UMD Global 250 meter Land Water Mask User Guide 1
Contents
1. www2 jpl nasa gov srtm index html Digital Media accessed June 2006 Vermote E El Saleous N and Justice C 2002 Atmospheric correction of MODIS data in the visible to middle infrared first results Remote Sensing of Environment 83 1 amp 2 97 111 Vermote E F and Kotchenova S 2008 Atmospheric correction for the monitoring of land surfaces Journal of Geophysical Research Atmospheres 113 12 Wan Z Zhang Y Zhang Q and Li Z 2002 Validation of the land surface temperature products retrieved from Terra Moderate Resolution Imaging Spectroradiometer data Remote Sensing of Environment 83 1 amp 2 163 1802. N Wolfe R 200 Global Land Water Mask Derived from MODIS Nadir BRDF Adjusted Reflectances NBAR and the MODIS Land Cover Algorithm Slater J 2006 personal communication April 11 2006 Strabala K 2004 MODIS cloud mask user s guide Retrieved Dec 1 2004 from http cimss ssec wisc edu modis 1 pdf CMUSERSGUIDE PDF SWBD 2005 Shuttle Radar Topography Mission Water Body Data set http www2 jpl nasa gov srtm index html accessed June 2006 Vermote E El Saleous N and Justice C 2002 Atmospheric correction of MODIS data in the visible to middle infrared first results Remote Sensing of Environment 83 1 amp 2 97 111 Vermote E F and Kotchenova S 2008 Atmospheric correction for the monitoring of land surfaces Journal of Geophysical Research Atmospheres 113 12 Wan Z Zhang Y Zhang Q and Li Z 2002 Validation of the land surface temperature products retrieved from Terra Moderate Resolution Imaging Spectroradiometer data Remote Sensing of Environment 83 1 amp 2 163 180 4 Contact Information Data can be found at the GLCF http landcover org data watermask and LP DAAC special collections ftp emodisftp cr usgs gov GlobalLandWaterMask For further information on the product generation see Carroll et al New Global 250m land water mask in International Journal of Digital Earth submitted Feb 2009 Contact Mark Carroll for further information markc umd edu In Press Internatio
3. The Shuttle Radar Topography Mission SRTM collected 30m interferometric Synthetic Aperture Radar data over the course of 11 days in February 2000 For security reasons data were released to the public at the degraded 90m resolution except for the US The purpose of the mission was to create a new consistent fine resolution Digital Elevation Model DEM with nearly global coverage The process of converting the raw data to a DEM created as a byproduct the identification of water bodies Water bodies had to be identified so that consistent elevation values could be maintained for non land areas The water bodies were given an elevation 1m below the elevation of the surrounding shoreline and rivers were given a consistently decreasing value to create an even flow The result was a reliable depiction of water bodies for a large portion of the globe It was then decided to release this depiction as a separate data set called the SRTM Water Body Data set or SWBD SWBD 2005 Most of the remotely sensed data products depicting water have been derived from spectral data that were then classified The use of SRTM data to create a water mask represents a different method of using remotely sensed data to create a global consistent mask than any of the products shown in Table 1 Recently at MODIS Science team meetings in October 2006 and April 2008 the science community s needs for a new land water mask were discussed and it was agreed that a mask create
4. averaged SWBD data for regions between 60 S to 60 N 2 2 1 80 to 90 N Tiles in row v00 80 to 90 N were handled separately because most of the water in this area remains frozen even in summer due to the high latitude and in some cases there are ice shelves that extend from the land to the ocean In the MODIS tile grid there are only 4 tiles in this region which contain land Because of the small number of tiles and the complex landscape an inverse mapping approach was adopted whereby water was determined by first mapping the visible land and the area outside the projection The remaining area was initially labeled as water and was reclassified as land if it could be determined that it was indeed ice over land This was accomplished by creating a decision tree with 3 classes land ice and water and applying it to the 4 16 day composites that comprise July and August for 5 years from 2003 2007 The information from 2003 2007 was combined to yield a single static map for each of the 4 tiles The EOS DEM for MODIS contains the old water mask and was found to have substantial locational shifts which made it unsuitable to use in determining elevation For this reason interior ice sheets were digitized because no other consistent DEM product was found to determine elevation The NSIDC 1km DEM for Greenland was used in the area of the McKinley Sea in the Northeastern corner of Greenland due to the existence of an ice shelf 2 3 60 to 9
5. old EOS 10N 0 8 Improved representat water mask The large lake in the north center Manitoba in Canada igure F Reindeer Lake on the border between Saskatchewan and 1s New Water New Land N Coincident Water Coincident Land km 40 a fo lt ct Figure 9 Comparison of the new 250m water mask with the old EOS water mask for 4 MODIS tiles in the Mid Atlantic region of the United States References Breiman L Friedman J H Olshen R A amp Stone C J 1984 Classification and regression trees New York Chapman amp Hall Carroll M Townshend J Hansen M DiMiceli C Sohlberg R Wurster K 2006 Vegetative Cover Conversion and Vegetation Continuous Fields In Ramachandran B Justice C O Abrams M eds Land Remote Sensing and Global Environmental Change NASA s Earth Observing System and the Science of ASTER and MODIS Springer Verlag in press DiMarzio J A Brenner R Schutz C A Shuman and H J Zwally 2007 GLAS ICESat 1 km laser altimetry digital elevation model of Greenland Boulder Colorado USA National Snow and Ice Data Center Digital media accessed December 2008 ESRI 1992 The Digital Chart of the World for use with ARC INFO Data Dictionary ESRI Redlands CA Haran T Bohlander J Scambos T Fahnestock M 2005 MODIS mosaic of Antarctica MOA image map Boulder CO USA National Snow and Ice Data Center Digital media accessed December 2008 Ho
6. omits Antarctica and in the north this omits most of Alaska the northern parts of Canada Europe and Asia as well as Greenland In addition the SWBD was created as ArcView shapefiles in Geographic projection and subsetted into 1 squares This format is acceptable for local or small regional studies but is cumbersome for doing large area studies Note that there are over 12 300 individual files necessary to get the full coverage of land surface for the SWBD If one tries to stitch together a large number of these enough to make a single MODIS tile for example in most cases the software ARCGIS 9 will crash because of the daunting number of individual shapes In addition despite best efforts there are still data gaps in the SWBD These gaps can occur when there are mid stream islands and or where cloud cover was persistent pers comm James Slater SWBD team April 11 2006 An attempt was made by the SWBD team to use the Landsat Geocover data to fill these gaps but gaps remain where the Geocover data was also too cloudy to make a determination A global 250m data set in 16 day composites for the entire 8 years of Terra data and 6 years of Aqua data Collection 5 is online at the University of Maryland This data set MOD44C was originally created as the input to the MOD44A Vegetative Cover Conversion and MOD44B Vegetation Continuous Fields products For a full description of this products see Carroll et al 2006 During the composit
7. reviewed publications See the acknowledgements for access information 2 1 Area from 54 S to 60 N Initially the SWBD was reprojected to MODIS Sinusoidal projection converted from vector to raster and stitched into MODIS tiles at the native 90m spatial resolution These 90m resolution tiles were aggregated to 250m resolution by absolute averaging to yield percent water content per pixel Gaps in the SWBD derived 250m map were detected and filled in an automated way using the methodology shown in Table 2 Figure 4 shows an example of a gap being detected and filled using the methodology in Table 2 Use the SWBD converted to raster and subset into MODIS tiles as the base mask Group areas of contiguous water pixels into discrete water bodies 3 Create a reference map using year of 250m daily water and land hits o From the MOD44C composites for a year compute the sum of land hits and the sum of water hits o Those pixels with at least 100 total observations and greater than 75 water hits are considered water 4 Working within a 10 x 10 pixel kernel o Search for discrete water bodies that terminate within the kernel o If found use the reference map created from a year of daily water hits to find suitable observations to connect the water bodies o Constraint if the total number of water pixels in the kernel before adding from the reference exceeds 20 there are 100 pixels in a 10x10 kernel do not try to connect Thi
8. the complex landscape with permanent sea ice and frozen interior water bodies the method applied to lower latitudes did not work sufficiently well in this region To solve this problem an inverse mapping approach was adopted whereby water was determined by first mapping the visible land and the area outside the projection The remaining area was initially labeled as water Mapping was done by creating a decision tree with 3 classes land ice and water and applying it to the 4 16 day composites that comprise July and August for 5 years from 2003 2007 Images from July and August were used to coincide with the timing when snow cover was minimal The information from 2003 to 2007 was combined to yield a single static map for each of the 4 tiles Interior ice sheets were determined by visual interpretation of MOD44C composites and referencing with the classified image Ice sheets were then mapped into the land water mask as land The EOS DEM for MODIS contains the old water mask and was found to have substantial locational shifts which made it unsuitable to use in determining elevation The NSIDC 1km DEM DiMarzio et al 2007 for Greenland was used in the area of the McKinley Sea in the Northeastern corner of Greenland due to the existence of an ice shelf 2 4 60 to 90 S The MOA grounding line vector data set has been reprojected from Polar Stereographic to Sinusoidal converted from vector to raster and subset into MODIS tiles The polyline sha
9. 0 S The MOA grounding line vector data set has been reprojected from Polar Stereographic to Sinusoidal converted from vector to raster and subset into MODIS tiles The polyline shapefile was converted to a polygon and rasterized such that any data inside the polygon was considered land and anything outside the polygon is considered water This reformatted product is included in the beta release of the new 250m water mask as the land water mask for Antarctica The grounding line is the point at which the ice sheet is still resting on solid rock Scambos et al 2007 The cryospheric community has used this reference in their products for a number of years 2 4 Quality Assurance Data Layer A QA layer was maintained that shows which data source provided the water pixel For example the area seen in red in figure 4 has a value that is distinct from the area shown in blue designating that it came from a different source in this case MODIS Users can utilize this layer to determine the utility of the data Success was determined by overlaying the new water mask with current MODIS surface reflectance data multi day composites to discover any gross errors of commission Additionally we will release a beta version early in the project in order to incorporate user feedback into the quality control process by investigating areas identified by end users as problematic 3 Product Details The new 250m water mask is a global raster data set in the Sin
10. ROIK Survey 1996 1 km resolution Vector Limited spatial resolution locational Land Processes accuracy varies by EDC Land Sea Mask_ DAAC 1996 1 km region Raster Limited spatial resolution significant BU MODIS Land Boston discontinuities in Sea Mask University 2004 1 km river networks Raster Lacks complete global coverage discontinuities SRTM Water Body remain in some major Detection SWBD NASA JPL 2005 90 m rivers Vector Table 1 Global surface water data sets Figures and Captions Tapajos River Juruena Teles Pires m New 250m Land Water Mask E Current EOS 1km Water Mask E Original EOS 1km Water Mask 1 0 km 20 Figure 1 Comparison of new 250m water mask with the original 1km MODIS EOS water mask and current Ikm MODIS EOS water mask updated by Boston University in 2002 Figure 2 MODIS Vegetation Continuous Fields VCF with current 1km MODIS EOS water mask overlain in blue The blocky appearance and discontinuous drainage lines are consistent with 1km raster water masks Central African Republic Democratic Republic of Congo Figure 3 SWBD shown in blue for rivers in central Africa note that gaps exist in the main stems of the rivers 3a shows a portion of the main stem of the Congo river to the northwest of Kisangani Democratic Republic of Congo 3b shows a portion of the Ubangi river where the Bomu river to the north defines the border between Democratic Republic of Congo and the Cent
11. S water mask 4227 3043 28 01 97 51 polygons Table 4 Comparison of the NLCD 2001 data set for Alaska United States to the new 250m water mask and to the old EOS water mask The results for the new 250m water mask show that 98 of the polygons intersect with NLCD polygons leaving only 2 of all 250m polygons outside of NLCD polygons However the new 250m water mask overestimates the surface area of water by 18 compared to the NLCD This overestimation is typically at the border of water bodies where the coarser MODIS spatial resolution overlaps the true land water boundary as compared to the finer resolution data Hence a mask created from finer resolution data could provide an even better representation of the water features Nearly 21 of the NLCD polygons did not have any intersection with new 250m mask polygons This number was higher than expected but upon further review the NLCD polygons that did not have intersections were mostly 1 2 pixel polygons It is likely that these were undetectable using MODIS data due to their small size relative to MODIS spatial resolution 250m and were picked up by the NLCD due to its finer native spatial resolution 30m Additionally the NLCD was not intended for the purpose of detecting water Water is merely a byproduct of identifying different classes of land cover so there may be errors in the NLCD resulting in false detections of water in that data set The NLCD was used for this analysis because
12. UMD Global 250 meter Land Water Mask User Guide Mark Carroll Charlene DiMiceli John Townshend Praveen Noojipady Robert Sohlberg 1 Introduction The new 250m land water mask was created in three sections using primarily 3 different data sources The main body of the product from 60 S to 60 N was created using the SRTM Water Body Dataset SWBD 2005 and supplementing with MODIS 250m data as necessary The area between 60 and 90 N was generated completely from MODIS 250m data While the area covering Antarctica between 60 and 90 S was generated using the Mosaic of Antarctica product The SWBD was used because of its fine spatial resolution and because of its consistent representation of the land surface Since the SRTM data was collected over a short time step 11 days it will provide a spatially coherent representation of surface water Additionally the cloud penetrating properties of the RADAR offers superior performance over spectral data alone particularly in cloudy areas such as the humid tropics Using this remotely sensed data product has the advantage of a single source of information unlike the vector data sets which are dependent on disparate sets of information to create a single data set The SWBD represents a significant improvement in the representation of land and water Unfortunately a variety of problems remain with this data set Foremost is coverage since it extends only from 55 S to 60 N In the south this
13. able 3 Water bodies in this region include Deep Ocean coastal bays inland rivers and inland lakes A total of 6 369 127 pixels were mapped as inland water in the new 250m water mask The ocean pixels were excluded from the statistical analysis The new water mask identified 1 274 106 pixels as water that were previously mapped as land This represents gt 68 000 km of new surface water area or 20 more water represented in the new map than was present in the old map Additionally nearly 330 000 pixels that were previously mapped as water were re mapped as land in the new mask This represents 5 of the total inland water pixels in the old mask or nearly 18 000 km Area Percent of Number mapped total Data Set Comparison of Pixels km pixels New 250m land pixels previously mapped as water in the old EOS Water Mask 329 922 17 705 5 18 New 250m inland water pixels previously mapped as land in the old EOS Water Mask 1 274 106 68 374 20 00 Total number of pixels mapped as inland water 6 369 127 Table 3 Comparison between the new 250m water mask and the old EOS 1km water mask remapped to a 250m grid for inland water in the Mid Atlantic region of the United States In areas north of 60 N features smaller than 2 to 3 MODIS pixels can be missed due to the spatial resolution of the MODIS instrument This can result in a feature that is represented by the finer resolution SRTM product up to the 60 N line
14. and then under represented by the coarser MODIS resolution This situation was intensely investigated by the developers and found to be a rare occurrence Data were used from multiple years of MODIS data to minimize any impact of flooding on the output product Small islands off the coast of mainland continents may be missed but this occurs rarely and should have little impact on downstream processing of data products which is the primary purpose of this product 4 Validation The new land water mask is intended to replace the Ikm MODIS EOS raster data set currently being used in MODIS data production As such the results from the new mask are primarily being judged against the mask that it is replacing However additional comparisons with other products have been performed Validation of the SWBD has already been performed by NASA JPL In summary the absolute vertical accuracy was determined to be 9m and the absolute geolocation accuracy was determined to be 8m Rodriguez et al 2006 Validation of the MOA has been performed by the developers of the MOA Haran et al 2005 The developers found no discrepancies greater than 125m for fixed objects in well mapped areas in more than 260 scenes Haran et al 2005 For purposes of this project this validation was accepted and not repeated Validation for the region between 60 and 90 N in North America was done using a 30m land cover classification These data are available for Alaska
15. averaging to yield percent water content per pixel The projection from the native Geographic projection to Sinusoidal projection can result in a loss of locational precision with increasing latitude However the conversion from vector to raster and subsequent aggregation from 90m resolution to 250m resolution was sufficient to minimize any discrepancies due to loss of precision with latitude Gaps in the SWBD derived 250m map were detected and filled in an automated way using the methodology shown in Table 1 e Use the SWBD converted to raster and subset into MODIS tiles as the base mask e Group areas of contiguous water pixels into discrete water bodies e Create a reference map using year of 250m daily water and land hits o From the MOD44C composites for a year compute the sum of land hits and the sum of water hits o Those pixels with at least 100 total observations and greater than 75 water hits are considered water e Working within a 10 x 10 pixel kernel o Search for discrete water bodies that terminate within the kernel o If found use the reference map to find suitable observations to connect the water bodies o Constraint if the total number of water pixels in the kernel before adding from the reference exceeds 20 there are 100 pixels in a 10x10 kernel do not try to connect This constraint helps avoid problems of connecting lakes Table 1 Description of gap detection and filling algorithm The SWBD did not provid
16. ces Breiman L Friedman J H Olshen R A amp Stone C J 1984 Classification and regression trees New York Chapman amp Hall Carroll M Townshend J Hansen M DiMiceli C Sohlberg R Wurster K 2006 Vegetative Cover Conversion and Vegetation Continuous Fields In Ramachandran B Justice C O Abrams M eds Land Remote Sensing and Global Environmental Change NASA s Earth Observing System and the Science of ASTER and MODIS Springer Verlag accepted DiMarzio J A Brenner R Schutz C A Shuman and H J Zwally 2007 GLAS ICESat 1 km laser altimetry digital elevation model of Greenland Boulder Colorado USA National Snow and Ice Data Center Digital media ESRI 1992 The Digital Chart of the World for use with ARC INFO Data Dictionary ESRI Redlands CA Haran T Bohlander J Scambos T Fahnestock M 2005 MODIS mosaic of Antarctica MOA image map Boulder CO USA National Snow and Ice Data Center Digital media Justice C Giglio L Korontzi S Owens J Morisette J Roy D Descloitres J Alleaume S Petitcolin F and Kaufman Y 2002 The MODIS fire products Remote Sensing of Environment 83 1 amp 2 244 262 Lehner B and Doll P 2004 Development and validation of a global database of lakes reservoirs and wetlands Journal of Hydrology 296 1 22 Salomon J Hodges J Friedl M Schaaf C Strahler A Gao F Schneider A Zhang X El Saleous
17. d at 250m resolution would meet many of the needs of the current users of MODIS data Additionally it would be valuable for future missions such as National Polar orbiting Operational Environmental Satellite System NPOESS and the NPOESS Preparatory Project NPP which will produce products at similar spatial resolutions The global raster dataset will be distributed in digital format through the Global Land Cover Facility website http landcover org 2 Methods The new 250m land water mask was created in three sections using 3 different data sources The main body of the product from 54 S to 60 N was created using the SWBD and supplemented with MODIS 250m data as necessary The area between 60 and 90 N was generated completely from MODIS 250m data while the area covering Antarctica between 60 and 90 S was generated using the Mosaic of Antarctica MOA product Haran 2005 The SWBD was used because of its fine spatial resolution and because of its consistent representation of the land surface Since the SRTM data were collected over a short time period of only 11 days it should provide a spatially coherent representation of surface water Additionally the cloud penetrating properties of the Radar offers superior performance over optical data alone particularly in cloudy areas such as the humid tropics Using this remotely sensed data product has the advantage of a single source of information unlike the typical vector data sets wh
18. e coverage between 55 S to 60 S however there is essentially no land surface in this area There are a total of 6 MODIS tiles that are produced to have land in them in this range and it was found that there was only 1 island not included in the SWBD in 1 tile This island was mapped using MODIS 250m data 2 2 60 to 90 N MOD44C 250m 16 day composites are also available for areas between 60 and 90 N where the SWBD is not available These data were used to create a new 250m resolution land water mask The data were classified using regression tree classification Breiman et al 1984 Training data were derived using the aggregated SWBD using a tile in the MODIS v03 tile row 50 to 60 N and the tree was applied to tiles in rows vO1 and v02 geographically nearby A total of 3 different trees were used 1 in North America in Europe and 1 in Russia Different trees were used in different geographic locations to accommodate locally different ground cover to maximize the efficiency of the tree The regression trees were applied to multiple time periods and the resulting classifications were averaged to increase the confidence that features were mapped correctly The regression tree yields a subpixel estimate of the water component of a pixel Features were determined to be water bodies if the averaged classification result showed 50 or greater water content This threshold is consistent with the threshold used to determine water using the
19. e entire 8 years of Terra data and 6 years of Aqua data Collection 5 is online at the University of Maryland This data set MOD44C was originally created as the input to the MOD44A Vegetative Cover Conversion and MOD44B Vegetation Continuous Fields VCF products For a full description of these products see Carroll et al 2006 During the compositing process the daily surface reflectance data Vermote and Kotchenova 2008 was interrogated using a decision tree algorithm to distinguish between water and land This daily depiction of water was stored in the 16 day composite data as a sum of hits labeled as water in the process These hits were then interrogated and used where ever gaps exist in the SWBD The MODIS mosaic of Antarctica MOA available from the National Snow and Ice Data Center NSIDC DAAC is a mosaic of MODIS 250m level 1b L1B data for the continent of Antarctica Haran et al 2005 This was generated using the Radarsat Antarctic Mapping Project Antarctic Mapping Mission 1 RAMP AMM data Haran et al 2005 as a reference to overlapping MODIS observations to create a fine resolution 125m image for the continent of Antarctica This vector shoreline product is available from the National Snow and Ice Data Center NSIDC Distributed Active Archive Center DAAC All data sets used here are available free of charge from various websites and have either been published or used in products that have appeared in peer
20. gins with 00 so to find California United States we see that we cross the horizontal to h08 and go down the vertical to v05 and the tile ID is hO8v05 Training data were derived using the aggregated SWBD using a tile in the MODIS v03 tile row 50 to 60 N and the tree was applied to tiles in rows v01 and v02 geographically nearby A total of three different trees were used one in North America one in Europe and one in northern Asia Different trees were used in different geographic locations to accommodate locally different ground cover to maximize the efficiency of the tree The regression trees were applied to multiple time periods and the resulting classifications were averaged to increase the confidence that features were mapped correctly Insert Figure 5 The regression tree yields a subpixel estimate of the water component of a pixel Features were determined to be water bodies if the averaged classification result showed 50 or greater water content This threshold is consistent with the threshold used to determine water using the averaged SWBD data for regions between 54 S to 60 N 2 3 Area from 80 to 90 N Tiles in row v00 80 to 90 N were handled separately because most of the water in this area remains frozen even in summer due to the high latitude and in some cases there are ice shelves that extend from the land to the ocean In the MODIS tile grid there are only four tiles in this region which contain land Because of
21. he land boundary is difficult to determine and errors may occur However given that the principal purpose of the mask is to ensure that terrestrial and oceanic algorithms are applied to the appropriate pixels this should not be regarded as a major deficiency Small artifacts may exist in areas where there were recurring cloud or terrain shadows that went undetected or where the sensor viewing geometry was far off nadir Both of these are minimized through the use of multiple composites from multiple years 6 Conclusions The new 250m water mask is a dramatic improvement over the current 1km raster mask that is used in MODIS data processing and many other purposes The product will be included in the MODIS Collection 6 reprocessing as the standard water mask used in the creation of many of the MODIS standard products It will also be incorporated into the MODIS Vegetation Continuous Fields product as well as the MODIS Land Cover product This product is not intended to be used for hydrologic modeling and caution should be used until the remaining discontinuities in rivers have been resolved The land water mask product was released as a beta product to the MODIS Science Team for evaluation purposes in February 2009 The product will also be suitable for use with similar coarse resolution satellite data from other systems It will be officially released by June 2009 and will be available in MODIS tile format through the special collections at the Land P
22. hio 2 In the World Vector Shoreline 2004 produced by the National Geospatial Intelligence Agency NGA and National Oceanic and Atmospheric Administration NOAA there was no update for a 10km shift in the location of the mouth of China s Huang Yellow River since 1978 due to the seaward growth of the delta 3 The water mask provided in EOS AM 1 Digital Elevation Model Data Sets 1999 produced by NASA Jet Propulsion Laboratory JPL is limited by its coarse spatial resolution of Ikm which results in insufficiently defined coast lines 4 An even more critical issue with the EOS AM 1 Digital Elevation Model Data Sets 1999 is that many rivers are offset from their actual location Such is the case with the Tapajos and Xingu Rivers in South America the location of which are in error by as much as 10 km figure 1 Insert Figure 1 5 The recent Boston University BU Water Mask 2004 which is now in use as the standard water mask used for products derived from the Moderate Resolution Imaging Spectro radiometer MODIS is limited by its spatial resolution of 1 km figure 2 This mask does not reliably depict continuous hydrologic networks but typically does label drainage systems in the correct locations In figure 1 the BU mask is in orange and can be seen behind the new 250m water mask in blue Insert Figure 2 6 The Global Lakes and Wetlands Database GLWD 2004 produced by Lehner and Doll combines existing vecto
23. ich are dependent on disparate sets of information to create a single data set The SWBD represents a significant improvement in the representation of land and water Unfortunately a variety of problems remain with this data set Foremost is coverage since it extends only from 54 S to 60 N In the south this omits Antarctica and in the north this omits most of Alaska the northern parts of Canada Europe and Asia as well as Greenland In addition the SWBD was created as ArcView shapefiles in Geographic projection and subsetted into 1 squares This format is acceptable for local or small regional studies but is cumbersome for doing large area studies Note that there are over 12 300 individual files necessary to get the full coverage of land surface for the SWBD If one tries to stitch together a large number of these enough to make a single MODIS tile for example in most cases the software ARCGIS 9 will crash because of the daunting number of individual shapes In addition despite best efforts there are still data gaps in the SWBD Figure 3 These gaps can occur when there are mid stream islands and or where cloud cover was persistent pers comm James Slater of the SWBD team April 11 2006 An attempt was made by the SWBD team to use the Landsat Geocover data to fill these gaps but gaps remain where the Geocover data was also too cloudy to make a determination Insert Figure 3 A global 250m data set in 16 day composites for th
24. in the United States in the National Land Cover Dataset NLCD Homer et al 2001 These data were created within the last 5 years using Landsat data from 1990 2000 and are being made available to us from the USDA Forest Service Table 4 shows the results of the analysis of the NLCD data set compared to the new 250m water mask and also the old EOS 1km mask The NLCD was aggregated from 30m to 250m by exact averaging A pixel from the NLCD was determined to be water if it contained 50 or greater water This aggregated map was compared to both the new 250m mask and the old EOS mask The old EOS 1km mask was resampled to 250m resolution using nearest neighbor resampling All three datasets were converted from raster to polygon data and the polygons were dissolved to join neighboring polygons After the dissolve polygons were selected based on location where new 250m mask polygons intersect NLCD water polygons and similarly where the old EOS mask polygons interest NLCD water polygons The results are displayed in table 4 Commission error was calculated by intersecting polygons total polygons and the Omission error was calculated by Total NLCD polygons intersecting polygons Total NLCD polygons Data set Total of of polygons Commission Omission polygons intersecting Error Error NLCD Total NLCD water 122114 polygons New 250m water mask 98514 96552 1 99 20 93 polygons Old EO
25. ing process the daily surface reflectance data Vermote and Kotchenova 2008 was interrogated using a decision tree algorithm to distinguish between water and land This daily depiction of water was stored in the composite data as a sum of hits labeled as water in the process These hits were then interrogated and used where ever gaps exist in the SWBD The MODIS mosaic of Antarctica MOA available from the NSIDC DAAC is a mosaic of MODIS 250m level 1b L1B data for the continent of Antarctica Haran et al 2005 This was generated using the Radarsat Antarctic Mapping Project Antarctic Mapping Mission 1 RAMP AMM data as a reference to overlapping MODIS observations to create a fine resolution 125m image for the continent of Antarctica We anticipate the release of a vector shoreline of Antarctica from this data set in February 2007 When this is released it will be evaluated as a replacement for the existing 1km product for Antarctica All data sets used here are available free of charge from various websites and have either been published or used in products that have appeared in peer reviewed publications see acknowledgements for access information 2 Methods 2 1 60 S to 60 N Initially the SWBD was reprojected to MODIS Sinusoidal projection converted from vector to raster and stitched into MODIS tiles at the native 90m spatial resolution These 90m resolution tiles were aggregated to 250m resolution by absolute
26. it was derived independently from the MODIS data set and the accuracy is stated by the developers as 90 Homer et al 2001 The EOS water mask showed that 72 of the polygons matched polygons from the NLCD The old EOS water mask missed nearly 98 of all polygons shown in the NLCD The poor performance of the old EOS 1km mask relative to the NLCD is attributable to the coarse spatial resolution compared to the small size of the lakes in Alaska the region covered 5 Remaining Issues There are still some remaining issues that could not be alleviated with the new water mask These issues include discontinuities in small rivers which occurred infrequently in rivers that have sections smaller than 250m in width and hence were difficult to detect with MODIS We will investigate in the future whether Landsat and Aster can be used automatically to fill the gaps possibly intelligent interpolation procedures based on the known rules of behavior of drainage patterns Persistent floating sea ice was often labeled as land typically occurring in areas north of 75 N latitude We attempted to clear these by manually digitizing the features if they were labeled as ocean in the old EOS 1km water mask but some may remain The QA layer maintains the information for how each pixel was derived and does show if a pixel was derived by digitization Where the ice shelf extends into the sea from the land as in Greenland some Islands north of Siberia and Antarctica t
27. line of latitude The spatial continuity across the line is remarkable and the improvement over the existing 1km data set figure 7b and 7c is evident This example shows that while there may be some disparities between the MODIS data and SWBD the differences are quite minor This result was consistent with results found in other areas across the globe Insert Figure 7 Substantial improvement in spatial detail of the new mask has already been shown in figure 1 for areas where the SWBD was used Comparable improvement in spatial detail is seen in the northern latitudes where there is a high density of small lakes Figure 8 shows this improved representation for central Canada west of Hudson Bay as compared to the 1km mask Similar improvements are seen in Scandinavia and Siberia Insert Figure 8 The mapping of Antarctica is done using the vector representation of the grounding lines for the ice sheets Evaluating this with data from MODIS is difficult due to the limitations of visible data The cryospheric scientists in the MODIS Science team requested that the data be represented in this way so we honored that request Insert Figure 9 A quantitative comparison of the old 1km water mask and the new 250m water mask was undertaken for 4 adjoining MODIS tiles in the Mid Atlantic region of the United States tiles h11v04 h11v05 h12v04 h12v05 Figure 9 shows the results of this comparison visually and the numerical results are shown in T
28. map of surface water at 250 m spatial resolution This effort is automated and intended to produce a dataset for use in processing of raster data MODIS and future instruments and for masking out water in final terrestrial raster data products This new global dataset is produced from remotely sensed data and provided to the public in digital format free of charge The data set can be found on the Global Land Cover Facility GLCF website at http landcover org This dataset is expected to be a base set of information to describe the surface of Earth as either land or water which is a fundamental distinction upon which other descriptions can be made 1 Introduction Accurate depiction of the land and water is critical for the production of land surface parameters from remote sensing data products Without such a reliable mask there will be areas of water to which terrestrial algorithms will be applied and conversely areas of land to which water algorithms are applied Among the important parameters requiring a mask include the cloud mask Strabala 2004 land surface temperature Wan et al 2002 active fires Justice et al 2002 and surface reflectance Vermote et al 2002 Many global databases have been created to depict global surface water but these databases still fall short of the needs of the terrestrial remote sensing community especially for products with a 250m spatial resolution Existing global databases of water boundarie
29. mer C Huang C Yang L Wylie B Coan M 2001 National Land Cover Database USGS Eros Data Center Digital Media accessed December 2008 Justice C Giglio L Korontzi S Owens J Morisette J Roy D Descloitres J Alleaume S Petitcolin F and Kaufman Y 2002 The MODIS fire products Remote Sensing of Environment 83 1 amp 2 244 262 Lehner B and Doll P 2004 Development and validation of a global database of lakes reservoirs and wetlands Journal of Hydrology 296 1 22 Rodriguez E C S Morris J E Belz 2006 A global assessment of the SRTM performance Photogrammetry Engineering and Remote Sensing 72 249 260 Salomon J Hodges J Friedl M Schaaf C Strahler A Gao F Schneider A Zhang X El Saleous N Wolfe R 2004 Global Land Water Mask Derived from MODIS Nadir BRDF Adjusted Reflectances NBAR and the MODIS Land Cover Algorithm IEEE Geoscience and Remote Sensing Symposium IGARSS 04 Proceedings Alaska Sept 2004 241 Scambos T A T M Haran M A Fahnestock T H Painter and J Bohlander 2007 MODIS based Mosaic of Antarctica MOA data sets Continent wide surface morphology and snow grain size Remote Sensing of the Environment 111 2 3 242 257 Strabala K 2004 MODIS cloud mask user s guide Retrieved Dec 1 2004 from http cimss ssec wisc edu modis1 pdf CMUSERSGUIDE PDF SWBD 2005 Shuttle Radar Topography Mission Water Body Data set http
30. nal Journal of Digital Earth Scheduled for publication December 2009 A New Global Raster Water Mask at 250 meter Resolution M L Carroll J R Townshend C M DiMiceli P Noojipady R A Sohlberg Department of Geography University of Maryland College Park MD Corresponding author Abstract Accurate depiction of the land and water is critical for the production of land surface parameters from remote sensing data products Certain parameters including the land surface temperature active fires and surface reflectance can be processed differently when the underlying surface is water as compared to land Substantial errors in the underlying water mask can then pervade into these products and any products created from them Historically many global databases have been created to depict global surface water These databases still fall short of the current needs of the terrestrial remote sensing community working at 250m spatial resolution The most recent attempt to address the problem uses the Shuttle Radar Topography Mission SRTM data set to create the SRTM Water Body Data set SWBD 2005 The SWBD represents a good first step but still requires additional work to expand the spatial coverage to include the whole globe and to address some erroneous discontinuities in major river networks To address this issue a new water mask product has been created using the SWBD in combination with MODIS 250m data to create a complete global
31. ows v00 v03 were all tested in this manner and obvious discontinuities were determined to be rare and were resolved by additional discrete mapping of the specific local region using decision tree classification Validation efforts are discussed in the validation section section 4 of this text 3 Results The new 250m water mask is a global raster data set in the Sinusoidal projection subset into tiles matching the MODIS tile grid There are 3 discrete values represented e 0 Land ae i Water e 253 Fill outside the projection This dataset is intended to replace the old EOS 1km Land Water mask originally created in the mid 1990 s and updated in 2002 The 2002 update was global except for 80 to 90 N where no data were available at that time and was performed by Boston University Salomon et al 2004 This update solved numerous errors including many misplaced rivers in South America but was limited by the 1km spatial resolution and the inability to solve problems in the far north due to lack of data Figure 6 shows the difference between the new 250m water mask and the old EOS Land Water mask for an area of northern Greenland The old EOS mask seen in 6b is shifted 35km from where the water actually exists the new 250m water mask seen in 6c corrects this issue Insert Figure 6 Joining the SWBD and the MODIS 250m data produces a heterogeneous data set Figure 7a shows part of the Scandinavian Peninsula spanning the 60
32. pefile was converted to a polygon and rasterized such that any data inside the polygon was considered land and anything outside the polygon is considered water This reformatted product is included in the beta release of the new 250m water mask as the land water mask for Antarctica The grounding line is the point at which the ice sheet is still resting on solid rock Scambos et al 2007 The cryospheric community has used this reference in their products for a number of years 2 5 Quality Assurance Data Layer A QA layer was created that shows which data source provided the water pixel For example the area seen in red in figure 4 has a value that is distinct from the area shown in blue designating that it came from a different source in this case MODIS Users can utilize the information in this layer to assist in the determination of the utility of the data Quality assurance was done by opening all tiles and performing a visual inspection Initial success was determined by visual comparison with MODIS 250m spectral data to determine if the water mask features did in fact overlay with known water features The new 250m water mask was found to have good agreement with known water bodies Spatial fidelity between tiles where different sources of data were used was tested by stitching together 4 MODIS tiles along the boundaries This process was repeated in a moving window from left to right across the MODIS tile grid shown in figure 5 The tiles in r
33. r maps for the purpose of representing surface water for climate modelers This is merely a compilation of existing maps most generated prior to 1996 and not updated for existing conditions for example Lake Chad and the Aral Sea are shown at historical extents Additionally the spatial resolution of the raster data set is only 1km The available vector data sets including GLWD and the Digital Chart of the World share a common set of original input files at a scale of 1 1 000 000 These were mostly derived from the US Defense Mapping Agency Operational Navigation Charts Lehner and Doll 2004 ESRI 1992 The latest update to any of the published data is 1992 according to the User s Guide Lehner and Doll 2004 The World Vector Shoreline was derived mostly at 1 250 000 and was a reasonable representation of the coastline at the time but is out of date and does not include interior lakes and rivers Inaccuracies in the location of rivers and coastlines are shared among the GLWD and others like the Digital Chart of the World because they share a common heritage This is particularly apparent in South America where the Tapajos River for example is shifted by as much as 10 kilometers In figure 1 the mask shown in cyan is the original Moderate Resolution Imaging Spectro radiometer MODIS Earth Observing System EOS water mask This mask also shares a common heritage with the vector data sets and exhibits the inaccuracy in location of this river
34. ral African Republic Figure 4 Shows the progression of the gap detection and filling for the SWBD 4a shows the mouth of the Amazon river in Brazil with a major gap in the SWBD 4b the area in red is water derived from MODIS 250m data that is being inserted in the gap detected in 4a 4c shows the finished product with all water in blue leaving a relatively seamless result 1 2 3 4 5 6 7 8 9 hoii 12 13 14 1 21 22 23 24 25 26 27 28 29 30 3 1 32 33 34 35 0 1 2 i Laet L D 5 f s tc mace f PH B 43 4 r et 7 iaa x j ii 4 A 12 A 13 14 4 Figure 5 The global MODIS Sinusoidal tile grid http landweb nascom nasa gov developers sn_tiles sn_bw_10deg html Figure 6 a Shows a composite of MODIS summer imagery for 2003 2007 for northern Greenland near the McKinsey Sea b Shows the composite image with the current Ikm MODIS EOS land water mask overlain in red c Shows the composite image with the new 250m water mask overlain in blue 60 degrees North latitude Figure 7 a overview of the Scandinavian peninsula b 250m resolution view of the old EOS 1km water mask c the new 250m water mask using the SWBD below 60 N and MODIS 250m data above C Coincident Land f lakes in Boreal Canada west of Hudson Bay as compared to the
35. rocesses DAAC and also available in alternate formats through the Global Land Cover Facility GLCF www landcover org 7 Acknowledgements The authors would like to acknowledge the use of the following free data sets in the creation of the new 250m Land Water mask e SWBD 2005 available from ftp eOsrp01u ecs nasa gov srtm version2 SWBD e NLCD 2001 http www mrlc gov nlcd_multizone_map php e NSIDC Ikm DEM Greenland available from ftp sidads colorado edu pub DATASETS DEM nsidc0305_icesat_greenland_dem e MOA available from ftp sidads colorado edu pub DATASETS MOA coastlinesA This work was funded under Grant Cooperative Agreement Number NNXO8AT97A Appendix Last Data Set Author Update Resolution Issues Type Global Self Consistent Hierarchical High National Resolution Shoreline Geophysical 200 m 1 km No rivers coasts and Database Data Center 2004 5 km 25 km_ inland lakes only Vector National Based on survey data World Vector Geospatial locational accuracy Shoreline Agency 2004 100 m varies by region Vector Regionally Accessible Nested Limited spatial Global Shorelines Rainer Feistel 1999 1km resolution Vector Continental Watersheds and River Networks for Use in Regional and Global Hydrologic and Climate Modeling University of 10 km 55 Very coarse spatial Studies Texas at Austin 2000 km 110km resolution US Raster Geological Limited spatial amp HYD
36. s Table 1 have been developed using one of two basic approaches In the vector based approach shorelines lake and river boundaries are determined using survey maps This provides a continuous vector around the water body in question In the raster based approach satellite imagery is used to determine the presence of water primarily through spectral classification The former approach results in a continuous representation of the land water boundary but is limited by the quality of the underlying survey data These data have been collected by many different organizations with varying techniques and quality of observations The latter approach usually reliably depicts larger water bodies but is compromised by drainage line discontinuities where the width of the river is smaller than the sensor s spatial resolution or when the water signal is mixed with that of adjacent vegetation Additionally spectral classification requires unobscured observation of the ground surface Areas such as the tropics with frequent and dense cloud cover can be difficult to depict Insert Table 1 The following examples illustrate the shortcomings of currently available surface water data sets 1 In Streams and Water Bodies of the United States 2002 produced by the United States Geological Survey USGS tributaries of the Ohio River are not included along the northeastern border between Kentucky and Ohio as well as the entire border between West Virginia and O
37. s constraint helps avoid problems of connecting lakes Noo Table 2 Description of gap detection and filling algorithm Insert Figure 4 Since the SRTM data were collected over a short period in February 2000 the MODIS data used for gap filling was chosen from years 2000 and 2001 in order to keep temporal consistency with water bodies that experience change over time The SWBD did not provide coverage between 54 S to 60 S however there is essentially no land surface in this area There are a total of 6 MODIS tiles with land in them in this latitudinal belt and there is only the southern part of the South Sandwich Islands that are not included in the SWBD in tile hl6v14 These islands were mapped using MODIS 250m data 2 2 Area from 60 to 80 N MOD44C 250m 16 day composites are also available for areas between 60 and 90 N where the SWBD is not available These data were used to create a new 250m resolution land water mask The data were classified using regression tree classification Breiman et al 1984 MODIS data are provided in standard subsets 10 square called tiles These tiles form a grid that is 36 tiles wide referred to as horizontal and shortened to h in tile ID s and 18 tiles 66 99 high referred to as vertical and shortened to v in tile ID s see figure 5 To find a tile ID one needs to cross reference the h or horizontal with the v or vertical Numbering in the grid be
38. uity where appropriate Daily MODIS data was characterized for 3 years of data 2000 2003 to generate a likelihood of water Areas that were detected as discontinuous but showed a high likelihood of water in MODIS were reclassified as water For areas north of 60 N the MODIS likelihood of water was used with some additional training and clarification using Landsat and ASTER data Data sets Used e SRTM Water Body Dataset SWBD 2005 accessed July 2005 e MOD44C 2008 UMD internal data set e Canadian Forest Service and Canadian Space Agency joint project Earth Observation for Sustainable Development of Forests EOSD is the production of a land cover map of the forested area of Canada http www4 saforah org eosdlcp nts_prov html accessed Oct 30 2008 e NLCD2001 for Alaska http www mrlc gov nlcd_multizone_map php accessed Jul 21 2008 e DiMarzio J A Brenner R Schutz C A Shuman and H J Zwally 2007 GLAS ICESat 1 km laser altimetry digital elevation model of Greenland Boulder Colorado USA National Snow and Ice Data Center Digital media File format File format is plain binary with 0 header bytes The MODIS Sinusoidal tile grid was used for compatibility with standard MODIS products Spatial resolution is 231 65635m and each tile is square 4800x4800 pixels A metadata file is associated with each data file and contains georeferencing information as well as legend production and citation information Referen
39. usoidal projection subset into tiles matching the MODIS tile grid There are 3 discrete values represented e 0 Land e 1 Water e 253 Fill outside the projection This dataset is intended to replace the old EOS 1km Land Water mask originally created in the mid 1990 s and updated in 2002 The 2002 update was performed by Boston University Salomon et al 200 and solved numerous errors including many misplaced rivers in South America The update was however limited to available data and hence no update was possible for 80 to 90 N because no appropriate MODIS data were available to them at that time Dramatic improvement is seen in this region in the new 250m water mask and represents an update to the original EOS Land Water mask from the 1990 s The UMD Global 250 meter Land Water Mask has been generated using data from the Shuttle Radar Topography Mission SRTM MODIS data from Terra and Aqua instruments Landsat and ASTER The base product between 60 N and 60 S was the SRTM Water Body Dataset SWBD which was converted from vector to raster 90m spatial resolution projected Sinusoidal projection and aggregated to 250m spatial resolution This data set was then subset into tiles using the MODIS tile grid and each tile was visually inspected for obvious errors Each tile was passed through a custom algorithm which detects discontinuities in water bodies Where discontinuities were identified MODIS 250m data was used to fill in the discontin
Download Pdf Manuals
Related Search