农业科学研究2025，Vol.46Issue(4)：20-29,10.DOI:10.13907/j.cnki.nykxyj.2025.04.003
内蒙古河套灌区土壤盐分反演的可解释机器学习模型构建
Construction of an Interpretable Machine Learning Model for Soil Salinity Inversion in the Hetao Irrigation District of Inner Mongolia
摘要
Abstract
Soil salinization in the Hetao Irrigation District of Inner Mongolia severely hampers the sustainable development of local agriculture.Accurately understanding the spatial distribution of soil salinity is crucial for improving agricultural productivity and pro-tecting the ecological environment.This study utilized multi-source remote sensing data,including Sentinel-2 image-derived indices(salinity indices:S1,S2,S3,SAIO,CRSI,SI1,SI2,SI3,SI4,SI5,and SI_T;vegetation indices:SAVI,NDVI,RDVI,GNDVI,TVI,DVI,NDGI,and ENDVI),and SRTM topographic factors(Aspect,Elevation,LS_Factor,Roughness,TWI,Slope,TPI,SPI).It initially screened feature variables using Pearson correlation analysis and constructed three soil salinity inver-sion models:Random Forest(RF),Light Gradient Boosting Machine(LGBM),and Extreme Gradient Boosting(XGB).Fur-thermore,it applied the Shapley Additive Explanations(SHAP)method to elucidate the contributions and directions of feature vari-ables on model predictions.The results indicate:① Pearson correlation analysis shows that the top ten feature variables most strongly correlated with the target variables are Elevation,S1,S2,S3,DVI,SI_T,NDGI,CRSI,SAVI,and TVI;② The XGB model performs the best in this study(training set R2:0.724 9;validation set R2:0.427 3),although further optimization is necessary to enhance its generalization ability;③ Different models exhibit varying capacities for capturing feature variables,with the terrain factor Elevation being the most significant contributor to the RF and XGB models,while the salinity index CRSI is the most significant contributor to the LGBM model.This study provides theoretical support for monitoring soil salinization in the Hetao Irrigation District and offers important insights for the precise prevention and control of local soil salinization and ecological protection.关键词
盐渍土/遥感数据/可解释机器学习/盐分反演Key words
saline soil/remote sensing data/interpretable machine learning/salt inversion分类
农业科技引用本文复制引用
马东祥,姜宇龙,王相平,王志鹏,李源奕,姚荣江,谢文萍,杨劲松..内蒙古河套灌区土壤盐分反演的可解释机器学习模型构建[J].农业科学研究,2025,46(4):20-29,10.基金项目
内蒙古自治区"科技兴蒙"行动重点专项(NMKJXM202401-01) （NMKJXM202401-01）
