Research article: Spatial Prediction of Soil pH Values in Hunan Province Based on Multi-source Environmental Variables and Machine Learning Models.

Abstract

This study aims to use multi-source environmental variables and machine learning models to spatially predict the soil pH values in Hunan Province, China. The study area is a main rice producing region of China, also known for its warm and humid climate, as well as its diverse soil types. The research employs three machine learning models: Random Forest(RF), Support Vector Machine(SVM), and Gradient Boosting Regressor Trees(GBRT), in conjunction with field soil sampling and remote sensing data, to predict soil pH values. The predictions are compared with the traditional Ordinary Kriging (OK) model. Through model evaluation, the best model is selected for spatial distribution prediction of soil pH, and SHAP analysis is used for importance analysis to determine the main factors affecting the pH of the red soil in the southern region. Our results showed that soils in the southeast and southwest of Hunan exhibited lower pH than those in northern regions, and land use is the most important factor affecting soil pH values in Hunan Province.