Assessment of soil salinity using explainable machine learning methods and Landsat 8 images

• DOI: 10.1016/j.jag.2024.103879
• Publication Date: 2024-05-03T23:10:51Z
• Authors: Samet Aksoy; Elif Sertel; Ribana Roscher; Aysegul Tanik; Nikou Hamzehpour
• Keywords: Physical geography; info:eu-repo/classification/ddc/550; Soil salinity; 550; 01 natural sciences; GB3-5030; Environmental sciences; XAI; Landsat-8 OLI; SHAP; Machine learning; GE1-350; Google Earth Engine; 0105 earth and related environmental sciences

The aim of this study is to comparatively analyze the performance machine learning (ML) algorithms for modeling soil salinity using field-based electrical conductivity (EC) data and Landsat-8 OLI satellite images with derived environmental covariates. We also interpret explain ML models without over-sampling methods Shapley (SHAP) values, an explainable approach that has not yet been utilized estimation tasks as problem. investigate two case areas from western southeastern Lake Urmia Playas (LUP) in Northwest Iran. Our uses 26 covariates, models, namely extreme gradient boosting (XGBoost) random forest (RF), methods: synthetic minority technique (SMOTE) (ROS). Results indicate XGBoost performs better compared RF terms both R2 RMSE. Additionally, visual interpretation maps demonstrated superiority XGBoost. SMOTE produced superior results than ROS no test cases. Finally, SHAP analysis illustrated vegetation indices made a greater contribution prediction West LUP, while visible bands contributed more Southeast LUP Region.