Soil Surface Salinity Prediction Using ASTER Data: Comparing Statistical and Geostatistical Models

This study was conducted to evaluate the performance of univariate spatial (ordinary kriging- OK), hybrid/multivariate geostatistical methods (regression-kriging- RK, Co-kriging- CK) with multivariate linear regression (MLR) in incorporation with ASTER data in order to predict the spatial variability of surface soil salinity in an arid area in northern Iran. The primary attributes were obtained from grid soil sampling with nested-systematic pattern of 169 samples and the secondary information extracted from spectral data of ASTER satellite images. The principal component analysis, NDVI and some suitable ratioing bands were applied to generate new arithmetic bands. According to validation based RMSE and ME calculated by a validation data set, the predictions for soil salinity were found ASTERmultivariate ASTERmultivariate to be the best and varied in the following order: RK > REG > Co-kriging ASTER> kriging. Overall, this comparative study demonstrated that RK approach was a better predicator than other selected methods to predict spatial variability of soil salinity. The overall results confirmed that using ancillary variables such as remotely sensed data, the accuracy of spatial prediction can further improved