森林工程2026，Vol.42Issue(3)：439-451,13.DOI:10.7525/j.issn.1006-8023.2026.03.001
基于地理加权逻辑回归模型的松材线虫病发生风险预测
Risk Prediction of Pine Wilt Disease Based on Geographically Weighted Logistic Regression Model
摘要
Abstract
This study aims to explore the spatial distribution characteristics of pine wilt disease in nine cities of eastern Liaoning Province(including Shenyang,Dalian,Anshan,Fushun,Benxi,Dandong,Yingkou,Liaoyang,and Tiel-ing),construct a prediction model for pine wilt disease risk areas using the geographically weighted Logistic regression(GWLR)model,and provide a scientific basis for the prevention and control of pine wilt disease.Based on the compart-ments data with pine wilt disease information of Liaoning Province in 2023,kernel density analysis was used to character-ize the spatial distribution of pine wilt disease,and thirty-two variables,including meteorological,topographic,vegeta-tion index,and social factors,were considered.Key factors for the prediction model were screened using Spearman′s correlation test,bidirectional stepwise regression,and variance inflation factor(VIF)analysis.Logistic regression and GWLR models were constructed to predict pine wilt disease,their prediction accuracies were compared and the spatial distribution of GWLR coefficients for different independent variables were analyzed.Pine wilt disease outbreaks in Liaon-ing Province were primarily concentrated around Fushun City.Five influential factors were key factors in the prediction model of pine wilt disease in nine cities of eastern Liaoning Province,including the distance from compartments to roads,distance from compartments to railways,annual average temperature,growing-season drought index,and soil moisture.The GWLR model significantly outperformed the traditional Logistic regression model,with an R² improvement of 0.42,a reduction of 0.14 in root mean square error(RMSE)and 0.10 in mean absolute error(MAE).The risk pre-diction showed high accuracy using GWLR:overall accuracy(OA)of 90.41%,producer′s accuracy(PA)of 95.34%and user′s accuracy(UA)of 86.79%for risk areas,and PA of 85.49%and UA of 94.83%for non-risk areas.The high-risk zones were concentrated in Fushun City,southeastern Tieling City,and northwestern Benxi City.Pine wilt disease in eastern Liaoning′s nine cities was predominantly distributed around Fushun.The GWLR model effectively captured the spatial heterogeneity of the disease,with significant spatial variations in the coefficients of independent variables.The distances from compartments to roads and railways exhibited obvious spatial heterogeneity in affecting disease inci-dence,while annual average temperature promoted disease incidence in most areas.This study provides theoretical sup-port for exploring the transmission patterns of pine wilt disease and formulating precise prevention strategies.关键词
松材线虫病/风险区/地理加权逻辑回归/Logistic回归/空间分布/影响因子分析/空间异质性/核密度分析Key words
Pine wilt disease/risk area/geographic weighted logistic regression(GWLR)/Logistic regression/spatial distribution/influencing factors analysis/spatial heterogeneity/kernel density analysis分类
农业科技引用本文复制引用
崔东阳,张智洋,程邦洲,徐钰,甄贞..基于地理加权逻辑回归模型的松材线虫病发生风险预测[J].森林工程,2026,42(3):439-451,13.基金项目
国家自然科学基金面上项目(32071677). （32071677）
