Local interpretable soil texture prediction based on modified geographically weighted regression AdaBoost method

Traditional global machine learning models face certain limitations, such as challenges in capturing spatial non stationarity of soil texture and insufficient interpretability. To address these issues, this study proposes an improved interpretability oriented geographically weighted regression adaptive enhancement (IGWRABoost) method and applies it to the spatial prediction of soil texture in Yongqiao District, Suzhou City, Anhui Province, China, which is an important wheat producing area on the Huang Huai Hai Plain. The AdaBoost sub model is constructed by incorporating geographic weighting into the local community. Subsequently, multiple local models underwent Gaussian weighted fusion during the prediction phase to capture the spatial non-stationary relationship between texture and environmental factors. The results showed that the IGWRABoost method achieved the highest R ² and lowest root mean square error (RMSE) and mean absolute error (MAE) values in soil texture prediction. Based on SHAP's global and local interpretations, it is further revealed that the Temperature Vegetation Index (TVDI) is the joint dominant factor for sand and clay, and the Normalized Difference Vegetation Index (NDVI) plays an important role in predicting sand content. The influence and intensity of each covariate vary significantly between different spatial locations. The IGWRABoost SHAP framework not only improves the accuracy of soil texture prediction, but also enhances the interpretability of the model. Therefore, this study provides a methodological reference for digital soil texture mapping and fine management of agricultural water resources in the Huang Huai Hai Plain and other similar agricultural regions.