Assessment of land degradation in different land uses by modeling soil salinity and soil erodibility coupled Vis-NIR spectroscopy and machine learning model

Accurate forecasting of soil salinity (ECe) and erodibility (K-factor) is of importance for food security and effective management for environmental issues. It is important to address land degradation. Hence, this study aims to employ visible and near-infrared (Vis-NIR) to precisely predict soil salinity and erodibility at different types of land use such as rangelands, bare lands, and farmlands, where land degradation is a significant issue. In this way, the multiple regression (MR) and random forest (RF) modelling techniques were used. 150 sampling points were selected to represent the diversity in land uses for soil analysis. At each point, the ECe and the K-factor from the RUSLE model were measured. A portable spectrometer captured spectral reflectance was employed to measure reflectance in the 400 to 2400 nm wavelength range. In this study, spectral reflectance data at each wavelength and combinations of different wavelength reflectance data were used as input variables. The results showed that the RF model, which incorporated combined factors as input variables, surpassed other models in predicting both ECe (RMSE = 4.85 and R² = 0.87) and K-factor (RMSE = 0.014 and R² = 0.60). Additionally, the analysis highlighted improved model performance across different land use types, particularly in farmlands and rangelands. Furthermore, the RF model proved more effective than the MR model in estimating soil erodibility at all land use types. This integrated approach showed the power of integrating Vis-NIR data and machine learning techniques to assess land degradation parameters. These results support the implementation of targeted agricultural practices by policymakers to address these challenges.