Spatial Prediction of Soil pH Through Machine Learning Algorithm: A Modeling Framework for Regional Soil Fertility Management

Soil pH plays a crucial role in crop production, directly influencing nutrient availability and microbial activity, thereby impacting environmental and ecosystem functions. High-resolution spatial data are essential for sustainable agricultural practices. This study represents the first attempt to generate a spatial distribution prediction map for soil pH across Rajasthan using 893 legacy soil profiles, along with digital auxiliary data and topographic variables. Six machine learning models – Random Forest (RF), Cubist, XGBoost, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) – were employed to predict soil pH at a 90 m resolution. Soil pH values measured in the study ranged from 5.55 to 9.88 at different depths. Among the models, RF demonstrated strong predictive performance, with R2 values of 0.928 for calibration and 0.55 for validation. However, the Cubist model achieved the accuracy, with R2 values of 0.75 and 0.38 for calibration and validation, respectively. Annual precipitation emerged as the most influential covariate for soil pH prediction. The predicted soil pH ranged from 5.5 to 9.8 across Rajasthan, with the spatial distribution aligning well with expected patterns, affirming the efficacy of model selection. However, uncertainty analysis indicated significant uncertainty in 60% of the study area due to limited soil profile data. Overall, the findings suggest that soil pH can be reliably predicted using machine learning techniques, and the resulting high-resolution maps can serve as valuable tools for stakeholders, policymakers, and agricultural practitioners aiming to implement precision farming strategies.