Predictive Analysis of Soil Textural Fractions Using Entropy and Regression Models in Digital Soil Mapping

Abstract

Digital soil mapping is an innovative and highly efficient technique that extracts valuable insights about soil by analyzing a combination of soil and environmental factors. A recent study conducted in Tamil Nadu demonstrated the accuracy and effectiveness of digital soil mapping in predicting soil properties both qualitatively and quantitatively. Around 440 points of soil database were generated through soil survey and from existing soil resources and the environmental variables were derived from satellite based remote sensing data. Thirty three covariates of soil and environment factors were generated and used as input in the See5 algorithm for predicting the texture class at surface and subsurface level. Out of 33 covariates, 25 and 24 covariates were effectively used for prediction of soil texture with an overall accuracy of 75 and 71.7 per cent and kappa coefficient of 0.65 and 0.58 at surface and subsurface class, respectively. The tangential curvature, profile curvature, geology, geomorphology and land use land cover classification were effectively used for surface soil texture prediction while the maximal curvature and geology were found to have high influence in predicting subsurface soil texture. Cubist model was used for quantitative prediction of sand and clay content using 200 point observations as training data and 160 as validation data. Two rule sets were generated to predict sand and clay separately revealing that green band was highly essential for prediction of sand content followed by red band, NIR band and SWIR band, while geology was the important variable for prediction of clay content. The prediction had an estimation error of 11.9% for sand and 8.6% for clay fraction indicating that the digital soil mapping is an efficient method to derive qualitative and quantitative information about soil properties, making it a promising tool for soil scientists and researchers.