Abstract

Digital soil mapping (DSM) has emerged as a revolutionary approach for understanding soil spatial variability by integrating remote sensing data with machine learning algorithms. This study presents a comprehensive framework for digital soil mapping using multispectral satellite imagery, terrain attributes, and advanced machine learning techniques including Random Forest (RF), Support Vector Machines (SVM), and Artificial Neural Networks (ANN). The research was conducted across a 5000 hectare agricultural region in the Indo-Gangetic Plains, utilizing Sentinel-2 and Landsat-8 imagery combined with field sampling data from 450 georeferenced locations. Results demonstrated that Random Forest achieved the highest accuracy with R² = 0.87 for soil organic carbon prediction, while SVM performed best for soil texture classification with 92% overall accuracy. The integration of topographic variables derived from digital elevation models significantly improved prediction accuracy by 15-20%. This study provides valuable insights into the effectiveness of different machine learning approaches for soil property mapping and establishes a robust methodology for large-scale soil characterization supporting precision agriculture applications.