To decrease the number of erroneous data points, model inversion was re-run only for data which fell out of the range of acceptable soil moisture. Re-runs incrementally changed first guesses for LAI . A second rerun incrementally changed LAI initial guesses, but with a first guess for soil moisture set to 0.1 (rather than 0.2). Once a data point fell within the soil moisture range of 0–0.55 m3 m-3, it was no longer included in inversion re-runs. The number of data points remaining out of range,after this process, are provided in Table 5 (calibration runs) and Table 7 (validation runs).
From Table 5, it is observed that all the calibration points for RADARSAT-2 and soybeans (HH–HV, VV–HV
and HH–VV–HV), UAVSAR and corn (HH–HV and VV–HV) and UAVSAR and soybeans (HH–VV–HV) have estimated soil moisture values less than 0.55 m3 m-3. Also, for RADARSAT-2 and corn (HH–VV, HH–HV and HH–VV–HV) and
UAVSAR and soybeans (HH–VV and HH–HV), very few points were out of range. However, for other inversion scenarios,
the number of points with erroneous soil moisture estimates was higher. When soil moisture estimates remained out of range, these data were removed and not used during model calibration or validation.Removal of these data assisted in building robust models which are not skewed by outliers in the data