Characterization of the Spatial Variability of In-Situ Soil Moisture Measurements for Upscaling at the Spatial Resolution of RADARSAT-2

This study characterizes the spatial variability of soil moisture measurements using statistical and geostatistical analyses for the transferability of the ground measurements to the scale of the spatial resolution of RADARSAT-2 images. It aims to better interpret synthetic aperture radar backscattering relationships to soil moisture. Coincident with RADARSAT-2 overpasses, soil and crop parameters were measured in July 2008 and August 2009 over two Canadian agricultural sites (Kenaston, Saskatchewan, and Lennoxville, Quebec). The measured soil moisture was used to determine the theoretical semivariogram models that fit the experimental semivariograms. Aninverse correlation is obtained between the soil moisture coefficient of variation (CV) and the range (spatial correlation) of the semivariogram, which can assess the degree of the spatial correlation between the samples of each field. Soil moisture measurements with high values of CV (20%-40%) are correlated within a distance less than 10 m and those with lower CV (10%-20%) are correlated within a larger distance varying between 12 and 46 m. The soil moisture measurements of each field were upscaled to the spatial resolution of RADARSAT-2 images (6 x 5 and 9 x 5 m(2)) using either simple kriging or ordinary kriging. The results were cross validated using the surface scattering component, which is extracted from the Freeman-Durden decomposition applied to fully polarimetric RADARSAT-2 images. They show that the kriging-based soil moisture better represents RADARSAT-2 surface scattering with strong clustered linear regressions (R-2 greater than similar to 0.6, RMSE lower than similar to 0.9 dB, and p-value of slope less than 0.05) than the nonkriged soil moisture samples.