Home

MEASUREMENTS OF PLANT STRESS IN

1. ayn y eat ima Figure A6 multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi the LabVIEW block diagram for the program used to acquire images and calculate reflectance and NDVI 153 Figure A6 Continued 154 Figure A6 Continued 155 Figure A6 Continued To run the program Figure A6 a folder must be created to save the images and a reflectance spreadsheet The reflectance spreadsheet must also be created with a layout of 19 cells wide and about 75 cells long The length depends on how long imaging will take place and the image capture frequency The cells should be filled initially with zeros The first column is used to hold the time of each reflectance measurement The second third fourth and fifth columns hold the Green Red NIR reflectances and NDVI respectively for the first region The seventh eighth ninth and tenth columns hold the Green Red NIR reflectances and NDVI respectively for the second region The twelfth thirteenth fourteenth and fifteenth columns hold the Green Red NIR 156 reflectances and NDVI respectively for the third region The seventeenth eighteenth and nineteenth columns hold the ITs for the Green Red and NIR color planes respectively for that specific image The user must also supply some inputs cam
2. de Jesus Jr Waldir Cintra et al Comparison of Two Methods for Estimating Leaf Area Index on Common Bean Agronomy Journal Vol 93 2001 989 991 SA guide to Reflectance Coatings and Materials North Sutton NH Labsphere Inc 2006 lt http www labsphere com data userFiles A 20Guide 20to 20kReflectance 2 OCoatings 20and 20Materials pdf gt 133 did Ot Dobeck Personal Communication 2007 ZERT field site manager Montana State University PE Dobeck Personal Communication 2008 ZERT field site manager Montana State University R Lawrence Personal Communication 2007 LRES Dept Montana State University J Shaw Personal Communication 2007 ECE Dept Montana State University 85 Shaw Personal Communication 2008 ECE Dept Montana State University J Lewicki Personal Communication 2007 Lawrence Berkeley National Laboratory 4J Lewicki Personal Communication 2008 Lawrence Berkeley National Laboratory Jensen John R REMOTE SENSING OF THE ENVIRONMENT AN EARTH RESOURCE PERSPECTIVE Upper Saddle River NJ Prentice Hall Inc 2000 Rossotti Hazel Colour Princeton NJ Princeton University Press 1983 BRocchio Laura Landsat 1 Landsat s History 6 Oct 2008 NASA 6 Oct 2008 lt http earthobservatory nasa gov Library GlobalWarmingUpdate printall php gt Paine David P and James D Kiser Aerial Photography and Image Interpretation Hoboken NJ John Wiley and Son
3. 91 spectralon scene Figure 60 Spectralon and vegetation scene viewed at the same angle to remove the effect of a variable illumination angle on the calibration panel This effect was proven at the end of the experiment using two spectralon panels to ensure the supposed erroneous reflectance measurements were not real To test this affect a 50 panel was laid flat and the 99 panel was set at 45 The 99 panel was used as the calibration target and the 50 was used as a test target It was found that the reflectance of the 50 panel changed significantly throughout the day very similarly to the vegetation The temporal change in reflectance had a cosine like behavior which is due to the fact that the projected area of each panel as seen by the sun was different and as the sun moved thru the sky each panel s projected area changed For analysis in this paper I have selected data points that match with solar noon This ensures that each point from each day has the same basic illumination conditions Plots of these NDVI data can be seen in Chapter 5 Statistical Analysis In addition to analyzing single spectral bands and NDVI I have statistically analyzed spectral band and NDVI combinations to find the best possible combination to model vegetation stress To do this I used the statistical computing program R This program allowed me to set date time as the response variable spectral bands and or NDVI as predictor variables and r
4. Eq 8 and Eq 10 can be combined to find the irradiance incident on the vertical card Enor a coslO eco a 1 vert A COS nin E This result can be used to substitute for Esun in Eq 7 85 x ES Eora cos erario Pa O pixel cos 0 DN 7 zenith int egration G t 12 This equation was used to find the gain factor for each channel of the MS 3100 imager To do this first Enor was modeled by a MODTRAN fit to band averaged pyranometer data for each channel A fit of MODTRAN data was needed since it does not match with pyranometer data throughout the period of a day Figure 54 Irradiance Wirn MODTRAN Data MODTRAN Data Fit x Pyranometer Data 1 1 1 1 1 6 00 AM 8 00 AM 10 00 AM 12 00 PM 2 00 PM 4 00 PM 6 00 PM Time Figure 54 Irradiance modeled by MODTRAN and measured by a pyranometer for one day The MODTRAN model used the same spectral band that is used by the pyranometer to detect broad band short wave solar radiation which is 300 3000 nm To obtain the irradiance fit Efi the irradiance from the pyranometer Epyranometer 18 subtracted from the solar irradiance modeled by MODTRAN Emoptran 300 3000nm fOr different times through the day and then a linear fit is applied to the irradiance difference and a polynomial fit is applied to the MODTRAN irradiance as indicated in Figures 55 and 56 E fit Emotran 300 3000nm gt anita 1 3 86
5. xviii ABSTRACT In response to the increasing atmospheric concentration of greenhouse gasses such as CO produced by burning fossil fuels which is very likely linked to climate change the Zero Emissions Research Technology ZERT program has been researching the viability of underground sequestration of CO2 This group s research ranges from modeling underground sequestration wells to detection of leaks at test sites One of these test sites is located just west of Montana State University in Bozeman MT at 45 66 N 111 08 W At this site experiments were conducted to assess the viability of using multispectral imaging to detect plant stress as a surrogate for detecting a CO leak A Geospatial Systems MS3100 multispectral imager implemented in color infrared mode was used to image the plants in three spectral bands Radiometric calibration of the output of the imager a digital number DN to a reflectance was achieved using a grey card and spectralon reflectance panels To analyze plant stress we used time series comparisons of the bands and the Normalized Difference Vegetation Index NDVI computed from the red and near infrared band reflectances Results were compared with rainfall soil moisture and CO flux data The experiment was repeated two years in a row the first from June 21 2007 to August 1 2007 and the second from June 16 2008 to August 22 2008 Data from the first experiment showed that plants directly over the leak
6. 200 data1 linear Iradiance Difference Wire 1 1 20 25 30 36 40 45 50 55 Zenith Angle degrees Figure 55 Linear fit for the MODTRAN Pyranometer irradiance difference Ert co E o O data2 cubic co co Q N o y co o Irradiance Wim l o o 1 1 20 25 30 35 40 45 50 55 Zenith Angle degrees Figure 56 MODTRAN modeled irradiance polynomial fit EmopTRAN 300 3000nm MODTRAN was also used to model the irradiance for each spectral band of the MS 3100 Emoprrana green 520 560 nm red 650 690 nm NIR 767 5 832 5 nm 87 Finally to find the actual irradiance Enor for each spectral band the irradiance fit is subtracted from the band averaged modeled irradiance Eq 14 Eora Emobrran a 7 Ep 14 Now Eq 14 is plugged into Eq 2 and the gain factor was found for each channel at the specific IT used when the grey card was imaged by the spectrometer and the MS 3100 These gain factors were then plotted along with a gain factor of zero for a zero IT to find the gain factor as a function of IT shown in Table 12 Table 12 Gain factor for specific ITs and gain factors as a function of ITs for each channel of the MS 3100 G 20 75ms G tintegration 9 00E 05 1E 4 tintegration 4 00E 05 eo 4E 5 tintegration ade 5 00E 05 6E 5 tintegration With the gain factors determined the reflectance of the vegetation scene can n
7. Dev Module i Temperature acquistion National NEDAQ Many amp processing functions Instruments 2007 image acquisition 1 multiple extract planes vi control program Written 2 calculate reflection 2007 image post NI LabVIEW scaffolding_grey_multiple_scenes 2 vi processing program nen 3 multiple extract planes amp calculate pe reflectance without temp a A Written sensor 2_with correction _test2 vi p g 4 calculate reflection 2008 image post Wri o ritten scaffolding_grey_multiple_scenes 3 processing program Spectrometer data Ocean SpectraSuite N A acquistion and Optics processing program p MODTRAN Irradiance modeling PCModwin NA program Silane oD Computes spatial non MATLAB 1 falloff_correction uniformity correction Written 2 create_refl_plot Plots reflectance NDVI Written Plots reflectance NDVI 3 create_refl_plot_min at min point each day Written R Linear Model Statistical Computation B awrence 35 DTControl File Tools About Image 2 Green Sensor Control Gain dB 0 E 0 0 Section Integration msec 3 10 00 3 10 00 2 10 00 i A 0 00 32 50 000 32 50 0 00 32 50 Overall Exp 10 Ba Min hax Output F abr Feature Control Section Display Modes _Mdeo trigger F RGB g White Bal Exp cre y Auto F hbno Red F wono Blue F ono Green g F Other Edit Other faie Aoguire ge f2 Aran a File Counter Image Display Mode oan Acquisition Data Multiplex wets ed Co
8. Green Red NIR minus and plus Each VI sets a reference to its specific color plane or arithmetic function and passes it to the buffer These VIs return a New Image ID for each of the color planes and functions which can be used in subsequent VIs to retrieve specific image data Panel 4 grabs data from the buffer and displays the images on the front panel IMAQ Get Buffer is called to grab the buffer buffer 0 containing the image data It returns an Image Out which is a multiplexed Color infrared image This is then displayed on the front panel using a synchronous display and is named CIR Next the three color planes must be extracted This VI takes the New Image IDs NIR Red and Green from panel three as inputs into the Red Green Blue Planes respectively This gives the standard Color infrared mode used by the remote sensing community This VI also takes the Image Out from JMAQ Get Buffer as an input to Image Source so that there is an image to extract the color planes from The three color plane images are then returned and sent to synchronous displays on the front panel labeled IR Red and 142 Green At this point all three color planes and the Color infrared images have been displayed on the front panel Next an JMAQ Subtract VI takes the NIR and Red color planes as Image Source A and Image Source B respectively It also takes the minus Image Name from panel three as the
9. Irradiance modeled by MODTRAN and measured by a pyranometer for one TEO 85 Polynomial fit for the MODTRAN Pyranometer irradiance difference Exit 86 MODTRAN modeled irradiance polynomial fit EMoDTRAN 300 3000nm omm o 86 Erroneous reflectance data due to differences in sun spectralon angle and sun scene lei coda 89 Differences in sun spectralon angle and sun scene angle oooooccnnncccinncccinnnccnnnss 90 Accurate reflectance data taken with spectralon laid flat oooonnnicnnnnnicinnno in 90 Spectralon and vegetation scene viewed at the same angle to remove the effect of a variable illumination angle on the calibration panel ee 91 2007 mown segment green red and NIR reflectances for regions 1 solid 2 CEN E 99 2007 mown segment Date versus NDVI for regions 1 green 2 red and 3 2007 un mown segment green red and NIR reflectances for regions 1 solid 2 dash and 3 00t ccoconnooonnnncnccnonnnoonananonocnononnnnononononcononnanononononccnon 101 2007 un mown segment Date versus NDVI for regions 1 green 2 red ANAS Dl luto aa dai 101 2008 mown segment green red and NIR reflectances for regions 1 solid 2 dash and S dO id ac 104 2008 mown segment Date versus NDVI for regions 1 green 2 red and 3 Figure 67 68 69 70 71 72 73 74 75 76 T11 78 xvi LIST OF FIGURES CONTINUED Page 2008 un mown segment green red
10. Then user starts the EGTA Set up image references for the three color planes MATLAB script node opens CCD spatial non uniformity correction arrays Set up and display each region s reflectance and NDVIs on four different real time graphs Close current image acquisition Loop back to acquire another image after the allotted image capture frequency Configure buffer image type and initialize asynchronous image acquisition Enter timed loop with the image capture frequency set by the user to control how often to image Initialize image capture session and configure image buffer list Enter IT control loop If all three are color plane ITs are correctly set continue with image acquisition Initialize image capture session and configure image buffer list Close current image acquisition Calculate average DN for spectralon using ROI MATLAB corrects image data for spatial non uniformity Grab image data from buffer Check if current IT gives average DN for spectralon that falls within the range set by user for each color plane Configure buffer image type and initialize asynchronous image acquisition If one or more ITs are not correct continue with IT control loop Set up image references for the three color planes Figure 27 Flow diagram for multiple extract planes amp calculate reflec
11. and NIR reflectances for regions 1 solid 2 dash and 3 dot ensina s aa a no nnnnnccnon 106 2008 un mown segment Date versus NDVI for regions 1 green 2 red ANAS Dl unta ia enaa a NA E A ENN 106 Green red and NIR reflectances for individual plants within un mown A ony catusasuaeunaeavatersg paadusunondeasasgaveuavenaan taneneaoes 108 Date versus NDVI for individual plants within 2008 un mown segment 109 2007 CO flux map of the ZERT CO2 Detection site adapted from J Lewicki Lawrence Berkeley National Laboratory 2007 ececeesseeeeeteeees 110 2008 CO flux map of the ZERT CO2 Detection site adapted from J Lewicki Lawrence Berkeley National Laboratory 2008 oonocccnnncccnnncccionncc nn 114 Position and number of soil moisture probes adapted from L Dobeck Chem Dept MSU ZO A a 115 Soil moisture for mown strip probes adapted from L Dobeck Chem Dept MSU 2008 cinsini a ance ae seeasabekea gasses E a NENS 115 Soil moisture for un mown strip probes adapted from L Dobeck Chem D pt MSU 2008 ii A az 116 Precipitation data adapted from J Lewicki Lawrence Berkeley National Eaboratory 2008 estes aei a aa r a o eaeh Erea i dina 116 Image of mown and un mown segments taken 9 July 2008 a and 9 August 2008 to visually illustrate the change in the health of the vegetation J Shaw ZO oa 118 Close up images of mown segment taken 9 July 2008 a and 9 August 2008 b to visually illustrate the
12. having been mown Again since it has not been mown there is a veil of tall nearly dead grass obscuring the stress signature from underlying vegetation The change in vegetation seen by inspecting Figures 77a and 79a beginning of the experiment and Figures 77b and 79b end of the experiment shows that the veil initially healthy and green dies by the end of the experiment Even though this veil exists the system was able to detect even more vegetation stress than in the mown segment This can be seen in the somewhat higher R values in the un mown segment 0 9033 compared to the mown 0 7273 The NDVI was able to statistically separate regions 2 and 3 and regions 1 and 3 as indicated in Table 19 These results together with the NDVI trends suggest that once the CO concentration has reached a certain level the effects on plants will be the same even if the CO concentration increases further This agrees with Arp 1991 in that 122 increased CO concentration and a sink source imbalance can lead to lowered photosynthetic capacity This shows that for an un mown segment the difference in the levels of CO flux in these regions cause a stress on the vegetation that is detectable up to a level somewhat higher than background This is evident because the NDVI in all regions are initially at the same point but by the end of experiment regions 1 and 2 are at nearly the same level and region 3 is much higher than the other two reg
13. nm Transmittance Figure 15 Transmittance of MS3100 channels in Color IR mode Green represents green red represents red and dark red represents NIR www geospatialsystems com 25 The path length from the lens to each of the CCDs is the same and is equivalent to the back focal length of the lens system Each of the CCDs is 7 6 mm x 6 2 mm with a pixel size of 4 65 um x 4 65 um which gives 1392 x 1040 pixels The red and NIR CCDs are monochrome CCDs while the green CCD is tri color and uses a demultiplexed Bayer Pattern to achieve the green signal The tri color CCD captures RGB images and outputs either RGB demultiplexed red demultiplexed green or demultiplexed blue images depending on an input provided by the user Even though I did not design and build the MS 3100 imager I have modeled it using the non sequential mode in the Zemax optical design code Figure 16 The prism was not modeled and the dichroic surfaces were modeled as ideal 1 e they reflect 0 and transmit 100 or reflect 100 and transmit 0 of light Zemax models dichroic surfaces by placing them on a perfectly transmitting glass plate The transmission and reflectance characteristics of each dichroic surface must then be set for wavelength and angle For example it is possible to set specific transmission and reflectance percentages for specific wavelengths at specific angles of incidence The CCDs are modeled with appropriate spectral trim filters Fig
14. should be un mown for remote plant stress detection when this kind of tall grass is prevalent though one is still able to see the effects of increased CO concentration on vegetation in the mown segment The NDVI was also able to explain variability in plant health very well This again shows the ability of the NDVI to explain the variability in 124 plant health in separate vegetation regions The NDVI was again able to detect the effects of hail and rain in the same fashion as it did in the un mown segment Summary In this study it has been found that the NDVI may have advantages over any other combination of spectral bands available on the imager used with or without NDVI for statistically detecting differing levels of plant stress and explaining the variability in plant health The NDVI is able to distinguish between both mown and un mown vegetation regions that have been stressed compared to non stressed regions but the NDVI is much stronger when the vegetation region has not been mown Also the NDVI explains variability in plant health better in un mown regions but not by much Spectral bands may not be the best solution to detect differences in the health of vegetation in different regions since they are best used to explain the variability in the spectral response of vegetation over time Another interesting result of this analysis is that differing levels of CO2 concentrations can lead to nourishment or stress depending on sink
15. 0 040 2100 0 015 0 013 2100 0 029 0 040 2400 0 022 0 015 2400 0 040 0 042 According to this chart we may be seeing a 5 change of the nominally 50 reflectance due to illumination angle which corresponds to a reflectance increase of 2 5 That is a nominally 50 reflectance panel will see a 5 change in its total reflectance which would make it appear as a 52 5 reflectance panel when viewed at large angles 66 By correcting for the spatial non uniformity and illuminating the spectralon at smaller angles we were able to obtain much better reflectance values within about 4 of the nominal values Given the low uncertainty in reflectance for the white 99 panel it was used as a known reference from which adjusted reflectance values were found for the 50 gray panel The resulting gray panel reflectances are shown in Table 9 for each channel of the MS 3100 imager These were found by requiring the calibration to yield the band averaged reflectances measured by Labsphere for our specific panel as shown in Table 10 Table 9 Reflectance values needed for the 50 reflectance panel to obtain 99 4 reflectance for the 99 panel Reflectance Reflectance panel Green 520 560 nm Red 650 690 nm NIR 768 833 nm 50 0 5075 0 5325 0 5525 Table 10 The percent reflectance for both spectralon panels as specified by the Labsphere data sheet These values are spectrally integrated across the
16. APPENDIX A In depth Discussion of LabVIEW Programs ocoocnncccnnncnnoncnonncnnncnnnnos 135 Table 10 11 12 13 14 15 ix LIST OF TABLES Page USB4000 Miniature Fiber Optic Spectrometer Optical Layout Explanation USB4000 Installation and Operation Manual oooonnocccnnoccnonancnonncnnnnnccnonnncnnnnnos 33 Listing of software programs used routines used within each program purpose for the program and routine and source of the software ee 34 Values used to calculate reflectance for each of the color planes during the 2007 experiment with the photographic grey Card ooooocnocccccoccccnoncnononanononanonnnanos 43 Serial settings needed to communicate with the MS3100 imager 47 Message format to query or set the MS3100 integration time eee 48 Values obtained in a test of the CCDs ability to quickly drain charge after viewing bright CONCUIONS 3 2 28 eds ence ee Ae Oa i 61 Bleeding effect caused by CCD charge walk off DN on a scale of 0 255 62 Change in reflectance of spectralon due to illumination angle measured from the surface normal for 99 and 60 panels Labsphere Inc 2006 65 Reflectance values needed for the 50 reflectance panel to obtain 99 4 reflectance for the 99 panel adi aie ceassniestedessenpcaocoeteawaneeuaane 66 The percent reflectance for both spectralon panels as specified by the Labsphere data sheet These values are spectrally integra
17. CO detection site Previous applications of multispectral imaging systems have shown the ability to remotely sense plant health Here I have shown the viability for this type of system to run continuously in the field in any sky conditions for CO detection where there is vegetation This was proven by linking increased CO concentration with plant stress or nourishment depending on sink source balance and by comparing regions of stressed vegetation to regions of non stressed vegetation to show that they are statistically separable using NDVI The NDVI has also been shown to be capable of explaining variability in differences in plant stress between vegetation regions I have shown that NDVI is better suited in this case for detecting differences in the health of vegetation within multiple regions than any combination of the bands represented by the MS3100 imager and NDVI This may be because the spectral bands explain the variability in the spectral response of vegetation over time very well but they exhaust their capacity to explain the variability between vegetation regions NDVI on the other hand has increased capacity to explain the variability between regions since it is not as well suited to explain the variability in the spectral response of vegetation These results can be seen on average in higher R and p values derived from linear regressions of Date versus NDVI The system also measured plant stress with enough accuracy to see the
18. Equal VI This VI returns a yes or no which is the control for the following conditional structure The program then uses the conditional structure which depending on if the DN is greater than or less than the upper limit it will start the process to lessen the IT or check the lower limit respectively If less than the spectralon average DN is compared to the lower limit using a Less Or Equal VI Then another conditional structure is used which depending on if the DN is greater than or less than the lower limit it will not change the current IT and exit this flat sequence structure with a true indicating the IT is set correctly or it will start the process to increase the IT respectively Since the process to increase or decrease the IT is the same I will explain just the increase 161 Table A2 Serial settings needed to communicate with the MS 3100 imager Input Setting Enable Termination Character Termination Character Timeout ms Baud Rate Data Bits Parity Stop Bits Flow Control Initially the imager is queried and current attributes are read When the imager is queried and the imager buffer is read 6 Bytes are written and 9 bytes are read If the imager is sent a command to change it s attributes 8 Bytes are written and 6 Bytes are read The messages are in hexadecimal and if querying or setting the IT the messages have the format shown in Table A3 The check sum is the
19. Geospatial System Inc 9 May 2007 lt http www geospatialsystems com gt U S Geological Survey 2008 Department of the Interior USGS 21 Feb 2008 lt http landsat usgs gov gt Ocean Optics 2007 Ocean Optics Inc 9 Oct 2007 lt http www oceanoptics com gt DTControl Software User Manual Rev C Auburn CA DuncanTech 1999 Installation and Operation Manual USB4000 Fiber Optic Spectrometer Dunedin FL Ocean Optics Inc 2001 Camera Link 2000 Silicon Imaging Inc 8 Sept 2008 lt http www siliconimaging com gt Remer Lorraine A Global Warming Earth Observatory March 2007 NASA 16 Sept 2008 lt http earthobservatory nasa gov Library GlobalWarmingUpdate printall php gt Media Advisory IPCC adopts major assessment of climate change science 2007 Intergovernmental Panel on Climate Change March 29 2007 10 Allam Rodney et al Capture of CO Carbon Dioxide Capture and Storage Ed Ziad Abu Ghararah Tatsuaki Yashima New York NY Cambridge University Press 2005 105 178 1 Anderson Jason et al Underground geological storage Carbon Dioxide Capture and Storage Ed Giinther Borm David Hawkins Arthur Lee New York NY Cambridge University Press 2005 195 276 Lewicki Jennifer L Jens Birkholzer and Chin Fu Tsang Natural and Industrial Analogues for Release of CO2 from Storage Reservoirs Identification of Features Events and Processes
20. Image Destination input The function returns Image Source A Image Source B which is the numerator for NDVI Next an JMAQ Add VI takes the NIR and Red color planes as Image Source A and Image Source B respectively It also takes the plus Image Name from panel three as the Image Destination input The function returns Image Source A Image Source B which is the denominator for NDVI Panel 5 finds the present time converts and saves image color planes minus and plus data into arrays of pixel values The arrays are saved with a time stamp so that each image s time will be saved To do this a Format Date Time String VI returns the present time in string format This is then concatenated with the color plane name or NDVI so that each image for each color plane has a distinguishable file name Five IMAQ Image to Array VIs are used to convert the color plane minus and plus images to arrays of pixel values The minus array is then divided by the plus array to find a NDVI array The color plane and NDVI arrays are then saved to separate files using the Write to Speadsheet File VI This VI needs the file path the Image Path set by the user concatenated with the file name found above and the array of pixel values as inputs Each of the images were saved so that post processing may be done The NDVI arrays were not used again since a better calculation was done using the calculate reflection
21. Type Grayscale U8 These VIs also need names to reference the images to be made I gave these the Image Names Green Red and NIR Each VI captures its specific color plane and passes it to the buffer These VIs now return a New Image ID for each of the color planes which can be used in subsequent VIs to retrieve specific color plane image data 158 Panel 1 4 grabs data from the buffer separates the three color planes and converts the images to arrays of pixel values IMAQ Get Buffer is called to grab the buffer buffer 0 containing the image data It returns an Image Out which is a multiplexed Color infrared image This is then displayed on the front panel using a synchronous display and is named CIR Next the three color planes must be extracted The Image Out is then passed to an IMAO ExtractColorPlanes VI This VI takes the New Image IDs NIR Red and Green from panel three as inputs into the Red Green Blue Planes respectively This gives the standard Color infrared mode used by the remote sensing community This VI also takes Image Out as an input to Image Source so that there is an image to extract the color planes from The three color plane images are then returned Three IMAQ Image to Array VIs are used to convert the color plane images to arrays of pixel values Panel 1 5 is for spatial non uniformity correction A MATLAB script node was used here It takes three inputs and assig
22. a cosine function versus Spectralon illumination angle The system has been shown to have a linear response in the expected range of operating values Though the super wide angle lens imparts a spatial non uniformity the correction works well enough to make an evenly illuminated scene look flat across all pixels each of the image arrays The spectralon panel acts as expected for a lambertian surface and the measurements made to verify this underscore the importance of considering separately the illumination angle and viewing angle Considering that the characterization of the imaging system shows that it has a linear response across the expected range of values that the non uniformity correction works well and that the Spectralon calibration panels are nearly perfectly Lambertian makes this system very well suited for accurate measurements of vegetation reflectance 73 EXPERIMENTAL SITE SETUP AND METHODS Zert CO Detection Site Setup The ZERT CO detection site Figure 8 was set up by a group of researchers to investigate multiple methods for detecting a CO leak To simulate a carbon sequestration site leak a horizontal 100 m pipe was placed about 2 5 m below the ground The detection techniques include but were not limited to a differential absorption laser system that measure CO concentrations CO flux chambers and hyperspectral and multispectral systems that measured the vegetation response to increased CO2 The ZERT site was
23. and Lessons Learned Lawrence Berkeley National Laboratory Berkeley CA Feb 2006 Carter Gregory A Responses of Leaf Spectral Reflectance to Plant Stress American Journal of Botany Vol 80 Mar 1993 239 243 Horlerd N H M Dockraya and J Barber The red edge of plant leaf reflectance International Journal of Remote Sensing Vol 4 1983 273 288 131 Rock B N T Hoshizaki and J R Miller Comparison of in situ and airborne spectral measurements of the blue shift associated with forest decline Remote Sensing of Environment Vol 24 1988 109 127 Curran P J J L Dugan and H L Gholz 1990 Exploring the relationship between reflectance red edge and chlorophyll content in slash pine Tree Physiology Vol 7 1990 33 48 Cibula W G and G A Carter 1992 Identification of a far red reflectance response to ectomycorrhizae in slash pine International Journal of Remote Sensing Vol 13 1992 925 932 Carter Gregory A and Alan K Leaf Optical Properties in Higher Plants Linking Spectral Characeristics to Stress and Chlorophyll Concentration American Journal of Botany Vol 88 2001 677 684 Sinclair T R M M Schreiber and R M Hoffer Diffuse Reflectance Hypothesis for the Pathway of Solar Radiation Through Leaves Agronomy Journal Vol 65 1973 276 283 Gamon John A et al Relationships between NDVI Canopy Structure and
24. are the image array dimensions and the last dimension is the image number 63 Next two for loops are set up one the size of the x dimension and one the size of the y dimension of the image array so that each pixel is tested and will have a linear correction function Inside the loops the linear fit starts by placing the pixel values for all images for the 1 index of both dimensions of the image array pixel 1 1 into an array This is a 1x5 array of uncorrected pixel values Then the MATLAB polyfit function was used to make a 1 order fit of the uncorrected pixel values to the expected values The function returned the gain and offset slope and intercept of the best fit line which were placed in separate arrays at indices corresponding to the pixel indices The loop then continued until all pixels had a linear fit Finally the gain and offset arrays were saved to a file to be accessed when future images are captured The program was run for all three color planes The correction is applied during imaging by reading the uncorrected pixel values into a MATLAB script that uses Equation 4 shown with the NIR color plane as an example to correct the pixel values Uncorrected Figure 36 and corrected Figure 37 plots of the uniformly illuminated images for the Red color plane are shown below 64 Pixel Vlaue 1400 1000 1200 800 600 Y axis Pixel Index X axis Pixel Index Figure 36 3 D view of uncorrected red colo
25. by the copyright holder Joshua Hatley Rouse November 2008 Vv DEDICATION Dedicated to my dog Hector vi ACKNOWLEDGEMENT Many thanks to Dr Shaw and Paul Nugent for all the help with every aspect of my research These two always gave me a new path when the last was dead end Thanks again to Paul for all the little random help that he kind of had to give me since he got stuck with me in the office Also thanks to Tyler Larsen for the help in the hot sun Thanks to Ben Staal for the mechanical engineering help Thanks to Kevin Repasky and Rick Lawrence for helping me to get a specific plan going for this thesis relating 1t to what I would like to do for a job and for filling in technical details I would like to acknowledge ZERT for allowing me to do this work thru their funding Thanks to Eli Shawl for his construction knowledge vil TABLE OF CONTENTS INTRODUCTION isis daa 1 MULTISPECTRAL VEGETATION IMAGING ooococncccooncnnconncnnnnnnnnnnconnonnnonannnnnannononos 20 Spectral Response Ol Platini iii 20 Imagine Hardware uti a rd iii 23 Spectrometer HardWare a tenons dd ino 31 MANS aaa 33 2007 Experian 37 ZOOS Experi odio isa 43 IMAGING SYSTEM CHARACTERIZATION AND CALIBRATION cccoccccconcconncnnno 52 EXPERIMENTAL SITE SETUP AND METHODS cooococncoccconccncnnnnnonnconccnnannncnnncnncnnnonos 73 ZERT CO Detection Site Selina iii daysgaaasvstadsvendecevaneadensadesdace 73 2007 Experimental Setup and Imaging M
26. change in the health of the vegetation J Shaw 2006 aa a a a tanttagstantshag a e a a casa iniabtes 118 xvil LIST OF FIGURES CONTINUED Figure Page 79 Images taken 3 July 2008 a and 9 August 2008 b to visually illustrate the change in the health of the vegetation J Shaw 2008 Plant 10 s location is indicated by the blue circle Plants 8 and 9 locations are indicated by the A N 119 A1 Flow diagram for multiple extract planes vi the image acquisition program used In O diia 137 A2 multiple extract planes vi the LabVIEW block diagram for the program WSCC to ac quie images 112 OO Jadoo 0 etecis iio 138 A3 Flow diagram for calculate reflection scaffolding_grey_multiple_scenes 2 vi the reflectance and NDVI calculation program used in 2007 csc A ias 145 A4 calculate reflection scaffolding_grey_multiple_scenes 2 v1 the Lab VIEW block diagram for the program used calculate reflectance and NDVI in A5 Flow diagram for multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi the image acquisition reflectance and NDVI calculation program for 2008 oooooococcnocccooocccooncnononcconnncnonnncnnnnncnnnnos 152 A6 Flow diagram for multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi the image acquisition reflectance and NDVI calculation program used in 2008 sseseeeseseeesserererreesersrrsresseessese 153
27. effects of rain and hail in accordance with rain data taken by J Lewicki 2008 The imaging 127 system has high enough accuracy for these measurements due to strong calibration techniques using a spectralon calibration target that is imaged in every image of vegetation With this level of accuracy for the application at hand the need for a high data volume hyperspectral system is reduced In addition the system will run autonomously throughout the day there is no need for someone to run the system other than at setup and take down For a high degree of accuracy in plant stress detection strong calibration techniques must be used It is best for the calibration target to be imaged in every vegetation image so that every image has a reflectance standard for the calculation of reflectance This way it is possible to make reflectance measurements regardless of the sky conditions Spectralon is the best lambertian surface for this purpose in that it is spectrally flat across the spectral bands of interest in plant heath studies It was found for fully calibrated data throughout the day as the Sun changes angles the calibration target should be laid flat This also leads to the fact that the vegetation should be mown since then the calibration target and vegetation are nominally at the same angles with respect to the sun and the imager It was found that to optimize detection of increased CO concentrations it is best for the vegetation bei
28. filled with many people and experiments during the CO release Figure 44 The CO detection system I employed was one that measured the response of vegetation to increased levels of CO EC tower 07 Underground lasers eas SW end of pipe zone 6 zone 5 zone 4 zone 3 zone 2 zone 1 Figure 44 ZERT CO detection site layout L Dobeck Chem Dept Montana State University 2008 J Lewiki Lawrence Berkeley National Laboratory 2008 Kadie Gullickson Montana State University 2008 74 2007 Experimental Setup and Imaging Methods In 2007 the ZERT site was set up so that the center of a 100 m test vegetation strip ran perpendicular to the 100 m CO release pipe By doing this we created both a test and control vegetation area That is we hypothesized that the vegetation near the pipe would be most affected by the CO2 while the vegetation far away would feel less of an effect or no effect Unmowed Veg f Test Strip N Mowed Veg Test Strip Figure 45 Imager orientation at the ZERT Site in 2007 The imager was mounted on a 10 ft 3 m scaffolding about 3 m south of the intersection between the pipe and the vegetation and just to the west of the vegetation Figure 45 The imager was set to view at a 45 elevation angle so that the imager would view vegetation from 0 5 m north of the pipe to 13 5 m north of the pipe In azimuth the imager was pointed north north west and was recessed in a protective
29. for loop where each spreadsheet iteration is a reflectance NDVI calculation for one image Calculate average DN for each vegetation region and the calibration target using regions of interest User supplies path to folder of images and relfectance NDVI Read file name out of the file name array at the associated iteration and save this to reflectance NDVI spreadsheet as a reference for the calculated reflectance and NDVI Calculate and save reflectance spreadsheet The program can be started Complete all loop iterations computing all reflectances Open reflectance NDVI spreadsheet calculate and Reads file names from image folder and stores in 1 D array Query file name for color plane so that the correct algorithm can be applied Counts files for number of loop iterations needed save NDVI Figure 26 Flow diagram for calculate reflection scaffolding_grey_multiple_scenes 2 vi the reflectance and NDVI calculation program used in 2007 The program Figure A4 of the Appendix starts by reading the file names at the given folder path and storing them in a 1 D array so that they can be called one at a time for calculations An example file name is 125432 PMNIR where the first six digits are the time of the image and the first two letters after the space indicate morning or afternoon 12 54 32 PM The rest of the file name is the color plane the image re
30. for one day using grey card to calibrate all images After inspecting the grey card images I came to the realization that some of the images had nearly the same brightness for the grey card and had some vegetation that was unobstructed by the grey card The images I selected all had grey card DNs between 50 and 61 which correspond to 16 and 24 of the full working range of the imager I found 15 of these images that could be used to calculate reflectance This worked well and are apparently able to capture the effects of the CO on the vegetation These results are discussed in the next chapter 83 2007 Procedure Using Modeled Irradiance I thought it might be possible to scale a modeled solar irradiance for each day by each CCD s response to calculate reflectances for each color plane The first step is to find the gain factor for each imaging array in the camera To do so I simultaneously measured a grey card with the USB4000 spectrometer and with the MS 3100 The spectrometer data were used to find the reflectance of the grey card for each spectral band of the MS 3100 Using this reflectance I was able to determine the irradiance on the camera according to Bios 0 O _ pixel E irah a m gt 5 where Ecamera 18 the irradiance seen by the camera for a specific spectral band Esun is the irradiance of the sun modeled by MODTRAN for a specific spectral band p is the reflectance of the grey card for a specific spectral b
31. or setting the IT the messages have the following format Table 5 Message format to query or set the MS3100 integration time Query Set Attributes Message Byte Write Read Write Read Start Byte Start Byte Start Byte Start Byte LSB Size LSB Size LSB Size LSB Size MSB MSB MSB MSB Size Size Size Size Command Command Command Command Byte Byte Byte Byte Channel Channel Channel Number Number Number Status Checksum IT LSB IT LSB Checksum IT MSB IT MSB Status Checksum Checksum The check sum is the 8 bit two s complement of the message bytes The message bytes in a query are the command byte and channel number and the message bytes in a set attribute command are the command byte channel number IT LSB and IT MSB For a more detailed description of serial communication with the MS3100 consult the MS2100 MS2150 amp MS3100 Digital Multispectral Camera User Manual Next the 5 and 6 bytes are read and converted to decimal format The IT is then changed according to an algorithm that checks how far from the acceptable range the current IT is This algorithm also checks that the new IT is within minimum 0 and maximum 1024 values for the IT which correspond to 0 12 ms and 130 75 ms respectively There is a set of five conditional structures used to determine the difference between the spectralon DN 49 lower limit set by the user and the actual DN This gives six possible alterations to the IT Th
32. problem in 2008 since the system was implemented with automated exposure control and calibration panel auto referencing for every image Images of background conditions were acquired before and after the CO injections ended each year The imaging system I designed was mounted on a small tower with a 45 viewing angle so that it viewed the ground directly above the buried release pipe out to about 10 meters away from the pipe The system makes calibrated reflectance measurements of three spectral bands in the near infrared red and green using a Spectralon calibration target as a Lambertian reference These three bands were chosen since it is known that healthy plants are highly reflective in the near infrared while unhealthy plants are not In addition healthy plants usually have higher reflectance in the 18 green than the red while unhealthy plants have a much flatter response across these two bands The reflectance data were processed to create NDVI values as a function of time Eq 1 with nearly continuous operation during the daylight hours throughout the full experiment Both reflectances and NDVI were analyzed statistically to determine their effectiveness for plant stress detection In addition to the band reflectances NDVI also was used since it relies on the difference in the NIR and red reflectances that relate physically to plant health it is simple to calculate and use and it is historically one of the most commonly used in
33. publications dealing with the effects of added carbon dioxide to plant health but there might be similar affects from other gases Though carbon is a fertilizer in some cases and different gases will affect the spectrum of 15 vegetation differently it has been shown that there is an effect on the spectral response due to changes in soil gas content For example Noomen 2006 did a study on the effects of natural gas methane and ethane on the reflectance of maize Noomen 2006 converted reflectances to band depths for analysis It was found that natural gas and methane caused small decreases in band depth while ethane caused a marked decrease in the band depth of the 550 750 nm absorption region Noomen 2006 There also seemed to be a shift in the blue and red absorption features for the ethane treatment towards longer wavelengths Noomen 2006 She also found that there was a decrease in the reflectance at a water absorption band perhaps by ethane causing a decrease in water uptake This experiment may be analogous to a CO leak in that soil gas content would affect vegetation health leading one to believe that plant health could be a good indicator of a CO leak The recent trend for VIS NIR imaging imaging comprising the spectrum from blue through the NIR has been towards hyperspectral systems because of the spectral detail gained from having many narrow spectral bands The ability of these systems to map fine details such as separat
34. reasons First I found that the grey card is not as lambertian as hoped This problem arose from the position at which I held the grey card during imaging I would hold the grey card in front of the imager trying to do this at the same angle every time This did not work as well as I thought because the non Lambertian nature of the card led to a large variation of calibration target reflectance with grey card angle Also the sun angle changed throughout the day causing nearly specular reflection near midday This problem is evident when inspecting images some images of the grey card are almost black 0 reflectance while others are nearly white 100 reflectance This causes large jumps in the calculated reflectances almost every time the IT is changed Second the calibration of images taken after the initial grey card image was not stable because of the change in solar irradiance that occurred during the period when the IT was not changed This problem is evident by noticing the brightness of the vegetation 82 ramping down until midday and ramping up after midday This causes an erroneous overall shape to a full day s data Instead of seeing only the reflectance of the vegetation in the reflectance plots a secondary signature of the solar irradiance upon the grey card was added as indicated in Figure 52 60 50 40 Reflectance QU o 20 9 00 AM 11 00 00 AM 1 00 PM 3 00 00 PM 5 00 PM Time Figure 52 Reflectance
35. scaffolding_grey_multiple_scenes 2 vi program 143 Panel 6 acquires two temperatures one near the imager and one at the back of the computer converts them to Celsius and saves them to a temperature array A DAQmx VI was called to read the two temperature samples from the analog output of the NI usb 6210 digital analog I O module The samples were converted to Celsius using the following equation T ome 2 982 A sample Tesis og 25 2 In this equation Tsample is a 2 D representation of the temperatures sampled by the NI usb 6210 and Teeisius is a 2 D representation of the temperatures in Celsius The temperatures are wired to a 2 D Waveform Chart vi to display both temperatures in real time The temperatures are also wired to a Write to Spreadsheet File VI in case temperature corrections needed to be made At this point the flat sequence structure has finished so the present image acquisition can be closed To do this first IMAQ Extract Buffer is called with a 1 as the input to Buffer to Extract which clears the buffer Next IMAO Close is called to stop the current asynchronous acquisition closes all information pertinent to this acquisition and closes the IMAQ session At this point the system will loop back to panel one of the flat sequence structure after the allotted Image Capture Frequency has passed When the system is shut down for the day after all the images have been acquired the program will exit the imaging
36. than the edges The integrating sphere was used because it is the best source we have to create uniform illumination other than the sun but it is still not perfect for this application 71 Normalized Spectrometer Measurements as a Function of Viewing Angle Normalized Power Spectrometer Cosine Function 60 40 20 0 20 40 60 Theta degrees Figure 42 Normalized power measured by a spectrometer and cosine of the viewing angle Finally I kept the spectrometer fixed while rotating the spectralon around its vertical axis Figure 39 The spectrometer was placed very close to the panel The measurements were integrated over 590 670 nm and then normalized to 1 I measured the power from 70 to 70 in 10 increments The normalized power falls off very close to a cosine function Figure 43 This happens because the projected area of the spectralon illuminated by the integrating sphere decreases as a cosine In this case the projected area of the spectralon seen by the spectrometer stays constant since its FOV is filled by spectralon but the detected power decreases because of the decreasing irradiance on the panel 72 Normalized Spectrometer Measurements as a Function of Spectralon Angle Normalized Power o o o o o oD E am om mu oO a Ly a ho Spectrometer 0 1 Cosine Function 60 40 20 0 20 40 60 Theta degrees Figure 43 Normalized power measured by a spectrometer plotted along with
37. the imager to maintain the brightness of the spectralon calibration panel within a specific range set by the user This makes it possible to get good measurements in sunny or cloudy conditions since the spectralon is viewed in every image The vegetation and calibration regions are chosen with user defined ROIs The program will then calculate reflectance and NDVI for three regions and display the results in real time on a graph Finally the program saves the results to a spreadsheet 151 multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi Create a folder to place image MATLAB corrects image data pixel value arrays and a for spatial non uniformity spreadsheet for reflectances Calculate average DN for vegetation scenes and spectralon using ROI Finds present time for file name identification converts and saves each color plane as an array of pixel values User supplies camera interface name image capture frequency paths to image and reflectance files and limits z for apectraloniDNiasa Ii Grab image data from buffer control Then user starts the ARORA Set up image references for the three color planes MATLAB script node opens CCD spatial non uniformity correction arrays Set up and display each region s reflectance and NDVIs on four different real time graphs Close current image acquisition Loop back to acquire another im
38. throughout all regions the vegetation has been nourished For this segment it can be seen in Figure 65 that the NIR reflectance time series plots for regions 2 and 3 have positive slopes while region 1 has a negative slope The red reflectance for region 2 has a more negative slope than regions 1 and 3 The green reflectances are nearly the same though region 1 has a small negative slope while the other two have a small positive slope The red reflectances which began higher than the green end below the green reflectances A unique result of the 2008 103 analysis is that the NDVI values generally increased over time in all three regions of the mown segment As is discussed further in the Discussion section of this chapter this appears to be a result of significant rainfall and cooler temperatures during 2008 compared with 2007 The date versus NDVI regressions shown in Figure 66 agree with the red reflectance in that region 2 NDVI increases much more slowly over time compared to regions 1 and 3 The NDVI regressions for regions 1 and 3 have nearly the same slope but very different beginning and end points The regression coefficient of determination 0 7273 Table 17 was able to explain the variability reasonably well The regression was significant with a p value of lt 2 2e 16 Table 17 As expected the p values for both the intercept and slope regression coefficients Table 18 show that regions 1 and 2 and regions 2 and 3 were statist
39. titles used on the plots are changed to distinguish between each segment These programs will also plot a first order fit to the minimum points to help the viewer get a better idea of the total trends In addition 95 confidence intervals were plotted Finally R and p value values are output from these programs for analysis of data more specifically goodness of fit and statistical significance respectively 32 IMAGING SYSTEM CHARACTERIZATION AND CALIBRATION To ensure good calibration of images acquired in the field I tested the imaging system for potential causes of error Much of this characterization work came about as part of an investigation of calibration accuracy using two spectralon panels I wanted to test the system by using one spectralon panel to calibrate images while using the other as a test target since both are of known reflectance Spectralon is nearly perfect for this since it is a nearly lambertian surface The two spectralon targets used here are rated at 50 and 99 reflectance Equation 3 was used to calculate reflectance On a clear day I set up the imager so that it viewed both panels simultaneously Imager attributes were set to similar values used in the ZERT field Then using the LabVIEW VI multiple extract planes amp calculate reflectance without temp sensor vi which is the same as multiple extract planes amp calculate reflectance with temp sensor vi except without temperature measurements I placed th
40. would be added to the end for example if the file name mentioned above were a grey card image it would be 125432 PMNIRgrey It does this using the List Folder VI which takes the folder path as the input and returns the names of all the files An Array Size VI was used to find how many files there are within the folder This number will be used as the number of loop iterations needed to do the calculations for every file Next the for loop is entered where each iteration calculates reflectance and NDVI for a specific file Within the loop the first file name is read and is appended to the folder path so that it can be opened in the future The file name is read using an Array Subset VI that takes the array of file names and the number of the current iteration of the loop So for the first iteration this VI chooses the first file name in the array The file name is then displayed on the front panel so the user knows what the current file being used is This file name is then saved to the reflectance spreadsheet set up by the user This is done using Read rom Spreadsheet File vi Replace Subset vi and Write o Spreadsheet File vi The read uses the user defined reflectance path to open the file The file name is then placed into the spreadsheet where the row to save to in the spreadsheet is controlled by the loop iteration number and the column is the first Then the spreadsheet is written to the reflectance path replacing the old file 148 N
41. 02 23 2014 71 15905 46 1813 24 1354 91 1611 77 1204 37 1410 38 1053 82 1208 83 903 28 1097 36 752 73 895 69 6B2 1B 604 41 461 64 42 94 301 09 201 47 150 55 0 00 Figure 17 Power incident on modeled 3 chip imager detectors These pictures show detectors These pictures show the central beam and the edge of field beam indicating that the optical system simultaneously produces proper images on each of the three CCDs The top is the green color plane the bottom left is the red color plane and the bottom right is the NIR color plane One of the key requirements of this system was to image an area of approximately 10 m length or larger from the top of a scaffold whose height was best kept at or less than approximately 6 m Because of the very small size of the CCDs used in the MS 3100 imager it is quite difficult to achieve anything other than a very narrow field of view FOV Initially I was using the short focal length 14 mm Sigma lens sold with the MS 3100 imager which is usually a wide angle lens but with the small CCDs our full angle horizontal FOV was only 24 It turns out to achieve a shorter focal length with a lens that mates properly to the MS 3100 imager required that I use a combination of an old 20 mm f 3 5 Nikon lens and a 0 25x Phoenix Super Fish Eye lens adapter see Figure 13 to decrease the focal length to effectively 5 mm and increase the full angle horizontal FOV to 55 The 20 mm l
42. 07 0 0 eeeeeeseceseceeeeeeseeeeseeenaeenes 3 Basic block diagram of carbon dioxide capturing systems Allam et al 108 5 Location of CO2 sequestration sites Anderson et al 198 eesse 6 Basic block diagram of carbon dioxide capturing systems Anderson et al A A A N 7 Vegetation kill at Mammoth Mountain CA http pubs usgs gov fs fs172 DOS LIZ DO Pdf snsc eeni eini das e ici 8 Arial View of ZERT Site Dobeck Chem Dept MSU 2007 ooocccnnocccinccccconcns 9 Spectrum of a healthy gold unhealthy blue and dead grey plants Spectrum acquired with a USB4000 spectrometer made by Ocean Optics MM A A O a a O Sheet 12 Vegetation Test Strip for 2007 Shaw ECE Dept MSU 2007 eee 17 Vegetation Test Strip for 2008 Shaw ECE Dept MSU 2008 eee 17 Spectral Absorption and Reflection Characteristics of Plants http landsat US 8S POV sac hes as anes Sen laude aude Steve ance niu ce AA eae 20 Imaging system including the MS 3100 three CCD Imager made by Geospatial Systems Inc and the small computer to run the system 24 Figure 14 15 16 17 18 19 20 21 22 23 24 xii LIST OF FIGURES CONTINUED Page Schematic optical layout of the MS 3100 with color infrared setup Laa e EE RT 24 Transmittance of MS3100 channels in Color IR mode Green represents green red represents red and dark red represents NIR www geospatialsystems
43. 6 07411 07116 07 21 07 26 07 31 08 05 08 10 Date Figure 76 Precipitation data adapted from J Lewicki Lawrence Berkeley National Laboratory 2008 D 1 la 06 21 06 26 07401 The rain and soil moisture data support the hypothesis that the observed vegetation stress is likely caused by the CO that is the stress is not caused primarily by 117 a lack of soil moisture The soil moisture probes were not yet calibrated absolutely so the soil moisture in one region cannot be compared to that in another region However valuable information can still be retrieved from these data The soil moisture plots indicate that from the beginning 9 July 2008 to the end 7 August 2008 of the experiment the net change of soil moisture was a slight increase at each probe position Although the day to day trend is generally downward several large rain storms caused sudden increases of soil moisture sufficient to offset the longer term drying In fact after the two hail storms the soil moisture increased by at least 50 at each probe located within the un mown segment and almost doubled at each probe located within the mown segment The smaller relative changes in soil moisture for the un mown segment may indicate that the dense vegetation had to compete for water while the less dense vegetation in the mown segment did not thereby allowing the mown segment to become healthier The small net increase in soil moisture at each probe location suggests that an
44. 7 With this in mind there is growing interest among many organizations to do something to mitigate emission of greenhouse gases in the attempt to reduce or stop contributing to the warming of the Earth At this point in time most of our energy production comes from fossil fuels which create carbon dioxide when burned Therefore it is doubtful that humans can simply stop using fossil fuels as a source of energy in the near future However one option being explored at this time is the capture and geological sequestration of carbon dioxide The capture of carbon dioxide is being explored mostly for large scale fossil fuel power plants fuel processing plants and other industrial plants Allam et al 2005 Small scale capture at this point would be too difficult and expensive for example applied to individual cars To mitigate these small sources an energy carrier such as hydrogen or electricity could be produced at fossil fuel plants with capture technologies Allam et al 2005 The capture process basically emits non greenhouse gases such as O and N2 and compresses and dehydrates the carbon dioxide for easy shipment and sequestration Figure 4 The four basic types of capture systems are as follows Allam et al 2005 e Capture from industrial process streams e Post combustion capture e Oxy fuel combustion capture e Pre combustion capture Oxyfuel Industrial processes Raw material Gas Ammonia Steel Fig
45. 8 bit two s complement of the message bytes The message bytes in a query are the command byte and channel number and the message bytes in a set attribute command are which are the command byte channel number IT LSB and MSB For a more detailed desription of serial communication with the MS3100 consult the MS2100 MS2150 amp MS3100 Digital Multispectral Camera User Manual So for a NIR query a VISA Write VI is called to query the image using 0202 0015 02E9 as the Write Buffer string A wait is called followed by a Bytes At Port Property Node VI The output of this VI is wired to the Byte Count input to a VISA Read VI The imager attribute buffer is read and returned 162 Table A3 Message format to query or set the MS3100 integration time Set Attributes Message Byte Write Read 0 Byte Start Byte Start Byte Start Byte Start Byte 1 Byte LSB Size LSB Size LSB Size LSB Size MSB MSB MSB MSB 2 Byte Size Size Size Size Command Command Command Command 3 Byte Byte Byte Byte Byte Channel Channel Channel 4 Byte Number Number Number Status 5 Byte Checksum IT LSB IT LSB Checksum 6 Byte IT MSB IT MSB 7 Byte Status Checksum 8th Byte Checksum Next the 5 and 6 bytes are read and converted to decimal format The bytes are read using two String Subet VIs which requires the string read from the imager Offset of the wanted bytes and the Length number of bytes as inputs These strings are sent to Stri
46. Bleeding effect caused by CCD charge walk off DN on a scale of 0 255 Green DN Red DN NIR DN DN in read out direction 0 64 1 1 1 4 DN not in read out direction 0 1 0 1 0 25 Finally I investigated the spatial non uniformity of the CCDs To do this I viewed a spectralon panel with the FOV filled and I definitely saw a fall off of image brightness toward the edges When comparing the center pixels to the edge pixels there was almost a 50 fall off To correct this we decided to apply a unique linear signal dependent calibration function to each pixel of each CCD where the variable is the DN of the pixel To find the gain and offset of these functions I constructed a linear fit to five images one dark image with the lens cap on an image of each spectralon panel 50 and 99 reflective filling the FOV in direct sunlight and an image of each spectralon panel filling the FOV in shaded conditions A MATLAB script was created that could quickly handle the large image files used to process these calibration data The script starts by opening an image file Then it finds an average value the expected value for all pixels for a 20x20 array of pixels right in the center of the pixel array for each image It then places these values into a 1x5 array that contains the expected value for all pixels in each image Then the uncorrected image arrays are placed into a 1040x1391x5 array where the first two dimensions
47. CO the Earth has seen in about 650 000 years Remer 2007 The effects of this are seen in rising Earth temperatures glacial melt and rising sea surface levels In the past one hundred years the Earth s temperature has risen about 0 75 C Figure 2 and the rate of this increase has nearly doubled since the 1950s Remer 2007 It is believed that the Earth has lost 8 000 km of glaciers since 1960 Figure 3 Remer 2007 Sea levels around the world have risen by about 0 17 m during the Twentieth Century Remer 2007 Carbon Dioxide Concentration pre industrial carbon dioxide concentration 280 ppm Temperature Anomaly temperature baseline 1951 1980 24 Year Figure 2 Plots of the increase in carbon dioxide concentration and temperature NASA graphs by Robert Simmon based on carbon dioxide data from Dr Pieter Tans NOAA ESRL and temperature data from NASA Goddard Institute for Space Studies Remer 2007 Change in Volume km Year Figure 3 Plot of the decrease in volume of all Earth s glaciers Glacier graph adapted from Dyurgerov and Meier 2005 Remer 2007 Although there is still lingering debate concerning the relative importance of natural cycles and human caused climate change the science underlying the greenhouse portion of the climate is well understood and most scientists now agree that taking some prudent measures to reduce the growth of CO emissions into the atmosphere is warranted IPCC 200
48. Green Blue Processed Red Processed Green Processed Blue Processed Mono and channel off The Processed options are the Bayer de multiplexed data from the tri color CCD in the camera Next the camera definition file was copied from the DTControl CD to the NI IMAQ Data folder Then using MAX the camera definition was activated so that a link between the camera and NI software could be set up To do this Devices and Interfaces in the MAX Configuration Tree was selected This will show all of the NI devices installed on the computer Then by clicking NJ IMAQ Devices all the NI IMAQ devices will be shown With only one device connected this shows img0 NI PCI 1428 the interface assigned for the specific frame grabber being used Then by clicking on this the channel assigned for the imager Channel 0 is shown Finally the camera definition file must be selected To do this right click the Channel 0 select open camera definition file and then select the file copied earlier At this point the camera is ready to be used with LabVIEW but LabVIEW must be readied Once LabVIEW is installed NI Vision Acquisition and NI Vision Development Module must be installed for full image acquisition and processing capability NI Vision Acquisition software is an application that will add a few virtual instruments VIs the name of programs written in the LabVIEW graphical language used to acquire display and save images NI Vision D
49. IR Red and Green from Path 4 as inputs into the Red Green Blue Planes respectively This gives the standard Color infrared mode used by the remote sensing community This VI 165 also takes the Image Out from IMAQ Get Buffer as an input to Image Source so that there is an image to extract the color planes from The three color plane images are then returned The Panel 6 finds the present time converts and saves image color plane data into arrays of pixel values The arrays are saved with a time stamp so that each image s time will be saved To do this a Format Date Time String VI returns the present time in string format This is then concatenated with the color plane name so that each image for each color plane has a distinguishable file name Three IMAQ Image to Array VIs are used to convert the color plane images to arrays of pixel values The color plane arrays are then saved to separate files using the Write to Speadsheet File VI This VI needs the file path the Image Path set by the user concatenated with the file name found above and the array of pixel values as inputs Each of the images were saved so that post processing may be done Panel 7 is for spatial non uniformity correction A MATLAB script node was used here It takes three inputs and assigns them variable names The original array gets the name color plane name _scene for example the uncorrected NIR array would be called NIR_scene The gain corr
50. MEASUREMENTS OF PLANT STRESS IN RESPONSE TO CO USING A THREE CCD IMAGER by Joshua Hatley Rouse A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Electrical Engineering MONTANA STATE UNIVERSITY Bozeman Montana November 2008 11 OCOPYRIGHT by Joshua Hatley Rouse 2008 All Rights Reserved ili APPROVAL of a thesis submitted by Joshua Hatley Rouse This thesis has been read by each member of the thesis committee and has been found to be satisfactory regarding content English usage format citation bibliographic style and consistency and is ready for submission to the Division of Graduate Education Dr Joseph A Shaw Approved for the Department of Electrical and Computer Engineering Dr Robert C Maher Approved for the Division of Graduate Education Dr Carl A Fox iv STATEMENT OF PERMISSION TO USE In presenting this thesis in partial fulfillment of the requirements for a master s degree at Montana State University I agree that the Library shall make it available to borrowers under rules of the Library If I have indicated my intention to copyright this thesis by including a copyright notice page copying is allowable only for scholarly purposes consistent with fair use as prescribed in the U S Copyright Law Requests for permission for extended quotation from or reproduction of this thesis in whole or in parts may be granted only
51. Photosynthesis in Three California Vegetation Types Ecological Applications Vol 5 1995 28 41 Fuentes D A et al Mapping carbon and water vapor fluxes in a chaparral ecosystem using vegetation indices derived from AVIRIS Remote sensing of the Environment Vol 103 2006 312 323 Nakaji Tatsuro et al Utility of spectral vegetation index for estimation of gross CO flux under varied sky conditions Remote Sensing Environment Vol 109 2007 274 284 Lawrence R L and W J Ripple Comparisons Among Vegetation Indices and Bandwise regression in a highly Disturbed Heterogeneous Landscape Mount St Helens Washington Remote Sensing of Environment Vol 64 1998 1453 1463 Maynard Catherine Lee et al Modeling Vegetation Amount Using Bandwise Regression and Ecological Site Descriptions as an Alternative to Vegetation Indices GIScience and Remote Sensing Vol 43 2006 1 14 Noomen Marlen F Continuum removed band depth analysis for detecting the effects of natural gas methane and ethane on maize reflectance Remote Sensing of Environment Vol 105 2006 262 270 132 2 Muhammed Hamed Hamid Using Hyperspectral Reflectance Data for Discrimination Between Healthy and Diseased Plants and Determination of Damage Level in Diseased Plants Proceedings of the 31st Applied Imagery Pattern Recognition Workshop 2002 Arp J W Effects of source sink relations o
52. a 32 USB4000 Miniature Fiber Optic Spectrometer Optical Layout USB4000 Installation and Operation Manual esssesseessesessessesssesressrssrresresrrseresresseseeesreses 32 DTControl Main Camera Control Panel DTControl Software Users Manal Jrs an aaa A 35 Figure 25 26 Dis 28 29 30 31 32 33 34 35 36 xiii LIST OF FIGURES CONTINUED Page Flow diagram for multiple extract planes vi the image acquisition program used m2007 ile 38 Flow diagram for calculate reflection scaffolding_grey_multiple_scenes 2 vi the reflectance and NDVI calculation program used in 2007 ay iii 41 Flow diagram for multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi the image acquisition reflectance and NDVI calculation program used in 2008 00 0 ceeeeeessceeseceeeeseeeeeneeesaeenes 44 A plot of the average digital number for each color plane as a function of integration time The full range of integration times 1 130ms is shown A plot of the average digital number for each color plane as a function of integration time The full working range of integration times 1 20ms is Shown DTi iii 55 A plot of the average digital number for each color plane as a function of gain The full range of gains 2 36dB or 1 585 3981 on a linear scale is show here ina lmearscale is i 56 A plot of the average digital number for each color plane as a func
53. age after the allotted image capture frequency Configure buffer image type and initialize asynchronous image acquisition Enter timed loop with the image capture frequency set by the user to control how often to image Initialize image capture session and configure image buffer list Enter IT control loop If all three are color plane ITs are correctly set continue with image acquisition Initialize image capture session and configure image buffer list Close current image acquisition Calculate average DN for spectralon using ROI MATLAB corrects image data for spatial non uniformity Set up image references for the three color planes Grab image data from buffer Check if current IT gives average DN for spectralon that falls within the range set by Configure buffer image type user for each color plane and initialize asynchronous image acquisition If one or more ITs are not correct continue with IT control loop Figure A5 Flow diagram for multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi the image acquisition reflectance and NDVI calculation program used in 2008 152 amd pa nanah parpena ar T a T ensa A ar pa nand ga masraf UNC 20910 mo gn reiha ro pro a a caian ad ii fon nary naa men arca pa 00 qua mi OS e pd heyei marp ms vie
54. ain and hail and nourishments and stresses The multispectral imaging approach demonstrated in this experiment therefore offers potentially lower cost does not require as much operator time and effort can image continually throughout daylight hours in clear or cloudy conditions and can image over a wide range of angles It was unfortunate that our system was pointed such that the field of view happened to almost entirely miss one of several isolated CO hotspots that happened to occur in only a few small areas of the test field The imager was able to see just the edge of one hotspot which gave us the chance to show the system s ability to detect small amounts of elevated plant stress Overall the data gained from this experiment demonstrate the effectiveness of a tower mounted multispectral imager for detecting an underground carbon dioxide leak via plant stress on a continuous basis The balance of this thesis is organized as follows Chapter 2 presents a description of the multispectral imaging system developed and used for this study Chapter 3 discusses imager system characterization and calibration Chapter 4 presents the experimental setup of the imaging system at the ZERT CO detection site and imaging methods for the 2007 and 2008 experiments Chapter 5 communicates the experimental results and a discussion of implications Chapter 6 wraps everything up with conclusions and future work 20 MULTISPECTRAL VEGETATION IMAGING Sp
55. alling a MATLAB script node to open spatial non uniformity corrections arrays discussed in detail in Chapter 3 for each color plane and save these to arrays for later use These are linear correction functions that are implemented as offset and gain arrays for each pixel of each color plane MATLAB was implemented multiple times in this program to speed up its process time A timed loop was then called so that an image capture frequency typically 3 minutes could be set by the user on the front panel Each iteration of the loop corresponds to one image for each color plane Then a flat sequence structure was used to insure that VIs would run at specific times 46 Panel contains the entire IT control code A conditional loop is used to continually change the IT until all three color planes are within the user specified range Within this loop there is a flat panel structure Having a secondary flat panel within the main flat panel could cause confusion in this discussion so for clarity this discussion will refer for example to panel 1 2 when referencing the main panel 1 and secondary panel 2 Panel 1 1 initializes an image capture session and configures a buffer list to place the images for processing Panel 1 2 configures the buffer for the specific image type and starts an asynchronous acquisition When the imager is running in 8 bit mode the image type is Grayscale U8 Panel 1 3 sets up image references for each of the three color p
56. an be used to find the location of the probe in the vegetation region using Figure 73 Dobeck 2008 These plots show that directly after a hail storm the vegetation health decreases but as the water saturates the ground and the moisture is taken in by the vegetation within a day or two the detectable plant health increases This increase in health can also be seen after significant rainfalls which are indicated with vertical black lines in all 2008 reflectance and NDVI figures 115 Mown Un mown Strip Strip N 15 5 10 Figure 73 Position and number of soil moisture probes adapted from L Dobeck Chem Dept MSU 2008 0 7 Probe 6 Probe 7 0 6 Probe 8 0 5 0 LL gt 2 2 0 4 oO gt 03 amp oO aa 0 2 0 1 7 8 7113 7 18 7 23 7 28 8 2 8 7 8 12 8 17 8 22 Date Figure 74 Soil moisture for mown strip probes adapted from L Dobeck Chem Dept MSU 2008 116 0 45 Probe 1 Probe 2 0 4 Probe 3 Probe 4 0 35 Probe 5 o 2 YA 03 O gt g BS 0 25 AA amp Y ab Y or 0 2 0 15 0 1 T T T T r 7 7 8 7 13 7 18 7 23 7 28 8 2 8 7 8 12 8 17 8 22 Date Figure 75 Soil moisture for un mown strip probes adapted from L Dobeck Chem Dept MSU 2008 0 8 T T T T T T T T T O 6f D 4P 4 Precipitation inches 0 2 4 O17 4 1 1 1 1 1 Ji 1 07 0
57. and and Qpixe is the projected solid angle of one pixel of the MS 3100 The MS 3100 provided a digital number that corresponds to the reflectance of the grey card for each spectral band The digital number obtained from the camera can be converted to the irradiance as seen by the camera by applying a gain factor and therefore can be equated to Eq 5 using ES ES Esas Pa O pixel camera A G 1 DN 6 integration TT where Gy tintegration is the gain factor of each spectral band of the camera which is a function of the IT and DN is the average digital number obtained for each spectral band of the camera Offset is not considered here since laboratory calibrations showed that this imager has a dark current that is basically 0 The gain factor can be obtained from these elements according to 84 x _ Ea Pa Q int egration i DN 7 G t Laa 7 Since the grey card was held vertically while it was imaged and the irradiance modeled by MODTRAN assumes a horizontal surface Esun must be modified to account for the angular variation of solar irradiance as indicated in Figure 53 MS3100 image of MODTRAN model grey card setup setup Grey card O 7 Qelevation 7A Enora Eocos Oreni 9 Evert EoCOS Gelevation 8 hor th Figure 53 Imager and MODTRAN geometry used to relate horizontal and vertical irradiances Next Eq 9 is solved for E E Ea 10 cos onin
58. and improve the field utility of this multispectral plant stress detection system Currently I have been using a somewhat expensive imager to capture the three spectral bands I believe that an imager could be built using a CCD imager along with a spectral filter wheel thus reducing the cost to a fraction of what the MS3100 imager cost Though this new imager would not image each of the three bands simultaneously it could image the three bands very quickly On the other hand if simultaneous imaging is needed this could also be built as I demonstrated with the modeled imager that employed the use of dichroic surfaces spectral trim filters and three CCDs basically the same imager as the MS3100 Also to reduce the price three bands may not be needed as I have shown not much was gained from the green spectral band so it may be practical to use only red and near infrared bands 129 To improve this system it might be helpful to introduce a slightly different red spectral band Data from previous studies Carter Responses 1993 Horler Dockray and Barber 1983 Rock Hoshizaki and Miller 1988 Curran Dungan and Gholz 1990 Cibula and Carter 1992 suggests that it may be more useful to have a band that is centered on 710 nm instead of 670 nm along with a smaller bandwidth The use of a 670 750 nm spectral band promises higher spectral sensitivity to stress agents A more self contained imaging system would also make this a more promising inst
59. arbon dioxide via leak detection ZERT is a partnership involving DOE laboratories Los Alamos National Laboratory Lawrence Berkeley National Laboratory National Energy Technology Laboratory Lawrence Livermore National Laboratory and Pacific Northwest National Laboratory as well as universities Montana State University and West Virginia University Spangler 2005 The ZERT research goals are as follows Spangler 2005 e Development of sophisticated comprehensive computer modeling suites which predict the underground behavior of carbon dioxide e Investigation of the fundamental geochemical and hydrological issues related to underground carbon dioxide storage e Development of measurement techniques to verify storage and investigate leakage e Development of mitigation techniques and determination of best practices for reservoir management A ZERT carbon dioxide detection field site Figure 8 was set up in Bozeman MT to study how CO will diffuse through the ground into the atmosphere how this affects the soil atmosphere gas content and plant life and if we can detect the additional CO2 For two years in a row 2007 and 2008 the ZERT program has simulated the leakage of a geological CO storage features by placing a 100 m pipe horizontally about 1 8 m beneath the ground as shown with the black line in Figure 8 The pipe was fitted with multiple packers that regulate the flow of CO to promote homogenous release along the length of
60. ard to calibrate the imaging system in the field Grey cards are designed to be lambertian reflectors meaning they reflect equal radiance at all angles The grey card we used was designed to reflect 18 of the incident light which is a common reflectance reference used by photographers For the 2008 experiment we used a Labsphere Spectralon standard Figure 20 calibrated to 99 reflectance Figure 20 Spectralon reflectance standard mounted on tripod for continuous calibration of the MS 3100 imager during the 2008 ZERT field experiment J Shaw 2008 In the 2008 experiment we chose to use spectralon instead of the grey card as a calibration target because the grey card was found to have significantly non Lambertian 30 reflectance without an acceptably flat reflectance over 500 865 nm Figure 21 The spectralon was calibrated to reflect 99 of the incident light for illumination angles down to 8 from normal below which specular reflection becomes significant Care must be taken when using spectralon in that it is very lambertian but its surface normal must be oriented the same as the scene in question or a cosine correction for the difference between the illumination and viewing angles must be applied 40r 35 30 Reflectance MN n T 400 450 500 550 600 650 700 750 800 850 900 Wavelength nm Figure 21 Reflectance spectrum of photographic grey card used to calibrate the MS 3100 imager during the 2007 ZERT fie
61. as finished so the present image acquisition is closed Now the system will loop back to Panel 1 of the flat sequence structure after the time corresponding to the image capture frequency has passed When the system is shut down for the day after all the images have been acquired the program will exit the imaging loop At this point calculate reflection scaffolding _grey_multiple_scenes 2 vi must be run to calculate reflectance and NDVI The NDVI is calculated again since the calculation done during the image acquisition program was not done correctly in that NDVI was calculated from unbalanced average DNs instead of radiometrically calibrated reflectances The program calculate reflection scaffolding_grey_multiple_scenes 2 vi calculates reflectance and NDVI for three regions using ROIs for one specific day Figure 26 There is an ROI for each color plane in each vegetation region studied and for each color plane for the calibration target which gives 12 ROIs The program will display each of the scene s average DN and the reflectance for each region To start this program the user must supply the path to the image folder and a path to a reflectance NDVI spreadsheet This spreadsheet must already be created with 15 columns and enough rows to hold data for every image taken during one day Once these paths are entered the program can be started 41 calculate reflection scaffolding_grey_multiple_scenes 2 vi Create a refletance NDVI Enter
62. ata from the un mown segment in 2007 show that the vegetation has been stressed as compared to the mown segment by a large margin according to NDVI values For this segment it can be seen in Figure 63 that the plot of the NIR reflectance over time for regions 2 and 3 have small negative slopes and the red reflectance time series plots have small positive slopes Region has a greater negative NIR slope and positive red slope indicating more stress The green reflectances for all three regions are nearly constant and equal In addition the red reflectances surpass the green reflectances by the end of the experiment The date versus NDVI regressions shown in Figure 64 agree with this in that the region 1 NDVI decreases much quickly over time compared to regions 2 and 3 The region 1 NDVI is initially greater than regions 2 and 3 by about 0 07 NDVI points but by the end of the experiment the region 1 NDVI is 0 05 and 0 11 points lower than the NDVI in regions 2 and 3 respectively The region 2 NDVI values also decrease throughout the CO release though not nearly as quickly as in region 1 The region 3 NDVI values decrease almost as much as in region 2 The regression coefficient of determination 0 7256 Table 15 was able to explain the variability moderately well The regression was significant with a p value of 1 27E 08 Table 15 The p values for both the intercept and slope regression coefficients Table 16 show that region 1 and 3 were sta
63. bon dioxide CO2 water vapor H20 and methane CH3 The energy absorbed by the greenhouse gases is then re emitted with some going back to the Earth surface to create a continual chain of energy transport between the Earth and the clouds or gases This is a good thing though because without the greenhouse effect the Earth s average equilibrium temperature would be about 18 C instead of 15 C Remer 2007 However more greenhouse gases in the atmosphere will lead to greater heat retaining capacity and higher equilibrium temperature surface Figure 1 Earth surface atmosphere solar radiation absorption and emission The yellow orange lines on the left indicate that most of the Sunlight is absorbed by the Earth s surface and atmosphere The red orange lines indicate the amount of thermal radiation emitted by the Earth s surface and atmosphere Energy flux in watts per meter squared Image adapted from Kiel and Trenberth 1997 by Debbi McLean Remer 2007 Scientists have found that humans have been increasing the concentration of greenhouse gases over the past 250 years at increasing rates Since 2004 humans have released 8 billion metric tons of CO a year into the atmosphere Remer 2007 According to the Intergovernmental Panel on Climate Change IPCC since the industrial revolution carbon dioxide levels have risen from about 280 ppm to about 380 ppm today about a 35 increase IPCC 2007 These are the highest levels of
64. box to ensure no direct sun light fell into the imager s field of view or no rain fell on the system The orientation can be seen in Figures 45 and 46 For the calibration of these images we 75 imaged an 18 reflective photographic grey card every time the integration time IT was changed and when the imager was removed from or replaced to the scaffolding so that reflectances could be calculated at a later time I had to change the IT late in the morning when the sun rose to higher angles in the sky to ensure the imager did not saturate and again anytime clouds passed in front of the sun Figure 46 View of 2007 setup showing imager orientation in respect to the vegetation test strip J Shaw 2007 I began taking images about two weeks before the CO leak June 21 until about two weeks after the release August 1 I took an image every ten minutes from about 9 am to 5 pm except during a period near midday when I took the imager down from the scaffolding and imaged specific plants along the outside edge of the un mown strip I used a cart to transport the computer and related electronics around the field While imaging the specific plants I also collected spectra of the plants with the USB4000 spectrometer This was done for 47 plants There were plants at 50 m south from the pipe 40 m south 30 m south 25 m south 20 m south and then every meter until at the 76 pipe The north side was also imaged in the same way There
65. by the red circle 119 2008 Mown Segment The 2008 CO flux map Figure 72 shows that the CO flux is somewhat above background in the mown segment region 1 is barely above background in region 2 and is at background in region 3 The NDVI trends upward Figure 66 in all regions confirmed by comparing Figures 77a and 77b Figures 78a and 78b and Figures 79a and 79b especially in region 2 Also regions and 2 and regions 2 and 3 are statistically separable as indicated in Table 18 This shows that the difference in the levels of CO2 flux in these regions causes a stress on the vegetation that is detectable over the background CO concentration Even though regions 1 and 3 have different levels of CO flux the system was not able to statistically detect this difference via plant stress although viewing the NDVI plot shows that the regions are separated by the same NDVI difference throughout the experiment a difference of approximately 0 1 These regions are not statistically different owing to similarities in the slope of the linear fits Again these data seem to agree with Arp 1991 since region 2 has a much greater increase in NDVI compared to the other two regions This may have come about because region 1 may have had an overload of CO and region 2 may have had just the right sink source balance Studying the reflectances for the mown section the green bands are nearly the same in both slope and intercept indicating it is n
66. cene E DN a dark Pa DN ER DN aan s P calibration target 3 2 calibration target Finally the reflectance 1s saved to the reflectance spreadsheet The row index is controlled by the iteration index and the column is controlled by the color plane of the image and the ROI index 1 2 or 3 The first column holds the time The second third and fourth columns will hold the Green Red NIR reflectances respectively for the first region The sixth seventh and eighth columns hold the Green Red NIR reflectances respectively for the second region The tenth eleventh and twelfth columns hold the Green Red NIR reflectances respectively for the third region Once all of the reflectance calculations have completed the arrays are opened once again and NDVI is calculated by accessing the appropriate cells The array is saved one final time and program finishes At this point MATLAB was used to plot the reflectances for each region 1 2 and 3 in the mown and un mown areas 43 Table 3 Values Used to calculate reflectance for each of the color planes during the 2007 experiment with the photographic grey card Dark Current DN Spectral Reflectance of Grey Card Color Plane Green 22 83 Red 18 172 NIR 17 373 2008 Experiment For the 2008 experiment a spectrally flat 99 reflective spectralon panel was mounted permanently within the imager s FOV to provide continuous calibration The imager was controlled with the pr
67. com ic secccayecensdajasnascansnaeeeatsvansocasedaseaessansecesserseredeuneees 24 Zemax model of a MS3100 3 chip multispectral imager Here green represents the green color plane blue represents the red color plane and red represents the NIR color plane showing that the dichroic syrfaces are m d led CURS 26 Power incident on modeled 3 chip imager detectors These pictures show the central beam and the edge of field beam indicating that the optical system simultaneously produces proper images on each of the three CCDs The top is the green color plane the bottom left is the red color plane and the bottom right is the NIR Color plane iciccccccscccesicaasisctectaceedsetcuivedeunntesssvendenttons 21 Camera Link High Speed Digital Data Transmission Cable http www siliconimaging com ARTICLES CLink9 20Cable htm 28 NI PCI 1428 base and medium configuration Camera Link frame grabber card used to acquire digital images from the MS 3100 imager www ni com 29 Spectralon reflectance standard mounted on tripod for continuous calibration of the MS 3100 imager during the 2008 ZERT field experiment J Shaw 2008 iaa 29 Reflectance spectrum of photographic grey card used to calibrate the MS 3100 imager during the 2007 ZERT field experiment 30 USB4000 Miniature Fiber Optic Spectrometer and Spectralon disk that were used together to measure reflectance spectra of vegetation and calibration panels www oceanoptics com i
68. coming light into three color planes green red and NIR using prisms dichroic surfaces and spectral trim filters Figure 14 The full spectrum light strikes the first dichroic surface which transmits red and NIR light with wavelengths longer than 600 nm and reflects all light with shorter wavelengths Then the transmitted long wavelength light strikes the second dichroic surface which transmits light with wavelengths longer than 740 nm to the NIR channel CCD and reflects all light with shorter wavelengths to the red channel CCD The short wavelength light that was reflected from the first dichroic surface is directed via a prism reflection to the green channel CCD Spectral trim filters are used to narrow each of these bands as follows green 500 580 nm red 630 710 nm and NIR 735 865 nm approximating Landsat bands The optical layout is shown in Figure 14 and the spectral response curves for the three channels are shown in Figure 15 24 Computer Camera Link Figure 13 Imaging system including the MS 3100 three CCD Imager made by Geospatial Systems Inc and the small computer to run the system Dichroic RED Surfaces __ CCD Sensor INFRARED Z CA A _ Trim Filters Prism CCD Sensors GREEN Figure 14 Schematic optical layout of the MS 3100 with color infrared setup www geospatialsystems com 1 0 9 0 8 0 7 0 6 0 5 0 4 0 3 0 2 0 1 0 7 r 500 600 700 800 900 Wavelength
69. comments that far red reflectance will increase considerably if chlorophyll levels decrease slightly He also noted that in the near infrared a change in reflectance would only be expected to result from a change in leaf anatomy or water content not chlorophyll levels Carter et al Leaf 2001 He summed things up by stating that specific stress agents do not have spectral signatures So one should be able to detect changes in chlorophyll concentrations leaf anatomy and water content by analyzing both the red and NIR portions of the spectrum or by analyzing an index that combines these bands Jordan 1969 Carter et al Leaf 2001 One such index that lends itself to vegetation remote sensing measurements is the Normalized Difference Vegetation Index Rouse et al 1974 or NDVI defined as NDVI Prin Prep 1 PNR Prep In this equation pym is the reflectance of a scene in the NIR portion of the spectrum and Prep s the reflectance of a scene in the red portion of the spectrum Considering Carter s work an index like NDVI should be useful for detecting plant stress Gamon et al 1999 noted that NDVI is a good marker for canopy structure chlorophyll content nitrogen content fractional intercepted or absorbed photosynthetically active radiation 14 and potential photosynthetic activity across many different types of vegetation Nakaji et al 2007 also found a correlation of 0 82 between NDVI and fractional intercep
70. correlated with increased CO2 concentrations resulting from a CO leak In both experiments spectral reflectance data were collected with the MS 3100 imager and the NDVI was calculated in three regions region near the pipe region 2 further from the pipe and region 3 furthest from the pipe The reflectances from individual spectral bands also were processed through statistical analysis of time series plots The reflectances are shown as exploratory figures to help explain the vegetation response especially relating changes in portions of the reflectance spectrum to changes in specific plant structures Also I ran linear regressions of date versus a linear combination of the three available spectral bands and or NDVI to find a combination that was best able to describe the variability in vegetation health and best able to statistically separate vegetation regions within a segment Coefficients of determination R and p values were used to determine how well the regressions fit the data if the regression was significant and if the spectral difference in vegetation regions were Statistically separable The possible differences in the spectral combinations were explored via both the intercepts and slopes from the linear regressions A difference in slope indicates a different vegetation response to stress A difference in intercept most likely means that the vegetation started at different health values The regression analysis showed that the NDVI
71. ctances for region 1 2 3 and NDVIs are clustered separately These are then displayed on the graph using a Waveform Chart VI Panel 11 finishes the current image acquisition process To do this first IMAQ Extract Buffer is called with a 1 as the input to Buffer to Extract which clears the buffer Next IMAO Close is called to stop the current asynchronous acquisition closes all information pertinent to this acquisition and closes the IMAQ session At this point the system will loop back to Panel 1 of the flat sequence structure after the allotted Image Capture Frequency has passed When the system is shut down for the day after all the images have been acquired the program will exit the imaging loop and close the program For image post processing I wrote a program calculate reflection scaffolding_grey_multiple_scenes 3 which loads the vegetation images and calculates reflectances and NDVI using ROIs This program is the same as calculate reflection scaffolding_grey_multiple_scenes 2 except a more efficient code to read images and to write the reflectance NDVI spreadsheet was adopted
72. ctrum are most sensitive to stresses To do this he measured the reflectances of six plant species that were stressed by four biological and four physiochemical stress agents Carter Responses 1993 Reflectance measurements were made using a scanning spectroradiometer with 768 channels Stress sensitivity was found by subtracting the reflectance of non stressed vegetation control from the reflectance of stressed vegetation for each channel of the spectroradiometer then dividing this difference by the non stressed reflectance at each channel Carter Responses 1993 He found that the green reflectance spectrum near 550 nm and the red reflectance spectrum near 710 nm both increased the same amount regardless of the specific plant species or stress agent Carter Responses 1993 The increase in reflectance near 700 nm agreed with previous data Horler Dockray and Barber 1983 Rock Hoshizaki and Miller 1988 Curran 13 Dungan and Gholz 1990 Cibula and Carter 1992 in that the red edge shifts towards shorter wavelengths when a plant is stressed Carter Responses 1993 states that there are maxima of the vegetation reflectance sensitivity to stress in the 535 640 nm and 685 700 nm regions of the spectrum In a later work by Carter et al Leaf 2001 he found that the 700 nm region was the most sensitive to stresses due to the loss of chlorophyll and the absorption characteristics of chlorophyll More specifically Carter et al Leaf 2001
73. d ROI 1 2 or 3 The second third and fourth columns will hold the Green Red NIR reflectances respectively for the first region The sixth seventh and eighth columns hold the Green Red NIR reflectances respectively for the second region The tenth eleventh and twelfth columns hold the Green Red NIR reflectances respectively for the third region Table Al Values used to calculate reflectance for each of the color planes during the 2007 experiment with the photographic grey card Value Dark Current DN Spectral Reflectance of Grey Card Color Plane Green 1 34 22 83 Red 1 43 18 172 NIR 1 79 17 373 150 The Arrays are opened once again after all reflectance calculations have finished and NDVI is calculated by accessing the appropriate cells The array is saved one final time and program finishes At this point MATLAB was used to plot the reflectances for each region 1 2 and 3 for each of the mown and un mown regions NDVIs for each region were plotted together for the mown and un mown separately 2008 Experiment For the 2008 experiment a spectrally flat 99 reflective spectralon panel was mounted permanently within the imager s FOV to provide continuous calibration The imager was controlled with the program multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi flow diagram shown in Figure 95 This program runs autonomously once started It changes the IT of
74. d of the experiment the region 1 NDVI has fallen past 0 45 lower by more than 0 10 than the region 2 NDVI and lower by more than 0 20 than the region 3 NDVI The regression coefficient of determination 0 45 Table 13 was not very strong but it was nearly as strong as any other band combination Again this relatively weak correlation may have been at least partly caused by the calibration technique that was used in 2007 Nevertheless the regression was significant with a p value of 0 000057 Table 13 The p values for both the intercept and slope regression coefficients Table 14 show that regions 1 and 3 and regions 2 and 3 were statistically separable Table 13 2007 mown segment Date versus NDVI regression R and p values 0 4472 0 00005651 Table 14 2007 mown segment Date versus NDVI regression p values that distinguish between vegetation regions Regions 1 2 Regions 2 3 Regions 1 3 Intercept Term Slope Term 0 074749 0 00443 0 023617 0 108908 0 00345 0 017171 99 Region 1 Region 2 Reflectance 21 1 7 16 7 21 7 26 7 31 Date Figure 61 2007 mown segment green red and NIR reflectances for regions 1 solid 2 dash and 3 dot 07 31 P 07 25 v a a 07 19 Region 1 07 13 o Region 2 Region 3 0 45 0 5 0 55 0 6 0 65 0 7 NDVI Figure 62 2007 mown segment Date versus NDVI for regions 1 green 2 red and 3 blue 100 2007 Un mown Segment D
75. dB and the integration time was set to 18 ms 25 ms and 33 ms for the green red and NIR planes respectively I set up the imager to view the integrating sphere through a rotating linear polarizer It was found that the imager s signal changes less than 1 for different polarizer angles as shown in Figure 35 116 3 116 2 116 1 116 115 9 Average DN n co 115 7 0 100 200 300 400 500 600 700 Angle degrees Figure 35 A plot of the affect of polarization angle on the imager s response None of the characteristics investigated so far were determined to be the cause of the abnormally high and low signals found when viewing the 50 and 99 Spectralon panels So after some more brainstorming I came up with some more possibilities First I thought maybe the CCDs were not able to drain charge quickly after viewing in bright 61 conditions which could lead to a build up of residual charge in the pixels and could be seen as an increase in the dark current Also there may be a bleeding effect as the charge is read out from the CCD causing a streak of higher pixel values in the direction of the walk off this effect is frequently observed as bright streaks emanating from bright portions of an otherwise dark image Lastly there may be a spatial non uniformity which is usually present with CCD imagers and may be augmented by the super wide angle lens adapter used to increase the small FOV of the imager To test t
76. dices for plant detection Generally speaking the greater the NDVI value the healthier the plant To obtain the three spectral bands simultaneously I used the MS3100 three CCD imager made by Geospatial Systems Inc This camera is able to simultaneously split incoming light into three different color planes via dichroic surfaces and a prism When the imager is run in color infrared mode the bands obtained are near infrared 735 nm 865 nm red 630 nm 710 nm and green 500 nm 580 nm These three bands were chosen to mimic the popular Landsat satellite bands Then software written in LabVIEW and MATLAB was used along with a National Instruments PCI 1428 Frame grabber card to acquire pixel values scale images create time series plots for each color plane and calculate NDVI It was determined that there was a correlation between higher levels of CO and reduced plant health since plants subjected to higher levels of CO close to the pipe had been stressed more than plants away from the pipe where there were lower levels of CO2 CO flux data provided by J Lewicki August 25 2008 I also found verified by experimental tests in Chapter 3 and Chapter 4 that automatically changing the 19 integration time and auto referencing a stationary spectralon panel greatly increased the accuracy and stability of reflectance measurements This improved calibration provided greater confidence in the data even making it possible to see the effects of r
77. dry up wilt and die leading to a decrease in chlorophyll and water content Conversely increased CO can decrease plant water content loss by stomatal regulation thereby increasing leaf thickness Arp 1991 No matter what the stress agent with a change in chlorophyll and water content there will be a notable change in the reflectance spectrum of vegetation 22 Vegetation reflectance is also altered by short term diurnal effects of temperature humidity and light levels on photosynthetic rate and plant respiration Huck et al 1962 notes that root respiratory rates were 25 50 higher during the daylight hours than at night when temperature and humidity were held constant Also leaf respiratory rates grow exponentially due to short term temperature increase and at the same temperature respiratory rates are greater in the afternoon than in the morning with the same levels of irradiance Atkin et al 2000 This will lead to the plant drying out throughout the day causing it to look somewhat stressed in the afternoon compared to the morning Raschke 1985 notes that decreasing humidity while holding temperature and irradiance constant throughout the day will decrease the photosynthetic rate especially when there are dry soil conditions thereby decreasing chlorophyll levels On the contrary it has been found that greater irradiance leads to a greater photosynthetic rate Kalt Torres et al 1987 This will increase absorption of red wavel
78. e 1 column At this point 1 is placed in the 6 column corresponding to the values from region 1 then 2 for region two and 3 for region three Then 3 is placed in the 7 column corresponding to the values from region 1 then 2 for region two and 1 for region three Now in R the current directory must be changed to the directory where the orefl txt file is held This file was then opened using the following code refl data read table refl txt header F refl data is a name that R will use to identify the spreadsheet for future use Since there are not headers in the refl txt 93 spreadsheet file indicated by header F in the previous code names must be attached to each of the columns The following code attaches V1 to column one V2 to column two and so on attach refl data The format of the refl txt spreadsheet file is discussed in Chapter 2 Section 4 Next the linear model was computed using the following formulation refl Im1l lm V 1 V2 V3 V4 V5 factor V6 V2 factor V6 V3 factor V6 V4 factor V6 V5 factor V6 In this equation refl lIm1 is a reference for R to identify the linear model and all of the statistics associated with the model and Im is the function to compute the linear regression The equation inside the parenthesis is the equivalent of Equation 16 where V1 is y is V2 is xg V3 is xp V4 is Xnr VS is Xnpv1 factor V6 is re
79. e A1 a folder must be created to hold the measured image arrays The user must also supply the following inputs temperature sensor channel numbers channels 1 and 2 on the USB 6210 camera interface name img 0 temperature path reference to temperature spreadsheet image path reference to folder that holds images and image capture frequency how often to image Once the program has received these inputs it starts by creating two channels for temperature acquisition These output data are held until later Next a timed imaging loop was used so that an image capture frequency typically 10 minutes could be set by the user on the front panel Then a flat sequence structure was used to insure that VIs would run at specific times name identification converts and saves each color plane as and initialize asynchronous Acquire temperatures convert Loop back to acquire another image after the allotted image 137 Figure A2 multiple extract planes vi the LabVIEW block diagram for the program used to acquire images in 2007 139 140 pun 22 A Da 250 Figure A2 Continued The LabVIEW program is a graphical routine divided into a sequence of panels as indicated in Figure A2 Panel 1 of the flat sequence initializes an image capture session and configures a buffer list to place the images for processing This was done using the IMAQ init and IMAO configure list VIs IMAQ init requires th
80. e Interface Name as specified by MAX which is img0 JMAQ init then returns the IMAQ Session Out an ID for all subsequent IMAQ processes The Session Out is passed to IMAQ configure list to create a buffer To simplify this section all IMAQ VIs will be passed the Session Out ID A property node set to type IMAQ is triggered to return the Image Type As mentioned before we are running the imager in 8 bit mode so the Image Type is Grayscale U8 This information is passed to the next panel of the flat sequence Panel 2 configures the buffer for the specific image type and starts an asynchronous acquisition First an IMAO Create V1 is called to setup the type of image 141 to be buffered This requires an Image Name and the Image Type found previously using the property node The Image Name image and the Image Type Grayscale U8 are passed The VI then returns a New Image ID which is the image reference that is supplied as input to all subsequent VIs used by Vision The New Image ID is then passed to the IMAQ Configure Buffer VI With the buffer now configured IMAO Start is called the start the asynchronous acquisition Panel 3 sets up image references for each of the three color planes and for two arithmetic functions used to find NDVI To do this five IMAO Create VIs are called Each VI has the Image Type Grayscale U8 These VIs also need names to reference the images to be made I gave these the Image Names
81. e ROIs on each of the spectralon panels The calibration target ROI was placed on the 99 panel and the scene ROI was placed on the 50 panel We would therefore expect to get 50 reflectance in the output file for each color plane but we obtained values near 60 Then we changed the test by making the 99 panel the scene and the 50 panel the calibration panel and we obtained values of about 80 So both tests were off by about 20 of what they should be These calibration panel tests were repeated with the imager attributes set to values that reduced the amount of light that reached the CCDs to a minimum This was done by decreasing the gain and integration time and increasing the F This made it possible to obtain correct reflectance values but it reduced the dynamic range of the system so much 53 that it would not be effective in the field the 99 panel only registered a DN of 15 whereas the full possible range of the system is 255 DN We were told by the manufacturers of the MS 3100 that the imager was very linear no matter what attribute was changed but we wanted to make sure From an educational engineering stand point this is important to me so that I have a better understanding of the response of the imager Therefore I tested DN vs integration time DN vs gain DN vs F DN vs radiance DN vs temperature DN vs polarization angle a pixel s ability to quickly drain charge when viewing a dim scene after viewing a very br
82. e greater the difference the greater the change applied to the IT The spectralon DN difference ranges are gt 70 70 50 50 30 30 20 20 10 and lt 10 The possible IT changes are add 50 20 14 28 11 11 8 33 and 6 67 respectively For the decrease in IT the IT changes would be subtracted from the original IT instead of added The message to set the IT is sent but first the new IT is converted to a hexadecimal number and the checksum byte of the message to change the IT is calculated A false state is returned after the IT has been changed so that the program knows it will need to check the spectralon DN at least one more time before imaging Once the NIR IT check has finished the red Panel 1 9 3 and then the green Panel 1 9 5 color planes are checked After all color planes have been checked both interior flat sequence structures are exited and the logical statements indicating whether or not each color plane s IT are correctly set are checked to see if all three are correct If not the program loops back to panel 1 1 if so the program moves ahead to panel 2 to begin the vegetation imaging Panel 2 of the main flat sequence initializes an image capture session and configures a buffer list to place the images for processing Panel 3 configures the buffer for the specific image type and starts an asynchronous acquisition Panel 4 sets up image references for each of the three color planes These have the image names Green R
83. e release pipe with half the strip mowed and the other half left un mown Figure 10 The multispectral imager viewed the vegetation to about 10 m past the pipe on the northwest side In 2008 there was a 30x20m vegetation area set up for vegetation testing Figure 11 Out of this vegetation area we imaged a 4x11m section Both years I imaged a mown and an un mown section The intent was to use northwest edge furthest from the imager as a control with little to no influence from the leaking CO and the section nearest the pipe as the primary test area During the 2008 test we also imaged three specific plants on the outer edge of the un mown strip just on the northwest side of the pipe to overlap with data acquired by another researcher using a hyperspectral system 17 Unmowed Veg f f Test Stri N N p Unmowed Veg CO2 CO2 Test Strip Leak Leak Pipe Pipe Center Mowed Veg Test Strip Mowed Veg Test Strip S W end Figure 10 Vegetation Test Strip for 2007 Figure 11 Vegetation Test Strip for 2008 The first CO injection took place from July 9 2007 to July 23 2007 Images were acquired from June 21 2007 to August 1 2007 though 15 days were skipped because of scattered cloud coverage that prevented the imager from achieving reliable calibration The second CO injection took place from July 7 2008 to August 7 2008 Images were acquired from June 16 2008 to August 22 2008 Cloud coverage was not a
84. each and a description of whether each was written as part of this project or obtained otherwise First the NI IMAQ framegrabber software was installed followed by the PCI 1428 framegrabber hardware Figure 19 This along with a Camera Link cable Figures 13 and 18 was used to interface the camera data output to the computer I then installed DT Control software Figure 24 which is used to control the imager and optionally can be used to acquire images This program makes it possible to control the gain and 34 integration time IT separately for each color plane overall exposure time bit depth display modes video mode not used triggers not used white balance not used autoexposure controls turned off and image acquisition controls not used The program also shows the status of the camera connections such as the port used control status of connection from last communication image acquisition frame grabber not used and status indicating if the imager is acquiring images Table 2 Listing of software programs used routines used within each program purpose for the program and routine and source of the software Software Routine Purpose Source NI Measurement Interface imager and National amp Automation N A temperature sensor Instruments Explorer with computer MAX NI IMAQ NI Vision Acq amp Man Image acquistion amp National NI Vision y processing functions Instruments
85. ection 2 Even still images selected with the grey card held at the proper angle gave a sufficiently diffuse approximately 18 reflection that provided a usable calibration for the imager NDVI data obtained from image regions that should be separable as discussed in Section 3 of this chapter did turn out to be statistically separable p value lt 0 05 with high coefficients of determination RS from 0 4472 to 0 7256 2007 Mown Segment Data collected from the mown segment in 2007 show that the vegetation became increasingly stressed as the release proceeded Closer to the release pipe there is a negative correlation while away from the pipe there is nearly a positive correlation For this segment it can be seen in Figure 61 that both the NIR and red reflectances for the 98 control region region 3 are nearly constant while the other two regions have decreasing NIR and increasing red reflectances All three regions have nearly constant green reflectances This indicates that regions and 2 have been stressed This interpretation of the reflectance data is backed up by the fact that the NDVI values Figure 62 for each region are initially about the same and after the start of the CO release the region 1 NDVI values decrease much more quickly than the other regions Region 2 NDVI values also decrease though not as quickly as in region 1 Region 3 has NDVI values that actually increase slightly throughout the experiment By the en
86. ection array gets the name color plane name _gain_array for example NIR_gain_array The offset correction array gets the name color plane name _offset_array for example NIR_offset_array These variables are then sent to MATLAB A new variable is created in MATLAB to place the corrected array into This is gets the name color plane name _scene_cor for example NIR_scene_cor The scene is corrected using Equation 4 which uses the NIR color plane 166 as an example Then the corrected scene array is an output of the MATLAB script node for more calculations Panel 8 sends corrected scene pixel values to the ROI average calculators in this case the spectralon and vegetation ROIs There are three vegetation ROIS Also reflectances and NDVI are calculated and saved along with the current IT First of all the array is displayed on the front panel using an intensity graph with scroll bars to determine the ROI The user is able to move the scroll bars to define the ROI The positions of the scroll bars within the array are read by Subvi vi which returns the x axis and y axis indices and lengths These values are passed to an Array Subset VI which selects the ROI array out of the entire array Then the average of the ROI array is found Reflectance is calculated using Equation 3 dark current values the reflectance of the spectralon 99 the spectralon average DN and scene average DN NDVI is found using Equation 1 The reflectanc
87. ectral Response of Plants It is possible to detect plant stress by inspecting the plant s spectral response pattern temporally The absorption and reflectance characteristics of a plant come about because of the interaction of light with the constituents of plants such as chlorophyll mesophyll cell structure and water content The plant pigments that dominate the reflectance spectrum are the chlorophyll inside the collenchyma that reflects green light and the spongy parenchyma in the mesophyll that reflects near infrared NIR radiation USGS 2008 Blue and red light are absorbed by the chlorophyll and then used for energy production in photosynthesis Figure 12 Consequently healthy plants are highly reflective in the near infrared while unhealthy plants are less so In addition healthy plants are more reflective in the green than in the red and blue while unhealthy plants have higher and flatter reflectance throughout each of these bands Figure 9 Spongy Parenchymal ph Palisade EAN ram i Collenchyma As N Mesophyil SR dad LLO a gt Vein Epidermis Cuticle Leaf Cross Section Figure 12 Spectral Absorption and Reflection Characteristics of Plants http landsat usgs gov By analyzing the reflectance of each band used in the MS3100 multispectral imager green 500 580 nm red 630 710 nm and NIR 735 865 nm temporally one is 21 able monitor the temporal evolution of plant stress When a plant becomes stre
88. ed and NIR Panel 5 grabs data from the buffer and separates the three color planes into separate images Panel 6 finds the present time and converts and saves image color plane data into arrays of pixel values The arrays are saved with a file name consisting of the color plane name concatenated with a time stamp so that each image s time will be saved Each of the images was saved so that post processing may be done Panel 7 is for 50 spatial non uniformity correction A MATLAB script node was used here The scene is corrected using Equation 4 which uses the NIR color plane as an example Panel 8 sends corrected scene pixel values to the ROI average calculators in this case the spectralon and vegetation ROIs there are three vegetation ROIs for each color plane The corrected arrays are displayed on the front panel using an intensity graph with scroll bars to determine the ROI The user is able to move the scroll bars to define the ROI Reflectance is calculated using Equation 3 using the dark current values listed in Table 3 the spectral reflectance of 99 the spectralon average DN and scene average DN NDVI is found using Equation 1 and the reflectances NDVI and ITs are saved to the reflectance spreadsheet using the format discussed earlier Panel 9 sets up the initial X and Y axes for the real time displays for reflectances and NDVIs The real time graphs are initialized on the first iteration of the imaging cycle Panel 10
89. eflectance NDVI Figure be started spreadsheet as a reference for Coan A the calculated reflectance and NDVI i Reads file names from image folder and stores in 1 D array Open reflectance NDVI spreadsheet calculate and Query file name for color plane so that the correct algorithm can be applied Counts files for number of loop iterations needed save NDVI A3 Flow diagram for calculate reflection scaffolding_grey_multiple_scenes 2 vi the reflectance and NDVI calculation program used in 2007 145 EA i i El had 104 eus0s Naz auas Hay i k had 49 NFRD 104 4046 NERO 94895 NITY Ted Era bas JOY 34305 NIJYS A 2905 NIIUD E 2 gt 8 E E Figure A4 calculate reflection scaffolding_grey_multiple_scenes 2 vi the Lab VIEW block diagram for the program used calculate reflectance and NDVI in 2007 146 Figure A4 Continued 147 The program Figure A4 starts by reading the file names at the given folder path and storing these in a 1 D array so that they can be called one at a time for calculations An example file name is 125432 PMNIR The first six digits are the time of the image and the first two letters after the space indicate morning or afternoon 12 54 32 PM The rest of the file name is the color plane the image refers to IR is Near infrared R is red and G is green For images that have the grey card present grey
90. egion number as the categorical variable R outputs linear regression coefficients along with statistical values such as adjusted R and p values The linear regression for a general fit can be seen in Equation 16 92 Y Po Boxe PrXr BurXnm Brovi Ano T GF Xp t 16 Xnr 57 Xypvr 57 Here y is the date time response variable is the y intercept linear regression coefficient fg is the slope for the green band xg is the green reflectance predictor variable Jp is the slope for the red band xp is the red reflectance Axi is the slope for the NIR band xnie is the NIR reflectance Pxpv1 is the slope for the NDVI and ris the region number categorical variable All the P values were calculated by R To find the best band combination I started by analyzing the p values for each of the predictor and categorical variables in Equation 16 The values that were not significant were removed one at a time and the regression was run again including only the significant values This approach leads to a better model to explain the variability in plant health and to statistically distinguish between vegetation regions To run R first this file must be altered for the statistical computation Inside the refl txt file each region s green red NIR and NDVI values must be placed in columns 2 3 4 and 5 respectively in descending order from region 1 to region 3 Also the corresponding date times must be placed in th
91. engths thereby increasing NDVI making the plants look healthier It is hard to say if the photosynthetic rate will increase or decrease on a whole Considering environmental factors and diurnal variations the CO flux water level temperature and rain data collected by my colleagues and myself are very important to understand what we are seeing in terms of plant stress To detect plant stress due to a CO leak I designed a system consisting of a 3 color CCD imager mounted on a scaffold operated by a small computer automatically in the field and calibrated by a photographic grey card or spectralon panel The imager collects green red and NIR portions of the spectrum separately for a vegetation scene and the calibration target simultaneously This data is then used by the computer to 23 calculate reflectances for each band and NDVI The system computes reflectances and NDVI for two vegetation segments mown and un mown vegetation split into three sections each one near the pipe one far from the pipe and one in the middle Then the reflectances and NDVI are plotted so the time evolution of the reflectance and NDVI could be analyzed Imaging Hardware We have developed and deployed a multispectral imaging system to detect plant stress The system is based on a Geospatial Systems Inc MS 3100 3 chip Charge Coupled Device CCD imager Figure 13 that images in three spectral bands of interest simultaneously The imager splits in
92. ens was chosen since it has a relatively short focal length and unlike newer lenses it does not have a tab near the threads that will not 28 allow them to connect to the MS 3100 The 20 mm lens was set to f 8 and focused at infinity The MS 3100 imager interfaces to a control computer with a Camera Link connection Figures 18 and 13 that allows high speed image data transfer to a National Instruments PCI 1428 digital frame grabber card Figure 19 This frame grabber allows data to be transferred in base 8 bit or medium 12 bit Camera Link configuration Although the MS 3100 can record 8 bit or 10 bit images we were not able to use the 10 bit mode because the frame grabber needed either 8 bit or 12 bit data Consequently the imager was run in 8 bit mode giving us 0 255 digital numbers DN or grey levels for each color plane a Generic End Detail MOR 26 Position Plug Both Ends 26 Position Cate 26 Position High Density See pinout for catie construction High Density Mini D Ribbon Mini D Ribbon MDR Male Plug MOR Male Plug q pi 2x 2x Thumbecrews Thumbscreas A Figure 18 Camera Link High Speed Digital Data Transmission Cable http www siliconimaging com ARTICLES CLink 20Cable htm 29 Figure 19 NI PCI 1428 base and medium configuration Camera Link frame grabber card used to acquire digital images from the MS 3100 imager www ni com For the 2007 experiment we used a Deltal photographic grey c
93. era interface name img 0 image path directory to save images to reflectance path image capture frequency how often the imager should image and the upper and lower limits for the average DN of the spectralon desired for each color plane determines how sensitive the IT control will be The program starts by calling a MATLAB script node to open spatial non uniformity corrections arrays discussed in detail in Chapter 3 for each color plane and save these to arrays for later use These are linear correction functions that are implemented as offset and gain arrays for each pixel of each color plane MATLAB was implemented multiple times in this program to speed up its process time A timed loop was then called so that an image capture frequency typically 3 minutes could be set by the user on the front panel Each iteration of the loop corresponds to one image for each color plane Then a flat sequence structure was used to insure that VIs would run at specific times Panel 1 contains the entire IT control code A conditional loop is used to continue changing the IT until all three color planes are within the user specified ranges Within this loop there is a flat panel structure Having a secondary flat panel within the main flat panel could cause confusion in this discussion so for ease of discussion when referencing for example panel 1 of the main and panel 2 of the secondary I will call it panel 1 2 157 Panel 1 1 initializ
94. es NDVI and ITs are saved to the reflectance spreadsheet using the format discussed earlier using a Write To Spreadsheet File VI Panel 9 sets up the initial X and Y axes for the real time displays for reflectances and NDVIs It will only do this the first iteration of the imaging cycle So it will check if it is the first iteration and if so it will step into the true statement of a conditional structure A Get Date Time In Seconds is called and returns the time in seconds since 12 00AM January 1 1904 This is converted to a true time year month day hour minute and second in the form of cluster data type using LV70 TimestampToDate Record VI Then only the hour minute and second are picked using an Unbundle By Name VI which unbundles cluster and selects the data specified by a name given by the programmer A property node for each vegetation region and NDVI is set up to initialize 167 real time graphs with the time as the X axis Offset input 0 as the X pin input 10 as the Xmax input and a cleared cluster of three elements to clear the graph as the History input If it is greater than the first iteration nothing happens in this panel Panel 10 updates the X axis and data points of the real time graphs An Elapsed Time VI is called to figure out the time difference since last time this VI was called This is an input to a property node setup to change the X scale Multiplier which updates the X axis Then the three refle
95. es an image capture session and configures a buffer list to place the images for processing This was done using the IMAQ init and IMAQ configure list VIs IMAO init requires the Interface Name as specified by MAX which is img0 IMAQ init then returns the IMAQ Session Out an ID for all subsequent IMAQ processes The Session Out is passed to IMAQ configure list to create a buffer To simplify this section all IMAQ VIs will be passed the Session Out ID A property node set to type IMAQ is triggered to return the Image Type As mentioned before we are running the imager in 8 bit mode so the Image Type is Grayscale U8 This information is passed to the next panel of the flat sequence Panel 1 2 configures the buffer for the specific image type and starts an asynchronous acquisition First an IMAO Create VI is called to setup the type of image to be buffered This requires an Image Name and the Image Type found previously using the property node The Image Name image and the Image Type Grayscale U8 are passed The VI then returns a New Image ID which is the image reference that is supplied as input to all subsequent VIs used by Vision The New Image ID is then passed to the IMAO Configure Buffer VI With the buffer now configured IMAO Start is called the start the asynchronous acquisition Panel 1 3 sets up image references for each of the three color planes To do this three IMAO Create VIs are called Each VI has the Image
96. ethod oooocccnnococcnncccnnncccnnnnnononcnonancninnnnons 74 2008 Experimental Setup and Imaging Method oooconocccnoncnonnconncnonnnnnnnona nono nonncnnnos 77 Procedures for Calculating Retenes ise 80 2007 Procedure Using Photographic Grey Card eeeesceeesscecsececeeeeeeseeeeeeneeeesaes 80 2007 Procedure Using Modeled Irradiance ooonnccnnnccnoncconnnonnnonrncnnnncnnnnonnccrnn nono ncnnos 83 2008 Procedure Using Spectralon Panels oooooncccnnococinocaconncncnoncnononanononcnonancncnnnnnos 88 EXPERIMENTAL RESULTS AND DISCUSSION coocococccocconncononnnonncnnccnnannncnnncnncnnncnns 95 207 APR o boo nd 97 2007 Mown SM usina sidra sets edades 97 2007 Un mown S egment id E a acid 100 2009 Experimental Resulta 102 ZOOS M wi Se met O ach ane E A E E enc 102 20058 UN MOWA SE MO etorre reca eona t iaa 105 Individual Plants Within Un mown SegMeNt oococnnocccconcccnoncnononcnononcnonnnanonanccinnncnos 107 ZO DISCUSSION id diia 109 2007 Mown Segment e na E R E E cates ol tl Ales conan acetate 110 2007 Un mown Segment anaranjada oca 111 2008 PISCUSSIOM tddi tedio 113 ZOOS INOW So MIST 28 0 tesa ec ncaa cg i E E evade ese eee Sanaa ia 119 2008 Un mown Segment ais 121 Individual Plants Within Un mown SegMeNt oococonoccconccccoonccnnoncnononancnnncnnnanacinnnanos 123 viii TABLE OF CONTENTS CONTINUED 6 CONCLUSIONS AND FUTURE WORK csssorcsossssseosasosnostessacesavsseossvseconnonssves 126 BIBLIOGRAPO Vo RA A E AR N 131
97. evelopment Module adds hundreds of VIs that allow high and low level acquisition display processing file I O pattern matching particle 37 analysis measurement tools amongst many others We used only a few of the VIs that came with the NI Vision Development Module Now the imager can be easily interfaced with LabVIEW LabVIEW was used for image and temperature acquisition in 2007 and image acquisition and processing in 2008 2007 Experiment For the 2007 experiment the programs multiple extract planes vi and calculate reflection scaffolding_grey_multiple_scenes 2 vi were used The first program acquires images of the vegetation test strip saves the temperature during an image acquisition and saves pixel value arrays of each of the color planes and the calculated NDVI The second program was used to calculate reflection and NDVI using only images of the vegetation with the photographic grey card calibration target in the image This program uses regions of interest ROIs to select the grey card and three separately analyzed vegetation regions within both the mown and un mown segments of the test area The second program provided only a few calibrated images each day but was used after two other attempted calibration methods did not work sufficiently well These initial calibration methods were based on 1 using gray card images recorded whenever the IT changed and 2 using continuous measurements of solar irradiance at the surface T
98. fers to IR is Near infrared R is red and G is green For images that have the grey card present grey would be added to the end for example if the file name mentioned above were a grey card image it would be 125432 PMNIRgrey Next a For loop is entered where each iteration calculates reflectance and NDVI for the ROIs for a specific file Within the loop the first file name is read and is appended to the folder path so that it can be opened in the future The file name is then displayed on the front panel so the user knows what the current file is This file name is then saved to the 42 reflectance NDVI spreadsheet set up by the user Now a query is performed on the file name to determine which color plane it is so that the correct algorithm can be applied Once the program has figured out what color plane the file is the file is opened and the pixel values are then sent to ROI average calculators which calculate the average DN of the selected ROI The array is displayed on the front panel using an intensity graph with scroll bars to determine the ROI The user is able to move the scroll bars to define the ROI The ROI average calculator will loop until the user presses the stop button insuring the best possible region has been chosen Next the reflectance is calculated using Equation 3 with values specific to each color plane s spectral band dark current and grey card reflectance These values are seen in Table 3 DN a s
99. for integration times over the range of 1 130 ms in increments of 5 ms The resulting plot Figure 28 shows that the red and NIR CCDs are fairly linear in response to a change in integration time but that the green plane has the smallest linear range I then noticed the response seemed very linear for all the CCDs in the range from 1 20 ms which is the working range used in the field for our experiments Figure 29 shows the response over this limited range of ITs The R values are 0 9812 0 9995 and 0 9998 for the green red and NIR color planes respectively Luckily the green plane is not as important to us as the red and NIR planes 250 o ua H O Green BBooooo O Red TE eo a A O x 200 x NIR o o x o o o O y o o o x 150 o N A a o o e ud a o yr T 100 o o x o x x O x o o x x s504 Os O a x x Q 71 A A A E A E 0 20 40 60 80 100 120 Integration Time ms Figure 28 A plot of the average digital number for each color plane as a function of integration time The full range of integration times 1 130ms is shown here 55 250p Green O Red x NIR 200 Green Linear Fit y 10 08 x 15 23 o Red Linear Fit y 4 45 x 2 01 abia NR Linear Fit y 3 27 0 85 150 Average DN 100 50 E fi 0 2 4 6 8 10 12 14 16 18 20 Integration Time ms Figure 29 A plot of the average digital number for each color plane as a function of integration time The fu
100. gee a e S A E 0 2 4 6 8 10 12 14 16 Gain linear Figure 31 A plot of the average digital number for each color plane as a function of gain A range of gains 2 12 dB or 1 585 15 85 on a linear scale is shown here in a linear scale 57 Next we plotted the average DN versus a change in F as shown in Figure 32 Changing the F by manually turning the aperture ring on the manual focus Nikon lens changes the diameter of the entrance pupil which in turn changes the amount of light that reaches the CCDs by allowing light to be gathered over a larger portion of the lens The gain was set to 5 dB 10 dB and 10 dB for the green red and NIR planes respectively The integration time was set to 3 ms 5 ms and 5 ms for the green red and NIR planes respectively The focus was set to 25 cm Here the average DN is plotted versus 1 F since this value is proportional to the power on the CCDs The R values are 0 9995 0 9995 and 0 9996 for the green red and NIR color planes respectively 120 Green O Red 100 F x NIR Green Linear Fit y 3612 7 x 0 2 Red Linear Fit y 3164 8 x 0 6 NIR Linear Fit y 2249 6 Average DN o o T O Ss 0 0 005 0 01 0 015 0 02 0 025 0 03 0 035 FA 2 Figure 32 A plot of average DN as a function of F 58 Next the radiance from the integrating sphere was changed while the camera settings were maintained constant The radiance was changed b
101. gh in the un mown region 1 is above background in region 2 and is at background in region 3 which is basically the same as the flux distribution in the mown segment This flux distribution shows great agreement with the NDVI Figure 64 for the un mown segment though the stress signature has been somewhat hidden by the presence of a veil of tall nearly dead grass obscuring the spectral signature of the healthier underbrush This can be seen by comparing the CO flux and NDVI trends to the mown segment Both segments have the 112 same flux and nearly the same starting NDVI values yet the un mown segment has much lower NDVI values at the end of the experiment Even though the veil of tall dead grass exists the system was able to detect vegetation stress In fact there was more stress detected in this segment than in the mown segment This can be seen in the higher R and lower p values in the un mown segment compared to those in the mown segment It is possible that the higher variability in the un mown segment was explained better in this case since NDVI is naturally able to distinguish between vegetation regions The NDVI was able to statistically separate regions and 3 as indicated in Table 16 Comparing these results to those from the mown segment suggests that an un mown segment reacts to increased levels of CO differently than a mown segment in that higher amounts of CO are needed to negatively affect the total health
102. gion B values are computed when the Im function is implemented To view these values refl lIm1 was entered Then summary refl Im1 was called which has multiple outputs but I especially paid attention to p values for each of the 2 values regression coefficients for the total linear regression which indicate if the band in question brings any significance to the regression and if the regression is itself significant I also analyzed the adjusted R which indicates how good the fit is These outputs would only compute statistics for regions and 2 factor V6 2 and regions 2 and 3 factor V6 so the program must be run again with V6 replaced with V7 This would output statistics for regions 2and 3 factor V7 2 anova refl lm1 was called to output the analysis of variance table This was analyzed to see if the categorical variables brought any significance to the model To plot the residuals the following code was used plot josh lm1 fit josh lm1 resid and 94 abline 0 0 To find the best combination of bands and NDVI the method of throwing out insignificant variables was applied and the program would be run again The results of this statistical analysis are presented and discussed in the next chapter 95 EXPERIMENTAL RESULTS AND DISCUSSION NDVI data from 2007 and 2008 indicate the ability of a multispectral imaging system to detect plant stress or nourishment that is
103. guishable at the p 0 05 level This shows that the difference in the levels of CO flux in these regions cause a stress on the vegetation that is detectable over the background which is expected Even 111 though regions 1 and 2 have different levels of CO flux the system was not able to statistically detect this difference via plant stress although the NDVI plot shows that the regions are initially the same and end up separated by an NDVI difference of more than 0 10 For the mown section the green and red bands were able to spectrally distinguish between regions and 3 and regions 2 and 3 This implies an ability of the green and red bands to explain variability in the spectral response of vegetation when the vegetation has been mowed However these bands were not able to explain variability in the vegetation regions as well as NDVI The NIR band alone distinguished between regions 1 and 2 regions 1 and 3 but not regions 2 and 3 This along with the agreement between NDVI and the green and red bands leads to a tentative conclusion that NIR reflectance alone is not able to accurately detect stress in a mown segment and there may be an angle dependent reflectance effect contributing to this Overall taking p values and R values into account the NDVI is the strongest and most consistent parameter for detecting plant stress in a mown segment 2007 Un mown Segment The 2007 CO flux map Figure 71 shows that the CO flux is hi
104. hanges To do this I set the imager s attributes to values similar to those used in the field the gain was set to 3 dB 5 dB 5 dB and the integration time was set to 18 ms 25 ms and 33 ms for the green red and NIR planes respectively The lens cap was left on so that we would measure only the dark current The imager was placed in a darkened thermal chamber and the temperature was increased from 18 C to 43 C in 5 C increments After each temperature change the imager waited until it was at an equilibrium temperature before recording images All pixels of the imaging array were averaged giving an average DN for each color plane I found that across the temperatures evaluated there was only a change of about 0 07 DN for the green color plane 0 04 DN for the red color plane and 0 02 for the NIR color plane as shown in Figure 34 I also found that the standard deviation of the spatial variation of dark noise was very low 1 79 DN for the green color plane 1 34 DN for the red color plane and 1 43 DN for the NIR color plane 1 12 gt a L o mn Dark Current Average Dn 1 04 15 20 25 30 35 40 45 Temperature C Figure 34 A plot of the effect of camera temperature on the imager s response 60 I also tested the response of the imager to a changing polarization state for the incoming light Again I set the imager s attributes to values similar to those used in the field the gain was set to 3 dB 5 dB and 5
105. he CCDs ability to drain charge the system was again set up to view the integrating sphere The dark current was first measured by placing the lens cap on the imager Then images were acquired every thirty seconds for an hour during which time the imager was continually illuminated with bright light A DN was read out at the beginning and end of the bright conditions At the end of an hour of bright illumination the dark current was measured again As can be seen in Table 6 the CCDs drain charge very effectively since the dark current returns immediately to the initial values after viewing bright light Table 6 Values obtained in a test of the CCDs ability to quickly drain charge after viewing bright conditions Green DN Red DN NIR DN Dark current before bright condition 0 02 0 5 0 7 DN at beginning of bright conditions 182 149 164 DN at end of bright conditions 1hr elapsed 189 9 156 3 170 45 Dark current after bright condition 0 02 0 5 0 7 To see if there is a bleeding effect the imager was moved to a larger distance from the integrating sphere so the sphere only filled a small center portion of the imager s 62 FOV Then the DN was measured around the bright spot created by the sphere showing that there is a very small bleeding effect in the read out direction to the right when viewing an image but this effect is only about 1 part in 255 as indicated in Table 7 Table 7
106. he lack of success of these methods was because the periodic grey card images did not track conditions as they changed with sun angle or clouds and because the location of the solar irradiance measurements was too far distant from the ZERT test site These two calibration methods will be described in more detail in Chapter 4 Section 2 38 multiple extract planes vi Create a folder to place image pixel value arrays and a spreadsheet for temperatures Initialize image capture session and configure image buffer list Finds present time for file an array of pixel values User supplies temperature Configure buffer image type sensor channels camera interface name image capture frequency and paths to image and temperature files Then user starts the program image acquisition to Celsius and save to the temperature spreadsheet Close current image and temperature acquisition Set up image references for the three color planes and arithmetic functions needed to find NDVI Enter timed loop with the image capture frequency set by the user to control how often to image Grab image data from buffer and display images on front panel capture frequency Figure 25 Flow diagram for multiple extract planes vi the image acquisition program used in 2007 To run the primary image acquisition program used in 2007 multiple extract planes vi Figure 25 a folder
107. he surface The lack of success of these methods was because the periodic grey card images did not track conditions as they changed with sun angle or clouds and because the location of the solar irradiance measurements was too far distant from the ZERT test site These two calibration methods will be described in more detail in Chapter 4 Section 2 136 multiple extract planes vi Create a folder to place image pixel value arrays and a spreadsheet for temperatures Initialize image capture session and configure image buffer list Finds present time for file an array of pixel values User supplies temperature Configure buffer image type sensor channels camera interface name image capture frequency and paths to image and temperature files Then user starts the program image acquisition to Celsius and save to the temperature spreadsheet Close current image and temperature acquisition Set up image references for the three color planes and arithmetic functions needed to find NDVI Enter timed loop with the image capture frequency set by the user to control how often to image Grab image data from buffer and display images on front panel capture frequency Figure Al Flow diagram for multiple extract planes vi the image acquisition program used in 2007 To run the primary image acquisition program used in 2007 multiple extract planes vi Figur
108. ically separable Reflectance data for this segment were also collected with the USB 4000 spectrometer but since the data were not conclusive they are not presented here This is discussed in more detail in Section 4 of this chapter In the figures below for the 2008 experiment the green solid line is the start of the experiment the red solid line is the end of the experiment the solid black lines are hail rain events and the dashed lines are rain events Reflectance 50 46 NIR A 40 104 Region 1 ter Region 2 14 Region 3 1 1 1 1 1 1 07411 07116 07 21 07 26 07 31 08 05 Date Figure 65 2008 mown segment green red and NIR reflectances for regions 1 solid 2 dash and 3 dot 08 15 Region 1 o Region 2 oaio Region 3 08 05 07 31 Date 07 21 07116 071 g m T 07 26 1 ale 1 a Lake 1 1 0 55 0 6 0 65 0 7 0 75 0 8 0 85 NDVI Figure 66 2008 mown segment Date versus NDVI for regions 1 green 2 red and 3 blue 105 Table 17 2008 mown segment Date versus NDVI regression R and p values Regression p value NDVI 0 7273 lt 2 2e 16 Table 18 2008 mown segment Date versus NDVI regression p values that distinguish between vegetation regions Regions 1 2 Regions 2 3 Regions 1 3 Intercept Term 0 000881 0 00436 0 604922 Slope Term 0 000305 0 01291 0 172055 2008 Un mown Segment Data from the un mo
109. ight one pixel smear and spatial non uniformity I found that the imager works very well under the conditions we exposed it to in the field The only problem that required compensation was a spatial non uniformity caused presumably by the extra wide angle lens adapter used to increase the imager s field of view However we corrected for this by finding and applying a linear fall off correction to each pixel Towards the end of the experiment I also characterized the angular reflectance properties of a spectralon panel to gain a better understanding of lambertian surfaces Initially I believed these problems could be attributed to one or more of the following a change in integration time gain F or radiance To test these we operated the imager looking into an integrating sphere to ensure spatially uniform radiance across the CCDs We then changed one of the attributes while keeping all others constant The average DN for the entire scene was then plotted versus the attribute change We also explored the spatial variation of DN for this uniform scene which led to the non uniformity correction described later 54 The integration time is the amount of time in milliseconds that the electronic shutter is open This can be set for each CCD separately The gain was set to 2 dB 4 dB and 4 dB for the green red and NIR planes respectively The F was set to 11 and the focus was set to approximately 25 cm initially ran this test
110. igures the buffer for the specific image type and starts an asynchronous acquisition First an IMAO Create VI is called to setup the type of image to be buffered This requires an Image Name and the Image Type found previously using the property node The Image Name image and the Image Type Grayscale U8 are passed The VI then returns a New Image ID which is the image reference that is supplied as input to all subsequent VIs used by Vision The New Image ID is then passed to the IMAQ Configure Buffer VI With the buffer now configured IMAO Start is called the start the asynchronous acquisition The Panel 4 sets up image references for each of the three color planes To do these three IMAQ Create VIs are called Each VI has the Image Type Grayscale U8 These VIs also need names to reference the images to be made I gave these the Image Names Green Red and NIR Each VI sets a reference to its specific color plane and passes it to the buffer These VIs return a New Image ID for each of the color planes functions which can be used in subsequent VIs to retrieve specific image data The Panel 5 grabs data from the buffer and separates the three color planes into separate images IMAQ Get Buffer is called to grab the buffer buffer 0 containing the image data It returns an Image Out which is a multiplexed Color infrared image Next the three color planes must be extracted This VI takes the New Image IDs N
111. ing vegetation into different levels of healthy and non healthy groups has been shown Muhammed 2002 used a spectroradiometer system with 164 channels in the 360 900 nm spectral region to measure reflectance Leaf damage levels were measured visually at the same time that reflectances were measured It was shown that using hyperspectral data there is the possibility of separating vegetation into 8 differing health levels corresponding to leaf damage levels from 0 59798 to 76 15 Mohammed 2002 He achieved 94 for a modified correlation coefficient and sum of squared differences Mohammed 2002 This is a great achievement but 16 dealing with the overwhelming amount of data inherent and cost associated with this type of system can be bothersome especially when a multispectral system may be able to delineate between stressed and non stressed plants sufficiently well with only a few spectral points Seeing the need for CO leak detection and potential ability of multispectral imagers to detect plant stress due to a CO leak I measured the spectral response of plants at the ZERT CO detection site and analyzed single bands and NDVI in a temporal fashion During 2007 and 2008 the ZERT CO detection experiment was held in a field just west of Montana State University in Bozeman MT Multispectral imaging data were collected during both experiments In 2007 a 100m vegetation test strip was set orthogonally to the center of the carbon dioxid
112. ions Regions 1 and 2 are not statistically different owing to similarities in the slope of the linear fits The NDVI is also able to explain variability in plant health in comparison to vegetation region much better than any band combination Overall taking p values and R values into account the NDVI is the strongest of the indicators examined here for detecting plant stress in an un mown segment The NDVI was again able to detect the effects of rain and hail in this un mown segment Within a day after each rain storm the NDVI values level out indicating that the system is able to detect this nourishment The two days of hail had less of an effect on the NDVI values in the un mown segment since the healthy vegetation in this segment was defended by the veil of tall nearly dead grass but there was still a small effect Again as in the mown segment the moisture from the hail storms was taken up by the vegetation and the NDVI value jumped right back up to a level consistent with that of the linear fit within 1 2 days Individual Plants Within Un mown Segment Plants 8 9 and 10 are located on the outside edge of the un mown strip just inside region at the junction of regions 1 and 2 The 2008 CO flux map Figure 72 shows that the CO flux where plants 8 9 and 10 are located is quite high Plant 8 was 123 closest to the highest CO concentration while plant 10 was the furthest away Again since this segment has not been mow
113. lanes I gave these the Image Names Green Red and NIR Panel 1 4 grabs data from the buffer separates the three color planes and converts the images to arrays of pixel values Panel 1 5 is for the spatial non uniformity correction implemented with a MATLAB script node The scene is corrected using Equation 4 shown for the NIR color plane but available for implementation with any color plane NIR _ scene _ cor NIR _ scene NIR _ gain _ array NIR _ offset _ array 4 Panel 1 6 sends corrected scene pixel values to the ROI average calculators which calculate the average DN of the selected ROI in this case the exposure control ROI First of all the array is displayed on the front panel using an intensity graph with scroll bars to determine the ROI The user is able to move the scroll bars to define the ROI The average of the ROI array is found panel 1 7 closes the current image acquisition panel 1 8 causes the program to wait for 5 seconds before moving on to the next panel and panel 1 9 contains a third flat sequence structure that checks the pixel values within the calibration target area and adjusts the IT if they are no longer within the 47 average DN range set by the user These panels will be denoted as 1 9 1 for the first panel of the third flat sequence structure Panel 1 9 1 is the NIR IT control Initially serial communication is setup with the imager for reading and writing attributes to the imager The se
114. ld experiment For the 2007 experiment we used a NI USB 6210 digital analog I O module along with a National Semiconductor LM335 temperature sensor to measure the ambient temperature around the imaging system This was done in case a temperature response correction was needed but we found later that it was not needed because measurements showed that the MS 3100 imager is very stable across the temperature range we experienced in the field To test this in between the 2007 and 2008 experiments we 31 placed the camera in a thermal chamber and recorded dark current images across a range of temperatures greater than what we would see in the field without observing any measurable change This is described in more detail in Chapter 3 Spectrometer Hardware Another optical sensor that was used extensively in this project was an Ocean Optics USB4000 spectrometer Figure 22 This is a fiber fed hand held spectrometer that was used to measure the reflectance of calibration panels and vegetation The optical layout of this spectrometer is shown in Figure 23 with explanatory labels listed in Table 1 The spectrometer uses a Toshiba TCD1304AP Linear CCD array detector and disperses light with a diffraction grating over a bandwidth of 350 1100 nm There are 3648 pixels each with a size of 8 um x 200 um The spectral resolution of a pixel at full width at half maximum FWHM is approximately 0 21 nm The aperture stop is 25 um The analog to digital c
115. ll working range of integration times 1 20ms is shown here Next we tested the linearity of signal with a change in gain The gain setting changes the settings on the electronic amplifiers used to amplify the signals read from the CCDs These amplifiers can easily have significantly nonlinear behavior The integration time was set to 3 ms 5 ms and 5 ms for the green red and NIR planes respectively F was set to 11 and the focus was set to 25 cm The gain values are in dB so I converted them to a linear scale gain linear p 6 10 Figure 30 shows that the CCDs responses over the full gain range are very nonlinear But if only the working range is considered it looks much more linear as shown in Figure 31 The R values are 0 9850 0 9848 and 0 9847 for the green red and NIR color planes Again when the system is in the field the gain is not changed so this was determined not to be the source of our observed calibration problem 56 300 260 8 o g E E Average DN a o o 0 500 1000 1500 2000 2500 3000 3500 4000 Gain linear Figure 30 A plot of the average digital number for each color plane as a function of gain The full range of gains 2 36 dB or 1 585 3981 on a linear scale is shown here in a linear scale 70r Green O Red 60 F x NIR o Green Linear Fit y 2 52 x 20 31 5 Red Linear Fit y 1 43 x 11 00 NIR Linear Fit y 1 00 x 7 54 z 40 o D 2 o 4 30 o x 20 x 10
116. loop All that is left is to close the temperature acquisition This is done using a DAQmx Clear Task VI At this point calculate reflection scaffolding_grey_multiple_scenes 2 vi must be run to calculate reflectance and NDVI The program calculate reflection scaffolding_grey_multiple_scenes 2 vi calculates reflectance and NDVI for three regions using ROIs for one specific day flow 144 diagram shown in Figure A3 There is a ROI for each color plane in each vegetation region studied and for each color plane for the calibration target which gives 12 ROIs The program will display each of the scene s average DN and the reflectance for each region To start this program the user must supply the path to the image folder and a path to a reflectance NDVI spreadsheet This spreadsheet must already be created with 15 columns and enough rows to hold data for every image taken during one day Once these paths are entered the program can be started calculate reflection scaffolding _grey_multiple_scenes 2 vi Create a refletance NDVI Enter for loop where each spreadsheet iteration is a reflectance NDVI calculation for one image Calculate average DN for each vegetation region and the calibration target using regions of interest User supplies path to folder of Read file name out of the file images and name array at the associated Jorelfectance NDVI iteration and save this to spreadsheet The program can r
117. lues and R values Table 18 Table 20 and Table 22 The spectral bands in 2008 seem to have increased statistical significance compared to the 2007 results presumably owing to the better calibration technique and more samples Though the bands have higher accuracy band combinations are still not as accurate as the NDVI for distinguishing between the stresses on different vegetation regions R values observed for all the data have been reduced by two outlying data points 21 June 2008 and 3 Aug 2008 these can be seen distinctively for the NIR band These data points do not appear to be erroneous points because they were consistently measured on those days However removing these two points cause R values to increase significantly The p values also change though not enough to change the outcome of the F tests 114 J Pipe End Pipe Center Pipe End Figure 72 2008 CO flux map of the ZERT CO2 Detection site adapted from J Lewicki Lawrence Berkeley National Laboratory 2008 The improved calibration techniques also lead to the system s ability to detect the negative and positive effects of hail and rain This can be seen by comparing the individual NDVI points in Figures 66 68 and 70 with the soil moisture data Figure 74 and 75 Dobeck 2008 and rain data Figure 76 Lewicki 2008 The soil moisture data are shown with separate plots for the mown Figure 74 and un mown Figure 75 segments The probe number on these plots c
118. must be created to hold the measured image arrays The user must also supply the following inputs temperature sensor channel numbers channels 1 and 2 on the USB 6210 camera interface name img 0 temperature path reference to temperature spreadsheet image path reference to folder that holds images and image capture frequency how often to image Once the program has received these inputs it starts by creating two channels for temperature acquisition These output data are held until later Next a timed imaging loop was used so that an image capture frequency typically 10 minutes could be set by the user on the front panel Then a flat sequence structure was used to insure that VIs would run at specific times name identification converts and saves each color plane as and initialize asynchronous Acquire temperatures convert Loop back to acquire another image after the allotted image 39 The LabVIEW program is a graphical routine divided into a sequence of panels as indicated in Figure A2 of the Appendix Panel 1 of the flat sequence initializes an image capture session and configures a buffer list to place the images for processing Panel 2 configures the buffer for the specific image type and starts an asynchronous acquisition The buffer image type is Grayscale U8 Panel 3 sets up image references for each of the three color planes and for two arithmetic functions used to find NDVI I gave these the image
119. n photosynthetic acclimation to elevated CO2 Plant Cell and Environment Vol 14 1991 869 875 8K imball B A J R Mauney F S Nakayama and S B Idso Effects of increasing atmospheric CO on vegetation Plant Ecology Vol 104 1993 65 75 Huck M G R H Hageman and J B Hanson Diurnal Variation in Root Respiration Plant Physiology Vol 37 1962 371 375 Kalt Torres Willy Phillip S Kerr Hideaki Usuda and Steven C Huber Diurnal Changes in Maize Leaf Photosynthesis Plant Physiology Vol 83 1987 283 288 l Atkin O K C Holly amp M C Ball Acclimation of snow gum Eucalyptus pauciflora leaf respiration to seasonal and diurnal variations in temperature the importance of changes in the capacity and temperature sensitivity of respiration Plant Cell and Environment Vol 23 2000 23 56 Raschke K and A Resemann The midday depression of C02 assimilation in leaves of Arbutus unedo L diurnal changes in photosynthetic capacity related to changes in temperature and humidity Planta Vol 168 1985 546 558 33M83100 Data Sheet Rochester NY Geospatial Systems Inc 2007 Multispectral Camera Info Sheet Rochester NY Geospatial Systems Inc 2007 Spectral and Polarization Configuration Guide Rochester NY Geospatial Systems Inc 2007 36MS2100 MS2150 amp MS3100 Digital Multispectral Camera User Manual Auburn CA DuncanTech Inc 1999
120. n there is a veil of tall nearly dead grass obscuring the stress signature Plants 8 and 9 seem to have been less obscured by the veil than plant 10 as indicated by Figure 79b This is shown by the lower initial NDVI values and rapid decrease in NDVI for plant 10 Even though this veil exists the system was able to detect vegetation stress somewhat better than the mown segment This can be seen in the higher R values in this un mown segment 0 7444 as compared to the mown 0 7273 Based on the location of these three plants relative to the highest CO concentration I would expect the NDVI trends to be opposite of what they are with plant 10 being the healthiest and plant 8 the least healthy So within this small segment at these higher levels of CO concentration this may suggest that the slightly higher concentrations are more of a nutrient than the slightly lower concentrations This agrees with Arp 1991 in that increased CO concentration and a sink source balance can lead to increased photosynthetic capacity For these three plants p values were calculated in comparison to un mown region 3 since it is a control vegetation area The p values show that plants 8 9 and 10 have been statistically separated from the un mown region 3 Plant 10 has the lowest p value and plant 8 has the highest because of the degree the imager was able to see through the veil of nearly dead plants into the healthier underbrush This implies that vegetation
121. names Green Red NIR minus and plus These are passed to the buffer for identification later Panel 4 grabs data from the buffer displays the images on the front panel and calculates values needed to calculate NDVI Initially the CIR image is displayed then the CIR image is split into its component color planes and they are displayed on the front panel The red color plane is subtracted from the NIR pixel by pixel to find the numerator for the NDVI calculation Eq 1 and the difference is normalized by the sum of these two images to obtain the NDVI Panel 5 finds the present time and converts and saves image color planes and NDVI data into arrays of pixel values The arrays are saved with a file name consisting of the color plane name concatenated with a time stamp so that each image s time is saved Panel 6 acquires two temperatures one near the imager and one at the back of the computer converts them to degrees Celsius and saves them to a temperature array The samples were converted to Celsius using the following equation VERE a 2 982 Vis 25 2 0 001 In this equation T ample is a 2 D representation of the voltages proportional to temperature sampled by the NI USB 6210 and Teeisius is a 2 D representation of the temperatures in Celsius The temperatures are read into a 2 D Waveform Chart vi to 40 display both temperatures on the screen in real time At this point the flat sequence structure h
122. ng To Byte Array VIs are called which convert each byte from hexadecimal to decimal Since this converts both hex numbers separately the MSB must be multiplied by 16 and then the MSB and LSB are added Now the IT in ms must be changed The maximum and minimum values for the IT are 0 1024 which correspond to 0 12 ms and 130 75 ms respectively There is a set of five conditional structures used to determine what the difference between the spectralon DN lower limit set by the user and the actual DN of the spectralon This gives six possible alterations to the IT The greater the difference the greater the change applied to the IT The spectralon DN difference ranges are gt 70 70 50 50 30 30 20 20 10 and lt 10 The possible IT changes are add 50 20 14 28 163 11 11 8 33 and 6 67 respectively For the decrease in IT the IT changes would be subtracted from the original IT The message to set the IT must be sent So the new IT is converted hexadecimal The checksum byte of the message to change the IT is calculated The message bytes are concatenated A VISA Write VI is called to set the IT using 0204 0014 02 as the first five bytes of the Write Buffer string The last three are the two LSB of the IT the two MSB of the IT and the checksum byte in that order A wait is called followed by a Bytes At Port Property Node VI The output of this VI is wired to the Byte Count input to a VISA Read VI The imager attribute buffer is
123. ng a spectralon target in every vegetation image and having more samples Table 23 Percentage change in the NDVI immediately after two hail storms for the 2008 mown segment 1st Hail Storm 2nd Hail Storm Region 1 Region 2 Region 3 Data from the USB4000 spectrometer has not been presented here because it was not conclusive When measuring reflectance very close to the vegetation it is very hard to take repeatable measurements especially with a fiber optic spectrometer with a small field of view whose width is approximately 3 8 cm at a distance of 30 5 cm Because of 121 this small field of view each time a measurement is taken the fiber sees different objects without extremely careful realignment For example one measurement might see a leaf of one type of vegetation a second measurement might see a leaf of another type of vegetation and a third reading might see part of the leaf in the first reading along with the ground This inconsistency results in very spurious data Using an imager at a distance will avoid this problem of spatial variation by averaging over a larger area 2008 Un mown Segment The 2008 CO flux map Figure 72 shows that the CO flux in the un mown segment region is high in region 2 it is above background and in region 3 it is at background This shows great agreement with the NDVI Figure 68 for the un mown segment though the stress signatures have been somewhat hidden due to the segment not
124. ng imaged to be un mown The veil of tall nearly dead vegetation will obscure the healthier spectral response of the underbrush This is because the tall un mown vegetation will die as summer heats up and dries out the environment However due to the CO sink source relationship with vegetation density higher CO2 concentrations will lead to higher levels of stress Even still it is possible to detect the effects of increased CO concentration on mown vegetation regions This can be seen as 128 CO as a nourishment for somewhat sparse vegetation as seen in the 2008 mown segment It was also found that using NDVI instead of some combination of spectral bands and NDVI will simplify and may optimize plant CO stress detection First of all it is simpler for someone with limited experience to temporally analyze NDVI than spectral band reflectances or linear combinations of bands and NDVI NDVI is simple in that the higher the value the healthier a plant so by relatively analyzing NDVI values for different vegetation regions one can easily determine if one region is stressed more than another region NDVI was shown to optimize stress difference detection as compared to the three spectral bands I employed owing this to its ability to statistically separate stressed and non stressed regions with a higher degree of accuracy in accordance with the CO flux maps made by J Lewicki 2007 and 2008 Some future work is warranted to lower the cost of
125. ns them variable names The original array gets the name color plane name _scene for example the uncorrected NIR array would be called NIR_scene The gain correction array gets the name color plane name _gain_array for example NIR_gain_array The offset correction array gets the name color plane name _offset_array for example NIR_offset_array These variables are then sent to MATLAB A new variable is created in MATLAB to place the corrected array into This is gets the name color plane name _scene_cor for example NIR_scene_cor The scene is corrected using Equation 4 which uses the NIR color plane as an example Then the corrected scene array is an output of the MATLAB script node for more calculations 159 NIR _ scene _cor NIR _scene NIR _ gain _ array NIR _ offset _ array 4 Panel 1 6 sends corrected scene pixel values to the ROI average calculators which calculate the average DN of the ROI selected in this case the exposure control ROI First of all the array is displayed on the front panel using an intensity graph with scroll bars to determine the ROI The user is able to move the scroll bars to define the ROI The positions of the scroll bars within the array are read by Subvi vi which returns the x axis and y axis indices and lengths These values are passed to an Array Subset VI which selects the ROI array out of the entire array Then the average of the ROI array is found Panel 1 7 closes the c
126. ntrol Section available only for selected jel d y B y Carral None Saw Sample framegrabbers Figure 24 DT Control Main Camera control panel DTControl Software Users Manual The gain control is used to amplify the output signal for each channel of the MS 3100 separately with a possible range of 2 0 dB to 36 dB The integration time IT is the amount of time the electronic shutter is open and can be set for each channel The IT has a range of 0 12 ms to 130 75 ms The overall exposure is used to change the brightness of an image without changing the relative brightness of each of the color planes When this is changed the individual ITs will be changed relative to each other The overall exposure is unitless and the min and max values are determined by each of the channels ITs and the difference between these times The bit depth can be set to either 10 bit or 8 bit However the frame grabber we used only allows for 8 bit or 12 bit data so we used 8 bits The display modes available for the MS 3100 are CIR RGB Mono Red Mono Green Mono Blue and Other We used CIR color infrared which has NIR Red and Green channels RGB is the typical 36 visible setup which has Red Green and Blue channels The three different Mono setups output the selected color plane to all three channels The Other selection is specific to this MS 3100 setup and allows one of eight different outputs for each of the channels Red NIR
127. of all plants alive and nearly dead in an image This agrees with Arp 1991 in that CO2 can be beneficial to vegetation except when there is a sink source imbalance In this segment there is somewhat dense vegetation so up to a point the CO is a nutrient as in regions 2 and 3 while past that point it is harmful as in region 1 These results show that for an un mown segment the difference in the levels of CO flux in these segments causes a stress on the vegetation that is detectable over a level somewhat higher than background This is evident because the mown segment detection was able to delineate between both regions and 3 where region 3 is background region 1 is a high level of increased CO and region 2 is a low level of increased CO Regions 2 and 3 and regions and 2 are not statistically different owing to similarities in the slope of the linear fits Overall taking p values and R values into account the NDVI is 113 the strongest indicator examined here for the detection of plant stress in an un mown segment 2008 Discussion 2008 mown un mown and individual plant results show good agreement with CO flux data Figure 72 Lewicki 2008 implying a decrease in plant health with high levels of CO and possible increase in health for low levels of increased CO2 The NDVI has superior accuracy when connecting these regions of CO flux to regions of possible nourishment for or stress on vegetation This is shown in p va
128. ogram multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi Figure 27 The block diagram for this program can be seen in Figure A6 of the Appendix This program runs autonomously once started It changes the IT of the imager to maintain the brightness of the spectralon calibration panel within a specific range set by the user thereby keeping the balance of incident light between the three CCDs constant This makes it possible to get good measurements in sunny or cloudy conditions since the spectralon is viewed in every image The vegetation and calibration regions are chosen with user defined ROIs The program will then calculate reflectance and NDVI for three regions and display the results in real time on a graph Finally the program saves the results to a spreadsheet 44 multiple extract planes amp calculate reflectance without temp sensor 2_with correction _test2 vi Create a folder to place image MATLAB corrects image data pixel value arrays and a for spatial non uniformity spreadsheet for reflectances Calculate average DN for vegetation scenes and spectralon using ROI Finds present time for file name identification converts and saves each color plane as an array of pixel values User supplies camera interface name image capture frequency paths to image and reflectance files and limits 3 for pecon DN a aiT Grab image data from buffer control
129. olid 2 dash and 3 dot Region 1 Region 2 08110 07 31 07 21 Date 07711 07 01 1 0 5 0 6 0 7 0 8 0 9 NDVI 0 3 0 4 Figure 68 2008 un mown segment Date versus NDVI for regions 1 green 2 red and 3 blue 107 Table 19 2008 un mown segment Date versus NDVI regression R and p values Regression p value NDVI 0 9033 lt 2 2e 16 Table 20 2008 un mown segment Date versus NDVI regression p values that distinguish between vegetation regions Regions 1 2 Regions 2 3 Regions 1 3 Intercept Term 0 538 6 32E 06 8 33E 07 Slope Term 0 721 8 36E 05 2 59E 05 Individual Plants Within Un mown Segment Within the un mown segment there were multiple plants that were individually numbered Among these plants 8 9 and 10 had exhibited evident stress in measurements made by other sensors The independent evidence of stress in these plants provided an opportunity to compare data from the tower mounted imager with data from a ground based sensor used to measure the reflectance spectrum of individual plants This comparison was accomplished by processing the MS 3100 imager data in small clusters of pixels near the location of these particular plants For these plants it can be seen in Figure 69 that the NIR reflectance for plants 8 and 10 hardly change during the experiment and for plant 9 it has a small negative slope The red reflectances for all plants are nearly the same but for plant
130. on p values that distinguish between individual plants and the un mown region 3 seessssssesrsesesresseeererresss 109 Percentage change in the NDVI immediately after two hail storms for the ZOOS MMO WI SCRIMICNL toalla 121 A1 Values used to calculate reflectance for each of the color planes during the 2007 experiment with the photographic grey Card ooooocnnococonocccconnnononcnonanannnnnos 151 A2 Serial settings needed to communicate with the MS3100 imager 163 A3 Message format to query or set the MS3100 integration time eee 164 Figure 10 11 12 13 xi LIST OF FIGURES Page Earth surface atmosphere solar radiation absorption and emission The yellow orange lines on the left indicate that most of the sun light is absorbed by the Earth s surface and atmosphere The red orange lines indicate the amount of thermal radiation emitted by the Earth s surface and atmosphere Image adapted from Kiel and Trenberth 1997 by Debbi McLean Remer O cane Wades So ccaln as Scan lascanteed us cae 2 Plots of the increase in carbon dioxide concentration and temperature NASA graphs by Robert Simmon based on carbon dioxide data from Dr Pieter Tans NOAA ESRL and temperature data from NASA Goddard Institute for Space Studies Remer 2007 cescecesccecsseceeeseecesseeceeseeeeseeeeaees 3 Plot of the decrease in volume of all Earth s glaciers Glacier graph adapted from Dyurgerov and Meier 2005 Remer 20
131. onverter A D resolution is 16 bits A fiber optic cable provides an easily hand held input for the device Data from the spectrometer are collected via a USB cable with a laptop computer For the measurements in this project the data were averaged over the MS 3100 bands for comparison with the multispectral imager data 32 ESTI Figure 22 USB4000 Miniature Fiber Optic Spectrometer and Spectralon disk that were used together to measure reflectance spectra of vegetation and calibration panels www oceanoptics com Figure 23 USB4000 Miniature Fiber Optic Spectrometer Optical Layout USB4000 Installation and Operation Manual www oceanoptics com 33 Table 1 USB4000 miniature fiber optic spectrometer optical layout explanation USB4000 Installation and Operation Manual www oceanoptics com Item Number Name 1 SMA 905 Connector Slit Filter Collimating Mirror Grating Focusing Mirror L4 Detector Collection Lens Detector OFLV Filters o 0 N O oa Pp W PD Imaging Software I used multiple programs to fully implement the system and the calculations needed to find reflectances Various aspects of the task were implemented with NI IMAQ DT Control NI Measurement and Automation Explorer MAX NI LabVIEW NI Vision Acquisition NI Vision Development Module NI DAQ Ocean Optics SpectraSuite PeModwin 4 0 MODTRAN MATLAB and solar position calculator Table 2 shows what routines are used by what programs the purpose of
132. ot detecting stress The red band was able to distinguish between regions 1 and 2 and regions 2 and 3 which matches the NDVI results The NIR band was able to distinguish between regions 1 and 2 and regions 1 and 3 again showing what may be an angle effect No combination of bands was able to explain variability as well as the NDVI The agreement between the NDVI 120 and the red band reflectance suggests that the green and NIR bands are not able to accurately detect stress in a mown segment Overall taking p values and R values into account the NDVI is the strongest of the indicators examined here for detecting plant stress in a mown segment In this segment the NDVI was also able to detect the effects of rain and hail Within a day after each rain storm July 16 July 17 August 3 August 4 and August 9 the NDVI value increases indicating the system is able to detect this nourishment The two days of hail decreased the NDVI values indicated in Table 23 but then as the moisture from these storms was taken up by the vegetation the NDVI value jumped right back up to a level consistent with the linear fit The amounts of precipitation for these two hail storms were incredibly higher than any of the other rain storms The largest rain storm delivered 0 12 in and the smallest hail storm delivered 0 95 in of precipitation It seems that this ability to detect precipitation effects comes about from the strong calibration technique based on imagi
133. ow a query is performed on the file name to determine which color plane it is so that the correct algorithm can be applied First the word grey is stripped off the end since this procedure to calculate reflectance only uses vegetation images with the grey card in the field of view This is done using a Match True False String V1 which takes the file name string format a true string and a false string The true and false strings are defined by the programmer and I only defined the true string as grey If either the true or false string matches any part of the file name string The matching selection and every character after the selection will be removed from the original string and returned in the output This leaves the file name with only the time and color plane designator Next the file name sting is reversed so the file name would now appear as follows RIMP 234521 Another Match True False String VI is used except with the true string RI This VI can also be configured to output a logical true or false If the true string matches any part of the file name it will return a true This output is connected to a conditional case structure which has a true and a false case If it is true then the program will continue to the actual calculation of NIR reflectance If it is false then it will check with another Match True False String VI if the file name indicates the image is of the red color plane R If this is true the program will continue
134. ow be computed To do this Eq 7 is solved for reflectance E G t Q sun A int egration pixel 1 5 de DN m Here Esun is found in the same fashion described above using the MODTRAN fit To model the irradiance of the sun Esun I used the MODTRAN version of PC Modwin I was then able to calculate reflectances using the original images taken at the ZERT site but not the grey card images To do this I created another LabVIEW program calculate reflection scaffolding 2 vi implementing Equation 15 This program calculates reflectance and NDVI values using the images and lookup tables as inputs The images are placed in a folder and that folder is specified on 88 the front panel of the LabVIEW program Then a new folder is created in that folder called refl in which the lookup tables are placed Four lookup tables are needed time txt int time txt irradiance txt and refl txt The time txt file is a 1 d array of all the times images were taken in one day with the header time The times are in the following format 92227 which would be 9 22 27am and 20117 which would be 2 01 17 pm The times are listed in chronological order The int time txt file is a 2 D array with four columns The first column is the same as the 1 D array time txt the second column is comprised of green channel ITs for each time listed in the first column the third column is red channel ITs and the fourth column is NIR channel IT
135. pe We placed the spectralon calibration panel on a mount set at a 45 elevation angle and about 3 m away from the scaffolding so that the imager would look nearly normal to 78 In CO2 Mown Strip Un mown A lek Pipe Center FOV Figure 49 Imager orientation at the ZERT Site during the 2008 CO release experiment the calibration target Again the imager was oriented to point in a north north west direction and was recessed in a protective box The images were calibrated by imaging the 99 reflective spectralon panel as part of every image This made it possible to calibrate each image during the experiment or to calibrate each image in post processing with higher accuracy An exposure control through IT adjustment was added to the imaging program to make a more autonomous and stable system The program checked the DN of the Spectralon panel and 1f 1t was within a range set by the user it would record an image if the DN of the Spectralon panel was not in the specified range it would change the IT and check again until all three color plane ITs were correctly set This exposure control gives the imaging system larger working range in that it would automatically change the IT so that images could be acquired no matter the sky conditions and keeps the DNs for the entire scene within 79 the linear response of the imager This along with the fact that the spectralon calibration target has a very stable reflectance at any
136. pidly than a cosine Figure 41 A cosine would be expected if the meter only viewed the spectralon but since it views the dark background in addition to the rotating spectralon the detected power falls off a little more quickly than a pure cosine i e there is a combination of cosine falloff of illumination and falloff from an increasingly large fraction of the detector s FOV being filled with dark background 70 Normalized Optical Power Meter Measurements as a Function of Spectralon Angle E Pp o 4 4 0 9 0 8 0 7 0 6 0 5 Normalized Power 0 4 0 3 0 2 Power Meter Cosine Function 0 1 80 60 40 20 0 20 40 60 80 Theta degrees Figure 41 Normalized power measured by an optical power meter and cosine of the spectralon rotation angle Next I rotated the spectrometer s fiber optic cable around the spectralon Figure 38 The spectrometer fiber was placed very close to the spectralon to ensure the small FOV was totally filled by the spectralon even at extreme angles The measurements were integrated over a spectral range of 590 670 nm and then normalized to 1 I measured the power from 70 to 70 in 10 increments Since the spectrometer FOV is filled over the same area as the spectralon is rotated the normalized power is basically constant Figure 42 The small amount of fall off about 10 seen here is due to non uniform illumination of the spectralon the center having a higher irradiance
137. r plane image viewing an evenly illuminated scene 120 110 100 30 Pixel Value 80 70 y 60 1500 ee s i 7 i ee a a Ton 1200 a eet n me Er 800 600 200 400 Y axis Pixel Index X axis Pixel Index Figure 37 3 D view of corrected red color plane image viewing an evenly illuminated scene 65 Even after the spatial uniformity correction the problem has not been corrected even though it has been improved However finally it was found that the spectralon reflectance changes with the illumination angle Labsphere Inc 2006 Consequently as the illumination angle increases there will be an increasing error in reflectance Initially I had kept the solar illumination angle somewhat high near 60 and tried to view the panels near normal incidence to ensure that the imager did not view a specular reflection As seen in Table 8 this geometry was causing an error in measured reflectance especially in the 50 panel assuming a 60 panel is similar to a 50 panel Table 8 Change in reflectance of spectralon due to illumination angle measured from the surface normal for 99 and 60 panels Labsphere Inc 2006 Sample SRS 99 99 Sample SRS 60 60 Wavelength AR 45 AR 61 Wavelength AR 45 AR 61 nm nm 300 0 009 0 010 300 0 026 0 064 600 0 006 0 005 600 0 021 0 050 900 0 001 900 0 033 0 051 1200 0 003 1200 0 027 0 042 1500 0 002 1500 0 024 0 042 1800 0 004 0 000 1800 0 020
138. rchers have been trying to model biophysical parameters such as leaf area index total biomass and gross CO flux estimation de Jesus et al 2001 Nakaji et al 2007 This thesis describes experimental research conducted to assess the potential utility of a platform based multispectral imager for detecting leaking CO through plant stress measurements Many remote sensing techniques for detecting plants or plant stress exploit a spectral signature called the red edge The red edge is the steep rise in vegetation reflectance that occurs near 700 nm with an inflection point connecting the low red reflectance and high near infrared NIR reflectance As vegetation is stressed the red edge shifts to shorter wavelengths and becomes less steep Figure 9 In Figure 9 the gold line is a reflectance spectrum measured for a healthy plant the blue line is the reflectance spectrum of an unhealthy plant and the gray line is the spectrum of a dead plant 12 Inflection Point so eo 700 800 900 Wavelenath nm Figure 9 Spectrum of a healthy gold unhealthy blue and dead grey plants Spectrum acquired with a USB4000 spectrometer made by Ocean Optics Inc Carter Responses 1993 completed a study of the reflectance spectrum of vegetation of different species to different stresses He hoped to define spectral signatures of specific stresses that could be applied to any vegetation and to define what regions of the spe
139. re DAR Light Port Spectralon Figure 39 Top view for spectralon test setup where light measuring device is fixed and the spectralon is rotated around its vertical axis The first test I did was to rotate the power meter around the spectralon Figure 38 I placed the power meter very close to the spectralon so that the field of view barely lapped over the edges of the spectralon I measured the power from 80 to 80 in 10 increments As expected for a Lambertian panel the power incident on the detector decreases as the angle increases away from normal Figure 40 The detected power falls off in a very nearly cosine fashion following the projected area of the panel 69 Normalized Optical Power Meter Measurements as a Function of Viewing Angle gr Ya 0 9 0 8 0 7 0 6 0 5 Normalized Power 0 4 0 3 0 2 Power Meter Cosine Function 0 1 80 60 40 20 0 20 40 60 80 Theta degrees Figure 40 Normalized power measured by an optical power meter who s FOV was slightly larger than the Spectral panel and cosine of the viewing angle Next I kept the optical power meter detector stationary while rotating the spectralon around its vertical axis Figure 39 Again the detector was placed close to the spectralon and the FOV extended just beyond the edges of the panel This time I was only able to measure angles from 70 to 70 As expected the power falls off with angle but does so somewhat more ra
140. read and returned A VISA Close VI is called to close the VISA serial session A false is returned after the IT has been changed so that the program knows it will need to check the spectralon DN at least one more time before imaging Once the NIR IT check has finished the red Panel 1 9 3 and then the green Panel 1 9 5 color planes are checked After all color planes have been checked both interior flat sequence structures are exited and the logical statements indicating whether or not each color plane s IT are correctly set are checked to see if all three are correct If not the program loops back to panel 1 1 if so the program moves a head to panel 2 to begin the vegetation imaging Panel 2 of the main flat sequence initializes an image capture session and configures a buffer list to place the images for processing This was done using the IMAO init and IMAQ configure list VIs IMAQ init requires the Interface Name as specified by MAX which is img0 IMAO init then returns the IMAQ Session Out an ID for all subsequent IMAQ processes The Session Out is passed to IMAQ configure list to create a buffer To simplify this section all IMAQ VIs will be passed the Session Out 164 ID A property node set to type IMAQ is triggered to return the Image Type As mentioned before we are running the imager in 8 bit mode so the Image Type is Grayscale U8 This information is passed to the next panel of the flat sequence The Panel 3 conf
141. rial port settings are set to match the imager s serial settings shown in Table 4 Next the panel reads the NIR Table 4 Serial settings needed to communicate with the MS 3100 imager Input Setting Enable Termination Character Termination Character Timeout ms Baud Rate Data Bits Parity Stop Bits Flow Control ROI average computed in Panel 1 6 and compares it to the NIR upper limit provided by the user The program then uses a conditional structure which depending on if the DN is greater than or less than the upper limit will start the process to lower the IT or check the lower limit respectively If the DN is less than the upper limit the spectralon average DN is compared to the lower limit and another conditional structure is used which depending on if the DN is greater than or less than the lower limit will either exit without changing the current IT with a true state indicating the IT is set correctly or it will start the process to increase the IT Since the processes to increase or decrease the IT are the same I will explain just the method to increase the IT 48 Initially the imager is queried and current attributes are read When the imager is queried and the imager buffer is read 6 Bytes are written and 9 bytes are read If the imager is sent a command to change its attributes 8 Bytes are written and 6 Bytes are read The messages are in hexadecimal and if querying
142. rument for use in the field at multiple sites within a carbon sequestration site At this point the imager and a small computer and touch screen monitor have been placed in a somewhat large box on top of scaffolding Instead an imager could be built on a small optical bread board and integrated alongside a microcontroller that employs wireless technology for communication with a monitoring center This could then be placed upon a pole in an orientation that would view a spectralon panel and vegetation that may be in a CO leak zone The spectralon panel would be placed in a protective box that automatically opens right before an image and closes after the image The system would need no at site attention other than the initial setup and could run continuously throughout the day Finally an automated stress check could be applied The system could check multiple regions for drastic changes in NDVI in comparison to other regions If there is a change an alarm could be sent to the monitoring site where at this point a more direct and expensive CO2 measurement system needing human interaction could be employed This would possibly be a more cost effective multiple deployment option to a more direct CO measurement system 130 BIBLIOGRAPHY Spangler Lee ZERT Zero Emissions Research Technology 2005 Montana State University 2 Sept 2008 lt http www montana edu zert home php gt MS 3100 Multispectral 3CCD Color CIR Camera 2007
143. s Inc 2003 Jordan Carl F Derivation of Leaf Area Index from Quality of Light on the Forest Floor Ecology Vol 50 1969 663 666 Robinson Andrew P Amy L Pocewicz and Paul E Gessler A Cautionary Note On Scaling Variables That Appear Only In Products In Ordinary Least Squares Forest Biometry Modeling and Information Sciences Vol 1 2004 83 90 134 APPENDIX A IN DEPTH DISCUSSION OF LabVIEW PROGRAMS 135 2007 Experiment For the 2007 experiment the programs multiple extract planes vi and calculate reflection scaffolding_grey_multiple_scenes 2 vi were used The first program takes images of the vegetation test strip saves the temperature during an image acquisition and saves pixel value arrays of each of the color planes and the calculated NDVI The second program was used to calculate reflection and NDVI using only images of the vegetation with the photographic grey card calibration target in the image This program uses regions of interest ROIs to select the grey card and three separately analyzed vegetation regions within both the mown and un mown portions of the test area The second program provided only a few calibrated images each day but was used after two other attempted calibration methods did not work sufficiently well These initial calibration methods were based on 1 using gray card images recorded whenever the IT changed and 2 using continuous measurements of solar irradiance at t
144. s The first row is a header for each of the columns The irradiance txt file is set up in the same fashion as the int time txt file except the cells are filled with the MODTRAN modeled irradiances The refl txt file is set up in the same fashion as the int time txt file except the cells are filled with the reflectance values found by the program This file must have numbers in the cells before the program is run or the file will not be written This procedure was very promising and should definitely work in most situations but the problem was that we did not have a pyranometer measuring solar irradiance at a location near the ZERT test site Because there was not a local station with a working pyranometer I tried using data from a pyranometer in Dillon MT Dillon is about 115 miles from Bozeman which certainly gave erroneous irradiance data when clouds were present 2008 Procedure Using Spectralon Panels Considering how well the reflectances calculated using the few vegetation images that also included the grey card worked out I decided a system using a highly diffuse 89 calibration target that could be imaged in every image would the best route to take So I placed a 99 reflective Spectralon panel in the FOV to be imaged along with the vegetation This made it possible for us to obtain accurate reflectance data using Equation 2 However as the experiment progressed we found that we were obtaining data with another calibration is
145. s 8 and 10 the red reflectances increase a little slower than for plant 9 The green reflectance s slopes are nearly the same though plant 8 has the shallowest slope and plant 10 has the steepest slope The date versus NDVI regressions shown in Figure 70 agree with the reflectances in that the NDVI slopes for each plant are nearly the same Since all three plants were located in nearly the same place each plant was compared to the 2008 un mown segment region 3 which serves as a control The 108 regression coefficient of determination 0 7444 Table 21 was able to explain the variability well The regression was significant with a p value of lt 2 2e 16 Table 21 The p values for both the intercept and slope regression coefficients Table 22 show that all plants are statistically separable from the 2008 un mown region 3 50 45 40 Reflectance Plant 8 Plant 9 Plant 10 07 11 07116 07 21 07 26 07 31 08 05 Date Figure 69 Green red and NIR reflectances for individual plants within un mown segment 109 Plant 8 o Plant 9 Plant 10 08 10 07 31 Q 2 i A 1 07 11 O 07 01 ER aa f f A i 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 NDVI Figure 70 Date versus NDVI for individual plants within 2008 un mown segment Table 21 Individual plants Date versus NDVI regression R and p values Regression p value NDVI 0 7444 lt 2 2e 16 Table 22 Individ
146. s as compared to the NDVI Although the NDVI is not best suited to explain variability in reflectance spectra it is strong when it comes to detecting statistical differences in stressed vegetation regions After exploring various combinations of band reflectances and NDVI I decided to use linear regressions only involving NDVI even though according to Robinson 2004 a regression involving a spectral band interaction term such as NDVI which involves both NIR and red reflectances without the individual bands is not statistically ideal I did this because when the bands were included the ability of the regression to statistically 97 separate between regions and the significance of each of the bands and NDVI towards the regression were diminished So for the application of multispectral imaging used to detect differences in vegetation health in multiple spatial regions NDVI was the strongest predictor Analysis of the measurements from the 2007 and 2008 field experiments also indicates that a hardy calibration technique may increase the accuracy of a plant stress detection system enough that the effects of a small increase in CO concentration rain and hail are detectable even in cloudy conditions 2007 Experimental Results The 2007 experiment resulted in only a limited amount of good data because of the non lambertian photographic grey card and poor reliability of the resulting calibrations discussed in more detail in Chapter 4 S
147. s stated by Anderson et al 2005 Information and experience gained from the injection and or storage of CO2 from a large number of existing enhanced oil recovery EOR and acid gas projects as well as from the Sleipner Weyburn and In Salah projects indicate that it is feasible to store CO in geological formations as a CO mitigation option It is believed that sequestration at a carefully chosen site one with the needed deep geological features would be able to retain up to 99 or more of the injected CO for at least 1 000 years The geological trapping features Figure 6 are as follows Anderson et al 2005 e Trapping below an impermeable confining layer caprock e Retention as an immobile phase trapped in the pore spaces of the storage formation dissolution in the in situ formation fluids e Adsorption onto organic matter in coal and shale e Trapped by reacting with the minerals in the storage formation and caprock to produce carbonate minerals a Geological Storage Options for CO oO co of or gan 10 cero co 5 Use of CO in enhanced cosi bed methane recovery 6 Other suggested cptens tasata of haies cavities Figure 6 Basic block diagram of carbon dioxide capturing systems Anderson et al 2005 Though the need to monitor the sequestration sites for carbon dioxide leaks arises mainly as a carbon control issue the safety of people and local flora and fauna is also a concern There have been nat
148. s surface With all of this detailed image data available there arose a need to quantitatively analyze the data Since the 1970s significant work has been done to more accurately analyze vegetation imagery Initially researchers with years of training would analyze vegetation imagery by viewing the images band by band or multiband This worked well but was not quantitative The next step was to use calibrated reflectances the percentage of incident sunlight reflected by objects With these data researchers began to see that objects have spectral signatures and more specifically that the spectra of healthy and non healthy vegetation of the same type were very different So by analyzing the relative levels of multiple spectral bands researchers were able to glean information on vegetation spectral signatures Knowing what portions of the vegetation reflectance spectra are most affected by stress led to the combination of multiple spectral bands into what are called vegetation indices Jensen 2000 Many at least 30 vegetation indices have been used over the years Each of these indices was created to examine different 11 characteristics of plant health by analyzing different parts of the reflectance spectrum Some look at overall plant health some look at water content some look at chlorophyll content and there are many others Realizing that each of these indices measures somewhat specific characteristics of plant health resea
149. source balance A wide field of view sensor such as the imager used in this study tends to reduce problems with spatial variations that induce randomness into measurements taken with narrow field sensors such as the fiber optic spectrometer used here Using strong calibration techniques such as viewing a spectralon calibration target in every image with an imaging system in a long term deployment allows detection of small changes such as the effects of rain and hail on the vegetation Having rain and soil moisture data was very helpful in determining the affect of CO on vegetation This data makes it possible to separate what may be CO stress or 125 lack of water related stress It also helps to show the importance of sink source balance and vegetation density when determining the affects of CO on vegetation For 2008 it was particularly helpful because according to soil moisture data the vegetation should have gotten healthier and it did for the mown region but at different rates The different rates can be directly related to CO concentrations The un mown segment had to compete for water so the CO levels present were detrimental 126 CONCLUSIONS AND FUTURE WORK It has been shown that a tower mounted multispectral imager viewing a vegetation scene and a reflective calibration target was able to detect plant stress temporally in response to a CO leak which was modeled to approximate a CO2 sequestration site leak at the ZERT
150. specified bands Labsphere Inc 2006 Labsphere specified reflectance Reflectance panel Green 520 560 nm Red 650 690 nm NIR 768 833 nm 99 0 9901 0 9894 0 9896 50 0 4859 0 5093 0 5304 I also characterized the angular variation of the Spectralon reflectance using the USB4000 spectrometer and an optical power meter I used an integrating sphere as the illumination source for the spectralon Four tests were run two with the power meter and two with the spectralon The power meter has a wide FOV meaning the area measured extended past the edges of the spectralon panel The spectrometer has a narrow FOV that was fully contained near the center of the Spectralon panel For each instrument I did the 67 following tests 1 keeping the spectralon still while rotating the light measuring instrument at a fixed radius around the spectralon panel Figure 38 and 2 keeping the light measuring instrument still while rotating the spectralon around its vertical axis Figure 39 The light measuring instruments had to be set below the spectralon and angled upward at about 40 from the Spectralon surface normal so that they would not block light from the integrating sphere RS 0 Integrating Sphere S e Light Port Spectralon y Figure 38 Top view of spectralon test setup where light measuring device is rotated around a fixed spectralon panel 68 Light Measuring Device Integrating Sphe
151. ssed we expect to see the NIR band reflectance decrease and the green and red band reflectances increase leading to a flatter total spectrum This leads to a decrease in both the red NIR reflectance difference and the normally steep slope of the red edge corresponding to a blue shift of the inflection point Carter 2001 between the red and NIR bands The Normalized Difference Vegetation Index NDVI Eq 1 takes advantage of this difference between the red and NIR bands We expect changes in the spectrum of the vegetation to come about from changes in long term environmental factors such as the soil atmospheric CO levels analogous to work done by Noomen 2006 seasonal water levels and seasonal heat It is possible for excess CO to be helpful or harmful depending on the CO sink source balance and the density of the vegetation Arp 1991 showed that even though CO stimulates photosynthesis long term high levels of CO2 could cause photosynthetic capacity to decrease when there is a source sink imbalance and dense plant growth This in turn will lead to a decrease in chlorophyll content Arp 1991 Conversely Kimball et al 1993 states when CO levels are doubled plant growth and yield increase by 30 This all suggests that depending on how close the plants are to the leak and the level of CO that they are exposed to there may be some plants that feel negative stress and some that are nourished Of course if it is hot and dry plants will
152. sue because of the differences in sun Spectralon angle and sun scene angle Figure 57 This came about because the spectralon was set on a mount at 45 while the scene or vegetation was nominally at 0 assuming all parts of the vegetation are in the same direction Figure 58 Reflectance Ea o q Oo Date Figure 57 Erroneous reflectance data due to differences in sun spectralon angle and sun scene angle 90 spectralon scene Figure 58 Differences in sun Spectralon angle and sun scene angle that gave rise to much of the curvature of the reflectance plots in Figure 57 However we were able to step around this in two ways 1 by picking the same solar time every day as a quick fix and 2 by coming up with a calibration to correct for the differing solar irradiance throughout the day as a total fix By the end of the experiment we found that by laying the spectralon panel flat we were able to obtain very accurate data throughout the day Figure 59 and Figure 60 wae 44 tor tte te se ttre AE PET TT a o tor CATA re st Pe wo ww E h oO 0 oO Mm M mn Reflectance A tte o tor ett teeee A AAA E tHe OO aye tt ee tyt 4g o 20 15 Hot eee te pi TOO OOo EE rro PESEE gt PAE t teteg tetet ogee toep ay 404 Date Figure 59 Accurate reflectance data taken with Spectralon calibration panel laid flat to remove effect shown in Figure 57
153. tance without temp sensor 2_with correction _test2 vi the image acquisition reflectance and NDVI calculation program used in 2008 To run the program a folder must be created to save the images and a reflectance spreadsheet The reflectance spreadsheet must also be created with a 45 layout of 19 cells wide and about 75 cells long The length depends on how long imaging will take place and the image capture frequency The cells should be filled initially with zeros The first column is used to hold the time of each reflectance measurement The second third fourth and fifth columns hold the Green Red NIR reflectances and NDVI respectively for the first region The seventh eighth ninth and tenth columns hold the Green Red NIR reflectances and NDVI respectively for the second region The twelfth thirteenth fourteenth and fifteenth columns hold the Green Red NIR reflectances and NDVI respectively for the third region The seventeenth eighteenth and nineteenth columns hold the ITs for the Green Red and NIR color planes respectively for that specific image The user must also supply some inputs camera interface name img 0 image path directory to save images to reflectance path image capture frequency how often the imager should image and the upper and lower limits for the average DN of the spectralon desired for each color plane determines how sensitive the IT control will be The program starts by c
154. ted photosynthetically active radiation in numerous vegetation types irrespective of sky conditions leading him to construct a linear regression equation to calculate absorbed photosynthetically active radiation with a root mean square error RMSE of less than 10 Fuentes et al 2007 did an experiment measuring the CO flux via eddy covariance towers and compared the results to NDVI trends calculated for that area They found that NDVI had a high correlation 0 981 with carbon flux Fuentes et al 2007 It was determined that NDVI was able to capture the effects of changing environmental conditions such as drought recovery and then fire on the carbon flux Fuentes et al 2007 According to this it is reasonable to believe that NDVI could see the effects of rain hail and small amounts of carbon dioxide on vegetation Maynard et al 2006 did a study on the ability of indices as compared to non transformed bands to accurately model biophysical parameters She compared linear regression models that estimate TTB total transformed biomass using NDVI and non transformed bands bands 4 and 7 of Landsat as the predictors The regression models were built using extra sums of squares F tests Lawrence et al 1998 and R values were used to determine the variability in biomass Maynard et al 2006 NDVI explained 41 of the variability while the non transformed bands explained 53 of the variability Maynard et al 2006 I am not aware of any
155. ted across the specified bands Labsphere Inc 2006 ceeseeceeseeceeeeceeeeeceeeeecseeeenteeeenaeeees 66 Reflectance of supposedly 18 grey card for each spectral band imaged by the WES 3 LOG secede sain a a a A A E TA tosses ogee A 81 Gain factor for specific ITs and gain factors as a function of ITs for each channel ofthe MISSION da 87 2007 mown segment Date versus NDVI regression R and p values 98 2007 mown segment Date versus NDVI regression p values that distinguish between VESES a 98 2007 un mown segment Date versus NDVI regression R and p values 102 Table 16 17 18 19 20 Pak 22 23 X LIST OF TABLES CONTINUED Page 2007 un mown segment Date versus NDVI regression p values that distinguish between vegetation regions seseseseeeseirerierersrtssrseresressreeresressesees 102 2008 mown segment Date versus NDVI regression R and p values 105 2008 mown segment Date versus NDVI regression p values that distinguish between VERE LATION Te SIONS is 105 2008 un mown segment Date versus NDVI regression R and p values 107 2008 un mown segment Date versus NDVI regression p values that distinguish between vegetation regions oooooocnnocccnoncncnoncnononcnononanononcconancccnnnacnnns 107 Individual plants Date versus NDVI regression R and p values 109 Individual plants Date versus NDVI regressi
156. the pipe The CO flow rate was 0 1 tons day and 0 3 tons day for 2007 and 2008 respectively These rates were chosen because they cover the maximum allowable leakage At this site many different carbon detection experiments were carried out some that directly measured CO in the soil ground water and atmosphere and some that indirectly measure CO through effects such as plant stress One of these techniques being explored as a potential mechanism to detect a CO leak is to measure the spectral reflectance of plants in the field with multispectral imagery to determine if they are stressed Figure 8 Arial View of ZERT Site Dobeck 2008 10 Researchers have used remote sensing as a tool for detecting plants and plant stress for a number of years Remote plant detection took a big step towards plant stress detection when Color IR film was invented during WWII Paine et al 2003 The US Army was not actually trying to detect plants they were trying to detect tanks people and things of that sort that were hidden in vegetation After the war people realized that this film format might be useful for vegetation detection Then in 1972 the first Landsat satellite implemented with a multispectral imager was put into orbit explicitly to monitor Earth s landmasses Rocchio 2008 Since then many more multispectral imagers and some hyperspectral imagers have been sent into space flown on airplanes and set up on towers to analyze the Earth
157. tial calibration target used in the 2007 experiment was a photographic grey card which was to serve as an approximately Lambertian reflector of known reflectance for calibration with Equation 2 Grey cards are designed to diffusely reflect 18 of visible light and are used by photographers routinely in a similar way that I used them to balance color and lighting ratios To get the best possible calibration I decided to use the USB4000 spectrometer to measure the actual band averaged reflectance for each spectral band of the MS 3100 especially since grey cards have been designed for the visible portion of the spectrum and their NIR reflectance was not known I found that the two visible bands were somewhat close to 18 but the NIR band was significantly high 81 Table 11 I then used these reflectance values in Equation 2 for the calculation of reflectance from images measured in the 2007 experiment Table 11 Reflectance of supposedly 18 grey card for each spectral band imaged by the MS 3100 Spectral Band Reflectance Green 17 37 Red 18 17 NIR 22 83 I initially thought by imaging the grey card every time I changed the IT on the imager I would be able to calibrate all images taken after the calibration image and before the next IT change Therefore I imaged throughout the experiment saved all the images and calculated reflectances and NDVI after the experiment This did not work entirely satisfactorily for two main
158. tion of gain A range of gains 2 12dB or 1 585 15 85 on a linear scale is shown here ma linear Seale AAA OS edn O ain eek 56 A plot of average DN as a function of 1 F H ccccessesssessessessesssessessesssessesseesseees 57 Average DN as measured by the imaging system versus current measured by the integrating sphere s detector sssssessssesssseessressessseesseeessseesseesseesseesseee 58 A plot of the affect of temperature on the imager s response 59 A plot of the affect of polarization angle on the imager s response 0 5 60 3 D view of uncorrected red color plane image viewing an evenly liada toasnsadevaseudives an a R a a NA 64 Figure 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 xiv LIST OF FIGURES CONTINUED Page 3 D view of corrected red color plane image viewing an evenly illuminated Ci ice 64 Top view for spectralon test setup where light measuring device is rotated around the spectral OM crisisen sinisini rrie E RER 67 Top view for spectralon test setup where light measuring device is fixed and the spectralon is rotated around it s vertical axis ooooonnoccnoncnnocnconcnnannconncnnns 68 Normalized power measured by an optical power meter and cosine as a f unction of viewing A A E R AE E EEE EEES 69 Normalized power measured by an optical power meter and cosine as a function of the spectralon Mt 70 Normalized po
159. tistically separable 101 50 Reflectance Region 1 Region 2 07 16 07 21 07 26 07 31 Date Figure 63 2007 un mown segment green red and NIR reflectances for regions 1 solid 2 dash and 3 dot po z sl Region 1 o Region 2 Region 3 07 25 Q 2 a 07 9 07 3 l 1 1 1 03 0 35 0 4 0 45 0 5 0 55 06 0 65 0 7 NDVI Figure 64 2007 un mown segment Date versus NDVI for regions 1 green 2 red and 3 blue 102 Table 15 2007 un mown segment Date versus NDVI regression R and p values Regression p value NDVI 0 7256 1 27E 08 Table 16 2007 un mown segment Date versus NDVI regression p values that distinguish between vegetation regions p Value Regions 1 2 Regions 2 3 Regions 1 3 Intercept Term Slope Term 0 1017 0 0986 0 2908 0 3432 0 0157 0 0187 2008 Experimental Results In the 2008 experiment better calibration techniques as discussed in Chapter 4 Section 2 led to good data being collected every day the system was operated correctly All linear regressions are statistically significant and have high coefficients of determination NDVI data obtained from image regions that should be separable as discussed in Section 3 of this chapter did turn out to be statistically separable p value lt 0 05 with high coefficients of determination R 2008 Mown Segment Data from the mown segment in 2008 show that
160. to the calculation of red reflectance If false it will continue to the calculation of green reflectance Once inside a conditional structure for the calculation of reflectance the file of pixel array values is opened using a Read from Spreadsheet File VI These pixel values are then sent to ROI average calculators which calculate the average DN of the ROI selected First of all the array is displayed on the front panel using an intensity graph with scroll bars to determine the ROI The user is able to move the scroll bars to define 149 the ROI The positions of the scroll bars within the array are read by Subvi vi which returns the x axis and y axis indices and lengths These values are passed to an Array Subset VI which selects the ROI array out of the entire array Then the average of the ROI array is found The ROI average calculator will loop until the user presses the stop button insuring the best possible region has been chosen Next the reflectance is calculated using Equation 3 with values specific to each P DN scene E DN dark s ibrafion ta 3 DN DN jar Pa calibration target A calibration target color plane s spectral band dark current and grey card reflectance These values are seen in Table Al Finally the reflectance is saved the reflectance spreadsheet using a Write to Spreadsheet File VI The row index is controlled by the iteration index and the column is controlled by the color plane of the image an
161. ual plants Date versus NDVI regression p values that distinguish between individual plants and the un mown region 3 p Value specific plant compared to 2008 un mown region 3 Plant 8 Plant 9 Plant 10 Intercept Term 0 00651 9 68E 05 7 45E 06 Slope Term 0 004116 0 000353 0 00125 2007 Discussion 2007 mown and un mown data results show good agreement with CO flux data Figure 71 Lewicki 2007 implying that CO is a stress on or is a nutrient for plant 110 health with increased CO levels depending on sink source balance Increased flux levels higher than background in specific vegetation regions shows up as increased stress on the vegetation The NDVI is the best suited method of connecting the regions of higher CO flux to regions of higher vegetation stress compared with any other spectral band This is shown in p values and R values Table 14 and Table 16 Pipe Center Figure 71 2007 CO flux map of the ZERT CO2 Detection site adapted from J Lewicki Lawrence Berkeley National Laboratory 2007 2007 Mown Segment The 2007 CO flux map Figure 71 shows that the CO flux is high in the mown region 1 above background in region 2 and at background levels in region 3 This shows great agreement with the NDVI results Figure 62 for the mown section Also the NDVI data for regions 1 and 3 and regions 2 and 3 are statistically separable according to Table 14 while regions 1 and 2 are almost distin
162. updates the X axis and data points of the real time graphs displaying all reflectances and NDVI data points on one of four graphs Each region s reflectances are shown on separate graphs and the NDVI values for each region are shown on a graph Panel 11 finishes the current image acquisition process At this point the system will loop back to Panel 1 of the flat sequence structure after the time corresponding to the image capture frequency has passed When the system is shut down for the day after all the images have been acquired the program will exit the imaging loop and close the program For image post processing I wrote a program calculate reflection scaffolding_grey_multiple_scenes 3 which loads the vegetation images and calculates reflectances and NDVI using ROIs This program is the same as calculate reflection 51 scaffolding_grey_multiple_scenes 2 except a more efficient code to read images and to write the reflectance NDVI spreadsheet was adopted The 2007 or 2008 reflectance and NDVI files can be viewed graphically using the MATLAB codes create_refl_plot_ segment or create_refl_plot _ segment _min The first program plots every reflectance and NDVI measurement for every day while the second program plots only the minimum reflectance and NDVI at times corresponding to the maximum solar angle for each day The value segment must be either mown un mown or plants8910 The only difference between these is that the
163. ural carbon leaks that have been studied to determine what might happen if a man made sequestration site leaked Even though these sites both natural and man made are able to almost completely sequester the carbon dioxide there is the possibility of a large leak due to some type of geological disturbance These disturbances can cause leaks in the forms of fissures springs vents and eruptions amongst others Lewicki et al 2006 This has happened at many naturally occurring CO geologic reservoirs causing flora and fauna to die For example at Mammoth Mountain CA for the past 30 years there has been a definite vegetation kill Figure 7 Lewicki et al 2006 According to Lewicki et al 2006 there has also been a case of one person with asphyxia and one report of a human death In the more extreme eruption cases there have been up to about 1 800 deaths Lewicki et al 2006 These cases show the need for monitoring systems at these sites Areas of dead and y O Figure 7 Vegetation kill at Mammoth Mountain CA http pubs usgs gov fs fs172 96 fs172 96 pdf In 2005 a large group of researchers came together and started a research group focused on developing the monitoring technologies that are required to move forward with practical carbon sequestration This Zero Emissions Research and Technology ZERT program is a research group dedicated to investigating the viability safety and reliability of geological sequestration of c
164. ure 15 intrinsic to the CCDs The path lengths from the back of the lens assembly to each CCD are set to 20 mm This optical model has been tested in Zemax using 15 000 analysis rays carrying a total of 3 watts for the wavelength ranges of interest The rays are randomly distributed across the desired range of wavelengths Two beams 1 5 watts each were modeled one centered on the optical axis and the other at the edge of the field Figure 16 shows that the colors are separated and sent to the correct CCDs with the green rays representing light within the green band red rays representing light within the red band and blue rays 26 representing light within the NIR band In Figure 17 the results are shown as power incident on the three detectors The green sensor detected 1 17 W the red sensor detected 1 19 W and the NIR sensor detected 0 64 W for a total of about 3 W This model provided a good optical design experience with advanced features of the Zemax code and enhances understanding of how the MS 3100 optical system directs color of the proper wavelength to the proper CCD array Dichroic Surfaces Figure 16 Zemax model of a MS 3100 3 chip multispectral imager Here green represents the green color plane blue represents the red color plane and red represents the NIR color plane showing that the dichroic surfaces are modeled effectively 27 2022 27 1820 04 1617 82 1415 59 1213 36 1011 14 4068 94 626 68 494 46 2
165. ure 4 Basic block diagram of carbon dioxide capturing systems Allam et al 2005 Sequestration is the next step in the process Geological sequestration is the process of injecting captured carbon dioxide into suitable rock formations where most of the Earth s supply of carbon is held in coals oil gas organic rich shale and carbonate rocks Anderson et al 2005 In this respect CO sequestration has been happening for millions of years The first test of injected carbon dioxide took place in Texas in the early 1970s Anderson et al 2005 This was done as a part of the enhanced oil recovery EOR program which was started to get more oil out of existing oil wells It worked well and still is working well but did not gain much recognition as a possibility for CO2 mitigation until the 1990s Since the EOR program started other similar sites have been put into place Figure 5 C Co storage proposed K CO EOR EGR ECBM YY CO EOR EGR ECBM proposed A Acid Gas sive monitoring 43 Number of projects if gt 1 7 Area with multiple projects 1000 Km Scale at Equator Figure 5 Location of CO sequestration sites Anderson et al 2005 Geological sequestration of CO is naturally occurring at many places across the world and has been tested at a few sites showing that sequestration of CO produced by humans is a possible method for decreasing the amount of carbon dioxide released into the atmosphere This wa
166. urrent image acquisition To do this first IMAO Extract Buffer is called with a 1 as the input to Buffer to Extract which clears the buffer Next IMAQ Close is called to stop the current asynchronous acquisition which closes all information pertinent to this acquisition and closes the IMAQ session Panel 1 8 only contains a Wait ms VI wired with 5000 so that the program waits 5 seconds before moving onto the next panel Panel 1 9 contains a third flat sequence structure set up to check if the current IT returns images with the calibration target registering the average DN set by the user and 1f 1t does not it will change the IT These panels will be denoted as 1 9 1 for the first panel of the third flat sequence structure Panel 1 9 1 is the NIR IT control Initially a VISA Open VI is called to open a serial communication with the camera for reading and writing attributes to the imager This VI takes the Serial Port Address associated with the imager as an input COM1 It then returns a VISA Resource Name Out that will be used by other VIs to access serial 160 communication with the imager A VISA Configure Serial Port VI is called to set the serial port settings to match the imager s serial settings these can be seen in Table A2 Also this VI takes the VISA Resource Name as an input Next the panel reads the NIR ROI average computed in Panel 1 6 and compares it to the NIR upper limit provided by the user using a Greater Or
167. viewing or illumination angle discussed in Chapter 3 the 2008 method is much more accurate and stable I began taking images on June 27 and finished on August 17 The CO leak started on July 9 and ended on August 7 I imaged from 9 am until 5 pm every three minutes or sometimes longer when IT calibration took place I imaged three regions within each of the mown and un mown vegetation strips one near the pipe 1 m south 2 7 m north of the pipe one far from the pipe 6 3 m 10 m north of the pipe and one in the middle 2 7 m 6 3 m north of the pipe These three regions within the vegetation strips are indicated in Figure 50 Included in the images were three plants of interest named plants 8 9 and 10 Figure 51 These plants were near the pipe along the outer edge of the un mown strip We chose to analyze these data because these plants were very close to a CO hot spot that became visibly apparent about half way into the 2008 experiment I also did this to provide an opportunity to compare data with another researcher using a hyperspectral system Figure 50 View of 2008 vegetation scene Mown and un mown segments 1 2 and 3 are shown here J Shaw 2008 80 Plant O 7 HotSpot Figure 51 View of Plants 8 9 10 on the outside edge of the un mown vegetation test strip during the 2008 experiment J Shaw 2008 Procedures for Calculating Reflectance 2007 Procedure Using Photographic Grey Card The ini
168. was the most consistent choice for explaining the variability in vegetation health and was strongest for statistically separating regions The regression equation can be seen in Equation 17 96 Date B Byoy NDVI 7 rs123 NDVI t i3 tsis NDVI 7 r5 NDW 17 Here Date is the response variable 4 is the intercept Pxpvris the slope NDVI is the predictor variable 7 1 2 is the vegetation region categorical variable that is affected by the relationship between the intercepts for regions 1 and 2 ts 1 2 is the vegetation region categorical variable that is affected by the relationship between the slopes for regions 1 and 2 and 11 1 3 Ts 1 3 and 712 3 Ts 3 3 are the same as above except these are applied to regions and 3 and regions 2 and 3 The NDVI was most consistent in that it had the highest F values for both the 2007 and 2008 un mown regions but had slightly lower values by lt 0 065 than combinations of red NIR and green NIR NDVI for the mown regions and the individual plants respectively More importantly the NDVI alone was best able to statistically separate vegetation regions in every case I believe this came about because whereas spectral band combinations are best for explaining variability in the reflectance spectrum Maynard 2006 Lawrence 1998 in a linear regression including vegetation regions as a categorical variable there will be less variability left to explain the difference in the region
169. wer measured by a spectrometer and cosine as a function of viewing e de tdi sheds o e 71 Normalized power measured by a spectrometer and cosine as a function of the spectralon ale iia 72 ZERT COs detection site layout csi ria 73 Imager orientation at the ZERT Site in 2007 00 eee eeeeeeseecsecneeeeeeeeeneeeaeens 74 View of 2007 setup showing imager orientation in respect to the vegetation test strip Shaw ECE Dept MSU 2007 ccssccssssscscnsccsensecesnsccesnsscesecesnees 75 View of 2007 vegetation scene from scaffolding Mown and Un mown regions 1 2 and 3 shown here Shaw ECE Dept MSU 2007 eee 76 View of 2007 vegetation scene Mown and Un mown regions 1 2 and 3 are shown here Shaw ECE Dept MSU 2007 oooconcccccoocccconcnononcnononcncnnnaninnnnnos 71 Imager orientation at the ZERT Site in 2008 0oooococnnconocccoccnonnnnonncnoncnannconanonnnonns 78 View of 2008 vegetation scene Mown and un mown regions 1 2 and 3 are shown here Shaw ECE Dept MSU 2008 coooccccoocccoococccoccccconcnononcnonancnonananinn 79 View of Plants 8 9 10 Shaw ECE Dept MSU 2008 oooonccncccoconococonicnncancnos 80 Figure 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 XV LIST OF FIGURES CONTINUED Page Reflectance for one day using grey card to calibrate all images ee 82 Imager and MODTRAN setup to equate horizontal and vertical irradiances 84
170. were negatively affected quickly while plants far from the pipe were affected positively Data from the second experiment showed that the net effect of leaking CO2 depends on the relationship between CO sink source balance and vegetation density Also due to the strong calibration techniques employed in 2008 the imaging system was able to see the effects of water and hail on the vegetation We have also found a way to image continuously through the day not having to worry about clouds or sun to scene scene to imager angle effects This system s easy setup automation all day imaging capability and possibility for low cost makes it a very practical tool for plant stress measurements for the purpose of detecting leaking CO2 INTRODUCTION Greenhouse gases make life sustainable on Earth by trapping some of the Sun s incoming short wave radiation in an Earth surface atmosphere energy transfer system the greenhouse effect The warming of the Earth starts with short wave radiation entering the atmosphere About 30 of this radiation is reflected back into space by clouds atmospheric particles reflective ground surfaces and the ocean surf so about 70 of the short wave solar radiation is absorbed by land air and ocean Remer 2007 These Earth features then emit this energy as long wave thermal radiation Almost half of this reemitted radiation is transmitted out of the atmosphere and more than half is absorbed by greenhouse gases such as car
171. were no plants at 19 m south and over the pipe since there were paths in these positions It took about 2 hours to image all of these plants from 10 am until 12 pm When imaging from the scaffolding I imaged three regions within each vegetation strip one near the pipe 0 5 m 4 5 m from the pipe one far from the pipe 9 m 13 5 m from the pipe and one in the middle 4 5 m 9 m from the pipe as indicated in Figure 47 and 48 Un mown Figure 47 View of 2007 Vegetation Scene from Scaffolding Mown and Un mown segmentss 1 2 and 3 are shown here with black lines J Shaw 2007 77 J Shaw 2007 2008 Experimental Setup and Imaging Method During the 2008 experiment the test vegetation area was set on the south west end of the pipe about 30 m south west of the 2007 location It ran 30 m along the pipe and 10 m to the north west and south east of the pipe Figure 49 To make the two experiments similar a 1 5 m wide mown strip and a 1 5 m wide un mown strip were established on the north east side of the vegetation test area The strips ran from 2 m south of the pipe to 10 m north of the pipe I placed the imager on a 10 ft 3 m scaffolding about 3 m south of the intersection between the pipe and the vegetation and just to the west of the vegetation Figure 49 The imager was set at a 45 elevation angle and a 320 azimuth angle so that the imager would view vegetation from m south of the pipe to 10 m north of the pi
172. wn segment in 2008 indicate that throughout all regions the vegetation has been stressed For this segment it can be seen in Figure 67 that the NIR reflectance slopes for all regions are nearly the same though in region one it begins and ends much lower than in the other regions The red reflectance for region 2 and 3 are very similar about 12 Region 1 approximately 7 starts much lower than regions 2 and 3 All regions end at nearly the same reflectance approximately 15 The green reflectance s slopes are nearly the same though region starts and ends much lower than the other two regions Again the red reflectances end higher than the green reflectances The date versus NDVI regressions shown in Figure 68 agree with the red reflectance in that region 2 NDVI increases more slowly over time compared to regions and 3 The NDVI for regions 1 and 3 have nearly the same slope but very different beginning and end points The regression coefficient of determination 0 9033 Table 19 was able to explain the variability well The regression was significant with a p value of lt 2 2e 16 Table 19 The p values for both the intercept and slope regression coefficients Table 20 show that regions 1 and 3 and regions 2 and 3 were statistically separable 106 50 Reflectance Region 1 Region 2 07 16 07 21 07 26 07 31 Date Figure 67 2008 un mown segment green red and NIR reflectances for regions 1 s
173. y long term change in NDVI was not likely caused by the soil moisture evaporating transpiring and that the vegetation stress was more likely a result of the increased CO concentration This effect from the CO can be seen in the different NDVI slopes for each segment s three regions For example the CO2 may be a nutrient for the mown region since the vegetation had access to adequate amounts of water while the un mown segment did not have enough water so the excess CO may have been detrimental This agrees with the results from Arp 1991 which showed that sink source balance and vegetation density are both important in determining whether a given CO concentration 118 acts as a nutrient or as a stress There was also a small diurnal variation in soil moisture though this does not appear to strongly affect the vegetation ee Figure 77 Image of mown and un mown segments taken 9 July 2008 a and 9 August 2008 to visually illustrate the change in the health of the vegetation J Shaw 2008 7 if Figure 78 Close up images of mown segment taken 9 July 2008 a and 9 August 2008 b to visually illustrate the change in the health of the vegetation J Shaw 2008 gt ie Figure 79 Images taken 3 July 2008 a and 9 August 2008 b to visually illustrate the change in the health of the vegetation J Shaw 2008 Plant 10 s location is indicated by the blue circle Plants 8 and 9 locations are indicated
174. y altering the current to the light bulb in the sphere The gain was set to 3 dB 5 dB 5 dB for the green red and NIR planes respectively The integration time was set to 18 ms 25 ms and 33 ms for the green red and NIR planes respectively The F was set to 11 and the focus was set to 25 cm In Figure 33 the horizontal axis shows the current measured by a detector in the sphere and the vertical axis shows the average DN measured by the MS 3100 Again the green exhibits the most non linear behavior but the red and NIR planes are quite linear This means that in the field the camera settings should be adjusted so that the 99 spectralon panel registers less than about 180 DN for the green color plane to avoid saturation Since the green channel is obviously non linear past 180 DN and it was never run in that regime the linear fit only considers points of 180 DN and less The R values are 0 9963 0 9957 and 0 9976 for the green red and NIR color planes respectively 250 200 150 Average DN 100 O Green O Red x NIR 50 y 4 43 5 78 Green Linear Fit y 2 32 x 6 59 Red Linear Fit y 2 27 x 4 44 NIR Linear Fit 0 20 40 60 80 100 120 Detector Current microamps Figure 33 Average DN measured by the imaging system when viewing the integrating sphere plotted versus current measured by the integrating sphere s detector 59 Next I examined the response of the imager to temperature c

MEASUREMENTS OF PLANT STRESS IN

Contents

Download Pdf Manuals

Related Search

Related Contents