the compact haze mask if HOT gt mean HOT and to the large haz
299. und leaving radiance per band can be obtained in addition to the spectral solar fluxes by setting the parameter irrad0 2 It is calculated corresponding to the surface reflectance cube p x y named scene_surfrad bsq For a flat terrain it is L surf x y Elglobal p x y T 10 28 In case of a mountainous terrrain the direct and diffuse reflected radiation maps from the equations 10 26 and 10 27 are used L sur f dir x y Ear Eaif plx y T 10 29 Again the same output file name is used scene_surfrad bsq 10 1 4 Thermal spectral region Similar to the solar region there are three radiation components thermal path radiance L1 i e photons emitted by the atmospheric layers emitted surface radiance L2 and reflected radiance L3 The short form of the radiance equation in the thermal region can be written as 32 24 L Lp 7 Lpp T 4 7 1 e F a 10 30 where L at sensor radiance L Lp thermal path radiance T ground to sensor atmospheric transmittance E surface emissivity T surface temperature LB blackbody radiance at temperature T weighted with the channel s filter curve F thermal downwelling flux on the ground CHAPTER 10 THEORETICAL BACKGROUND 156 L c 0 DN L Ly A A Figure 10 5 Radiation components in the thermal region Li Lp La 7 Lgg T L3 T 1 Fr The second term on the right hand side of equation 10 30 is emit
300. varying visibility option was selected and if the sensor has a 2 2 um or 1 6 um band or at least a red and NIR band required for the automatic masking of the dark reference pixels compare chapter 10 4 2 Figure 5 30 Reflectance ratio panel for dark reference pixels CHAPTER 5 DESCRIPTION OF MODULES 74 The panel of figure 5 31 pops up for rugged terrain and contains the input parameters to BRDF correction as discussed in chapter 2 2 Figure 5 31 BRDF panel Figures 5 32 and 5 33 are associated with the value added products as described in chapter 7 This value added file contains up to 10 channels if the sensor has thermal bands In case of a flat terrain the air temperature has to be specified For a rugged terrain air temperature at some base elevation and the air temperature gradient as well as water vapor have to be defined If the value added option is selected another panel pops up Figure 5 34 It contains parameters for the leaf area index LAI model and FPAR model as described in chapter 7 Finally a job status window indicates the processing progress Note The job status window of ATCOR shows the percentage of processed image data and the estimated remaining time The time estimate is based on the processing time for the current band The time per band increases for channels in atmospheric water vapor regions it decreases in regions where interpolation is applied e g around 1400 nm However the time also de
301. vegetation sand asphalt based on the following criteria vegetation Pnir Prea gt 2 and Pnir gt 0 20 soil dry vegetation Pnir Pred gt 1 4 and Pnir Pred lt 2 0 and Pred gt 0 09 l sand asphalt Pnir Pred lt 1 4 and Prea gt 0 09 water Pnir lt 0 05 and P1 6um lt 0 03 To each class the user can assign an emissivity valid for the channel with the highest temper ature There is only one common emissivity class in case of night data or data from purely thermal channels The ANEM method provides accurate channel emissivities and surface temperatures if the classes are assigned correctly and the emissivity value assigned to the channel with the max imum temperature is close to the actual channel emissivity Maximum surface emissivities usually lie in the 10 5 13 um region After calculating the surface temperature the emis sivities for all channels are computed for thermal band imagery with at least 5 channels the ISAC In Scene Atmospheric Com pensation method is available A detailed description is given by Young et al 89 The CHAPTER 10 THEORETICAL BACKGROUND 158 method does not require ancillary meteorological data or atmospheric modeling It neglects the downwelling thermal flux and employs the equation L L TE Lpg T Lp T Lesur face 10 35 This approximation is justified for pixels with a high emissivity close to 1 i e blackbody pixels First the highest brightnes
302. ws the response function R of a ms channel and the dashed lines indicate the hs center wavelengths A eq 8 2 and they are documented in the corresponding log file created by program HS2MS i 1 k wy i EL 82 Y RaM j 1 Fig 8 2 describes the sequence of processing for the sensor simulation in the solar region After atmospheric correction with ATCOR the image_atm bsq contains the surface reflectance cube Program TOARAD then calculates the at sensor radiance for a different solar geometry or at mospheric parameters All parameters not specified as keywords see list of keywords below are taken from the mage inn file created by ATCOR The program HS2MS can be started to resample the radiance cube to a ms image by specifying the ms sensor i e channel filter functions and the ms noise equivalent radiance NER NER 0 is allowed so the resampled image product will only include the noise of the hs scene which will be reduced due to the inherent integration over several hs bands A channel constant NER or a file with channel dependent NER values may also be employed Figure 8 3 shows the GUI panel of program HS2MS Although the input image will usually be a hyperspectral scene with n gt 50 channels and the output a multispectral scene with m lt lt n channels this program can also be employed for the case of a multispectral input image with n lt 10 channels and a panchromati

Atmospheric / Topographic Correction for Satellite Imagery

Contents

Download Pdf Manuals

Related Search

Related Contents