Evaluation of digital soil mapping approach for predicting soil fertility parameters—a case study from Karnataka Plateau, India

Detailed maps of soil nutrients are important to identify the areas of fertility constraints and assist farmers on agricultural management measures for better crop productivity. In the present study, we evaluated a digital soil mapping approach for predictions of macronutrients (P2O5, K2O, and S) and micronutrients (Fe, Mn Cu, Zn, and B) along with other soil fertility indicators like soil pH, electrical conductivity (EC), and soil organic carbon in part of northern Karnataka Plateau (1373 km2). Five data mining algorithms, viz., multi-linear regression (MLR), cubist, random forest (RF), support vector machine (SVM), and random forest regression kriging (RFRK), were evaluated using 1297 surface samples (0–30 cm) collected from different land uses of the study area. Terrain attributes, Landsat 8 OLI data, present land use, and climatic datasets like annual average precipitation, maximum temperature, and minimum temperature were used as environmental covariates, and the models were calibrated using 80% of total field observations and validated using 20% of datasets. RFRK model performed well compared to other models for the prediction of soil fertility parameters (R2 = 9–48%). The order of performance for most of the soil parameters was RFRK > RF > SVM > Cubist > MLR. For the micronutrients, RFRK explained 36% of the variation for DTPA extractable Cu, 34% of variation for DTPA extractable Fe, and 23% of variation for DTPA extractable Mn. We prepared soil fertility parameters at 100-m spatial resolution, and the results were used for the calculation of soil fertility index (SFI) based on additive SFI approach. The high-resolution soil nutrient maps and SFI are useful for delineation of fertility-constrained areas and better-targeted fertilizer recommendations for nutrient management.