A COMPARISON OF MACHINE LEARNING MODELS FOR SOIL SALINITY ESTIMATION USING MULTI-SPECTRAL EARTH OBSERVATION DATA

Abstract. Soil salinity, a significant environmental indicator, is considered one of the leading causes land degradation, especially in arid and semi-arid regions. In many cases, this major threat leads to loss arable land, reduces crop productivity, groundwater resources loss, increases economic costs for soil management, ultimately probability erosion. Monitoring salinity distribution degree mapping electrical conductivity (EC) using remote sensing techniques are crucial use management. Salt-effected predominant phenomenon Eshtehard Salt Lake located Alborz, Iran. study, potential Sentinel-2 imagery was investigated monitoring salinity. According satellite's pass, different salt properties were measured 197 samples field data study. Therefore several spectral features, such as satellite band reflectance, indices, vegetation extracted from imagery. To build an optimum machine learning regression model estimation, three models, including Gradient Boost Machine (GBM), Extreme (XGBoost), Random Forest (RF), used. The XGBoostmethod outperformed GBM RF with coefficient determination (R2) more than 76%, Root Mean Square Error (RMSE) about 0.84 dS m−1, Normalized (NRMSE) 0.33 m−1. results demonstrated that integration data, appropriate could provide high-precision maps monitor problem.