Spatial estimation of soil moisture using synthetic aperture radar in Alaska

Abstract A spatially distributed Model of Arctic Thermal and Hydrologic processes (MATH) has been developed. One of the attributes of this model is the spatial and temporal prediction of soil moisture in the active layer. The spatially distributed output from this model required verification data obtained through remote sensing to assess performance at the watershed scale independently. Therefore, a neural network was trained to predict soil moisture contents near the ground surface. The input to train the neural network is synthetic aperture radar (SAR) pixel value, and field measurements of soil moisture, and vegetation, which were used as a surrogate for surface roughness. Once the network was trained, soil moisture predictions were made based on SAR pixel value and vegetation. These results were then used for comparison with results from the hydrologic model. The quality of neural network input was less than anticipated. Our digital elevation model (DEM) was not of high enough resolution to allow exact coregistration with soil moisture measurements; therefore, the statistical correlations were not as good as hoped. However, the spatial pattern of the SAR derived soil moisture contents compares favorably with the hydrologic MATH model results. Primary surface parameters that effect SAR include topography, surface roughness, vegetation cover and soil texture. Single parameters that are considered to influence SAR include incident angle of the radar, polarization of the radiation, signal strength and returning signal integration, to name a few. These factors influence the reflectance, but if one adequately quantifies the influences of terrain and roughness, it is considered possible to extract information on soil moisture from SAR imagery analysis and in turn use SAR imagery to validate hydrologic models.