| Performance of the Oh et al. (1992), Oh (2004) and Dubois et al. (1995a) soil moisture estimation models was evaluated with three RADARSAT-2 scenes over the Casselman, Ontario area. Root mean square error (RMSE), mean bias error (MBE), correlation coefficients and the number of valid pixels were used to assess the influence of boxcar speckle filter sizes and pre-extraction averaging filtering on the estimates. Highest agreement with the field data was obtained with speckle filter and pre-extraction filter sizes from 3 x 3 to 7 x 7. Bias of the estimates was found to be highly negatively correlated with incidence angle (r2= DAD, 0.15, 0.51). Correlations between the estimated and observed soil moisture contents were not significant for most dates and the correlations with soil properties data were also weak. The effects of tillage method on the relative error of estimates were significant with lower relative errors in low-roughness tillage methods. |
