|Thoma, D. - NATIONAL PARK SERVICE|
|Bryant, R. - STATICAL RESEARCH INC.|
|Rahman, M. - SASKATCHEWAN ENVIR. DIV|
|Holifield Collins, Chandra|
|Noriega, R. - UNIV. OF ARIZONA|
|Osman, I. - UNIV. OF ARIZONA|
|Skirvin, S. - UNIV. OF ARIZONA|
|Tischler, M. - US ARMY|
|Peters-Lidard, C. - NASA|
Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: September 20, 2007
Publication Date: January 14, 2008
Citation: Thoma, D., Moran, M.S., Bryant, R., Rahman, M., Holifield Collins, C.D., Keefer, T.O., Noriega, R., Osman, I., Skirvin, S., Tischler, M., Bosch, D.D., Starks, P.J., Peters-Lidard, C. 2008. Appropriate scale of soil moisture retrieval from high-resolution radar imagery for bare and minimally vegetated soils. Remote Sensing of Environment. 112:403-414. Interpretive Summary: The distribution of near-surface soil moisture is an important factor in a wide range of decision making related to flood prediction, prescribed burns, animal stocking rates, rangeland health and off road trafficability. It is possible to map surface soil moisture using satellite sensors that measure the radar backscatter from a non-vegetated soil surface. Studies have reported mixed results from this satellite-based approach because it appears that the accuracy depends on the scale of the map. This study offers a protocol for determining the minimum spatial resolution for soil moisture retrieval from radar imagery with known confidence. It is an image-based approach that doesn’t require any a-priori information about the surface. This work explained the poor performance of soil moisture retrieval approaches using SAR imagery at site scales and the good performance at watershed scales. This opens the door to use of soil moisture maps for many resource management applications that require both fine spatial resolution and high input accuracy.
Technical Abstract: This research investigates the appropriate scale for watershed averaged and site specific soil moisture retrieval from high resolution radar imagery. The first approach involved filtering backscatter for input to a retrieval model that was compared against field measures of soil moisture. The second approach involved spatial averaging filtered imagery in an image-based statistical technique to determine the best spatial scale for site-specific soil moisture retrieval. Field soil moisture was measured at 1225 m2 sites in three watersheds commensurate with 6.25 m resolution Radarsat image acquisition. Imagery was block median filtered at multiple levels to minimize speckle effects before retrieval of soil moisture. Results indicated that 5x5 pixel block median filter was the optimum filter level to reduce speckle effects and preserve spatial resolution for watershed-averaged estimates of soil moisture. However, median filtering alone could not be used to achieve acceptable accuracy for soil moisture retrieval on a site-specific basis. Therefore, spatial averaging of median filtered imagery was used to generate backscatter estimates with known confidence for soil moisture retrieval. This combined approach of filtering and averaging was demonstrated at watersheds located in Arizona (AZ), Oklahoma (OK) and Georgia (GA). The optimum ground resolution for the three watersheds was obtained when 3 x 3 block median filtered imagery was averaged to resolutions of 210 m, 252 m, and 294 m for AZ, OK and GA watersheds, respectively. This statistical approach does not rely on ground verification of soil moisture for validation and only requires a satellite image of the site. When applied at other locations, the resulting optimum ground resolution will depend on the spatial distribution of land surface features that affect radar backscatter. This work offers insight into the accuracy of soil moisture retrieval, and an operational approach to determine the optimal spatial resolution for the required application accuracy.