Soil Texture and Color Identification Using Artificial Intelligence Algorithm and Satellite Images

The demand for quality and low-cost soil information is growing because of necessity in land use planning and precision agriculture. The aim of this study is soil texture and color estimation using satellite images information as input variables of support vector regression and tree regression. NDVI, PDI, SAVI, MPDI, MSAVI, TCI, TVX, VHI, NVWSI, RVWSI, MVWSI, VCI are satellite indices which related to the region in Azarshahr (East Azarbaijan). Duncan's test at the 5% probability level indicates significant time differences between the indices. There was no significant difference among the average of indices in terms of soil texture diversity. The error criteria RMSE, RRMSE, MAPE and MSE decreasing regard to sand from tree regression to support vector regression was 15.43, 13.33, 16.41 and 28.7%, respectively. Determination of soil texture with soil texture triangle in the validation period indicated the agreement of soil texture between observation and support vector regression. Considering soil texture and color components, RPD statistic increased from tree regression to support vector regression by 12.43%, which indicates the efficiency of support vector regression against tree regression. RMSE, RRMSE and MSE decreasing from multiple linear regression to support vector regression in hue were 76.88, 77.4 and 94.6%, respectively and for tree regression were 72.15, 72.58 and 92.92%, respectively, which is indicative of better performance of two regression models relative to simple regression. Based on various aspects of analysis, support vector regression had better performance for soil color and texture determination than tree regression.