Soil salinity prediction using EOS AMI remote sensing satellite data

Abstract: The study was conducted to predict spatial variability of soil salinity in central Iran, using remotely sensed data. Also the efficacy of artificial neural network (ANN) and multivariate regression (MLR) analysis was evaluated in this field. The analysis was based on remote sensing data acquired from the EOS AMI remote sensing satellite. Performance of these two methods was compared to study linear and non-linear relationship between soil reflectance and soil salinity. In MLR analysis, stepwise method was applied and neural network was applied using sensitivity coefficient by arranging inputs through the backward propagation, and then modeling was done. The R2 and RMSE were 0.23 and 0.33 for MLR, and were 0.79 and 0.11 for ANN, respectively. Digital values of bands 6 and 9 were identified as the most important factors in MLR, whereas sum19,band10 were recognized the most important data to predict soil salinity using the ANN as indicated by higher R2 and lower RMSE in the case of ANN than for the MLR method. The results clearly indicated the performance of ANN models to detect existing of non- linear relationship of soil salinity with ASTER data at the study area.

Keywords:

Aster , Artificial neural network , Multiple linier regressions , Soil salinity , Yazd- Ardakan plain