How GLARS DSMs Were Made and How to Use Them
Overview
Among the unique demonstrated capabilities and products developed by the GLARS partenrs are high-resolution photogrammetric Digital Surface Models (DSMs) for the Great Lakes Basin. The individual DSM “strips” can be used to provide measurements of elevation change through time. The mosaic products offer a seamless, high-resolution reference model to serve as the foundation for further geospatial analysis and modeling.
GLARS DSM strips are constructed from sub-meter, stereoscopic satellite imagery collected by MAXAR’s Worldview satellite constellation. The imagery is provided at no cost to SharedGeo through the Polar Geospatial Center’s Nextview license. Stereoscopic images were processed into DSM strips using the open-source Surface Extraction from TIN-based Search-space Minimization (SETSM) software, developed by M.J. Noh and Ian Howat at The Ohio State University. The Blue Waters supercomputer located at the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign provided compute resources and invaluable systems support.
National Science Foundation funding supported the significant compute resources and development of the production pipeline and algorithms at Polar Geospatial Center through earlier projects including ArcticDEM and REMA.
Output data is produced by SharedGeo with long term distribution anticipated to handled by the NASA Commercial SmallSat Data Acquisition Program (CSDAP) as time-stamped, 2-meter resolution raster elevation files in 32-bit GeoTIFF format. In addition to the time-stamped DSM strip files, mosaic files are assembled from the registered and aligned DSM strips and tiled for distribution.
Source Data
GLARS DSM data is constructed from in-track and cross-track high-resolution (~0.5 meter) imagery acquired by the DigitalGlobe (MAXAR) constellation of optical imaging satellites and licensed through the NGA NextView contract.
Most GLARS DSM data was generated from the panchromatic bands of the WorldView-1, WorldView-2, and WorldView-3 satellites. A small percentage of data was generated from the GeoEye-1 satellite sensor. Many of the stereo pairs were collected in an “in-track” mode where the two images are collected in the same satellite pass only 1-2 minutes apart. Some pairs were identified as incidental “cross-track” pairs when their satellite collection geometry and time separation were suitable for DSM extraction.
Data Characteristics
Digital Surface Model
GLARS DSM products portray first-return elevation values that include vegetation, tree canopy, buildings, and other man-made surface features for the Great Lakes. Exercise caution when using GLARS DSM data for applications that are better served using a bare-earth Digital Terrain Model (DTM) because results may be inaccurate or otherwise misleading.
Interpolation & Filtering
All DSM products derived from SETSM software come from a TIN-based model and pixels are given data values as the TIN is written to a raster. Only edge artifacts are filtered out of the strip DEM data. Errors due to clouds or water may be identified by an additional bitmask file. More information on using those masks is below. Mosaic compilation uses a water mask derived from the Global Surface Water dataset or other more accurate sources when available to remove large water bodies.
Limitations
As with any optical imagery-derived elevation product, void areas or artifacts may appear where cloud cover, shadows, and unfrozen water bodies exist in the source imagery, or in regions of low radiometric contrast where pixel correlation cannot be resolved by the software. Errors in the water mask used for mosaicking will also result in areas of poor data quality.
Considerations
GLARS DSM mosaics have been assembled as median surface elevation values from source imagery collected over a period of several years and may include data collected throughout multiple seasons. Users should not assume that DSM data represent snow-free, leaf-off, or other temporally variable conditions, unless using a mosaic specifically noted for that purpose. Every effort has been made to introduce the best-available source images into the final product, notwithstanding the challenges of producing a synoptic dataset where source material from a single season or year is unavailable. Further imagery acquisition could improve the quality of GLARS DSM mosaics and their ability to detect change through pixel value variability as the temporal depth of the collection increases.
Mosaics are made using the median value of individual pixels, to focus on the robust central tendency and reduce influence of outliers. Median Absolute Deviation (MAD) at each pixel is also available, and shows the spread of values between DSMs strips collected at different times. Broader spread indicates likelihood of actual change, rather than image error.
Corrections
If you find any errors in the presentation of the GLARS DSM datasets, have questions, or need additional guidance, please contact SharedGeo.
Computer time provided through a Blue Waters Compute Allocation. DSMs produced using data from DigitalGlobe.