6) Linking multiyear maxima with plot-scale measurements: The extracted maxima of VIs and NonVIs were coupled with their correspondingly averaged plot-level EM38 readings or laboratory-measured soil electrical conductivity using SYSTAT, a software for statistical analysis and modeling, for salinity model development using multiple linear regression analysis at the confidence level of 95%. A positive correlation between salinity and LST, PCs and TCB, and a negative correlation between salinity and different VIs, especially GDVI and NDVI, were observed. Two types of salinity models were obtained: a) specific salinity models for vegetated and nonvegetated areas resulted from multiple linear regression modeling that was applied to two groups of samples located in vegetated and nonvegetated areas and b) integrated salinity models in which all samples in the same pilot site were input for modeling but vegetated and nonvegetated areas were separately treated.
7) Evaluation and application of the salinity models: To understand whether the models obtained are operational, the specific and integrated models were, respectively, applied back to the maxima of VIs and NonVIs of the period 2009–2012 to produce local-scale salinity maps. These maps were evaluated against the ground-measured data by linear regression model [29], [34]. If the agreement between the measured and predicted salinity is ≥80%, the models developed are considered operational at local-scale and the salinity maps are reliable.