Prediction and Mapping of Soil Texture at High Spatial Resolution in a Canal Irrigated Region Using Machine Learning

Abstract

Accurate prediction and mapping of soil texture are essential for sustainable land and water management in irrigated agricultural regions. This study aimed to estimate and map the spatial distribution of soil texture components viz. sand, silt, and clay, across the Cauvery command area in southern Karnataka, India, using geospatial techniques and machine learning. A total of 289 surface soil samples were collected and particle size analysis was performed using the international pipette method. A set of environmental covariates, including spectral indices derived from Sentinel-2 (10 m), terrain attributes, and remote sensing data, were used as predictors in a Digital Soil Mapping framework. The Random Forest algorithm was employed due to its robustness, ability to handle high-dimensional data, and resistance to overfitting. The model exhibited strong calibration accuracy for sand (R² = 0.958) and silt (R² = 0.930), with moderate prediction accuracy during validation, particularly for clay (R² = 0.282). Spatial distribution maps revealed distinct patterns in texture classes, largely governed by topography and depositional processes. Most soils were classified as clay loam, reflecting the region’s pedogenic conditions and irrigation history. This study highlights the potential of combining remote sensing data and machine learning for high-resolution soil texture mapping, offering valuable inputs for precision agriculture and soil health monitoring in canal-irrigated landscapes and helps policy makers for taking decisions.