Improved digital mapping of soil texture using the kernel temperature–vegetation dryness index and adaptive boosting

Existing soil texture mapping methods cannot accurately predict soil texture in complex geographical environments. To address this challenge, we propose a method that combines a kernel temperature–vegetation dryness index (kTVDI) with a gradient boosting algorithm to accurately predict the spatial distribution of soil texture. In this study, we collected 399 soil samples collected from Mingguang City in southeast China and made spatial predictions of soil texture based on remote sensing indices such as the kernel normalized difference vegetation index computed from Landsat8 data and topographic attributes computed via digital elevation model as environmental covariates. We validated model performance by mapping the spatial distributions of sand, silt, and clay particle fractions in the city (30-m resolution), using the boosting algorithms adaptive boosting (AdaBoost), gradient boosting decision tree (GBDT), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost). Among the environmental covariates, the kTVDI, digital elevation index, and salinity index have the highest importance values for soil texture prediction. The kTVDI is better for sand and silt prediction (especially sand). When combined with AdaBoost, the kTVDI can effectively improve the accuracy and consistency of the prediction model. Uncertainty analyses showed that the kTVDI was more effective at modeling soil texture in the plains. In summary, we present a new approach for accurately predicting the spatial distribution of soil texture and empirically validate its effectiveness and advantages for practical applications.