Rapid Prediction of Soil Electrical Conductivity in the Middle Indo-Gangetic Plains of India

Salts in the root zone have high spatial variability, change rapidly and adversely affect soil quality and crop productivity. In contrast to the time-intensive traditional methods for measuring electrical conductivity (EC), visible-near infrared (Vis-NIR) and mid-infrared (MIR) spectroscopy provide faster alternatives that can assist in creating strategies to reduce negative impacts on soil and plants. Soils were collected from the Indo-Gangetic Plains and analysed for EC_1:2.5 using conventional method. There was a wide variation in EC measured by the conventional method. So Partial Least Squares Regression (PLSR) was used to predict soil EC from spectral data, with the data divided into calibration (70%) and validation (30%) datasets. The partial least square regression (PLSR), random forest (RF), support vector regression (SVR) and multivariate adaptive regression splines (MARS) both in Vis-NIR and MIR region during calibration. The predictive performance of PLSR, RF, SVR, and MARS models for EC_1:2.5 in the Vis-NIR range showed PLSR as the best model (R² = 0.84, RMSE = 0.21, RPD = 2.44). In the MIR range, RF was considered fairly good (R² = 0.52, RMSE = 0.20, RPD = 1.43). Vis-NIR spectroscopy with PLSR algorithm predicted EC better than MIR spectroscopy and would be the method of choice for rapid estimation and prediction of EC in the study region.