Prediction of soil salinity in the Upputeru river estuary catchment, India, using machine learning techniques.

Abstract

Soil salinization is a widespread phenomenon leading to land degradation, particularly in regions with brackish inland aquaculture ponds. However, because of the high geographical and temporal fluctuation, monitoring vast areas provides substantial challenges. This study uses remote sensing data and machine learning techniques to predict soil salinity. Four linear models, namely linear regression, least absolute shrinkage and selection operator (LASSO), ridge, and elastic net regression, and three boosting algorithms, namely XGB regressor, LightGBM, and CatBoost regressor, were used to predict soil salinity. Cross-validation was performed by splitting the data into 30% for model testing and 70% for model training. Multiple metrics such as determination coefficient (R2), root mean square error (RMSE), mean square error (MSE), and mean absolute error (MAE) were used to compare the performances of these algorithms. By comparison, the CatBoost regressor model performed better than the other models in both testing (MAE = 0.42, MSE = 0.28, RMSE = 0.53, R2 = 0.92) and training (MAE = 0.49, MSE = 0.36, RMSE = 0.60, R2 = 0.90) phases. Hence, the CatBoost regressor model was recommended for monitoring soil salinity in India's massive inland aquaculture zone.