[Prediction of Soil Salinity and Analysis of Influencing Factors in Coastal Plains Based on Geographically Weighted Random Forests].

Abstract

Accurate monitoring of the spatial distribution characteristics of soil salinization and its influencing factors is crucial for combating soil degradation and ensuring global food security. Although studies have been conducted using machine learning to predict soil salinization, local modeling studies incorporating spatial information are still limited. Meanwhile, selecting influencing factors from a global perspective to develop precise prevention and control measures for the region is difficult. Therefore, taking the coastal plain of Hebei Province as the study area, a soil salinization prediction model based on geographically weighted regression （GWR）, random forest regression （RF）, and geographically weighted random forest regression （GWRF） was constructed by using multi-source data such as climate, topography, and hydrology, salinity index, vegetation index, and soil moisture index. The predictive performance of each model was systematically compared, and the variability of environmental variables in explaining the spatial variability of salinization was explored. The results showed that： ① The GWRF model was the best in predicting the spatial characteristics of soil salinization in the coastal area （R²=0.82, RMSE=0.10 g·kg^-1, MAE=0.06 g·kg^-1）. ② The degree of soil salinization in the coastal plain of Hebei Province increased from the inland to the coastal area, with soil salinization being the most severe in the eastern part of the coastal plain. ③ Significant differences were observed in the spatial distribution of the importance of different environmental variables. Overall, climate （mean annual precipitation and evapotranspiration） and depth to groundwater were important factors in predicting soil salinization in the coastal plain. This study provides a new perspective for the prediction and analysis of soil salinization in the coastal zone and also provides a scientific basis for regional ecological planning.