森林火灾预测模型研究

Study on Forest Fire Prediction Model

近年来,随着全球温室效应加剧、气候变暖和人为活动增加,造成全球森林火灾多发频发,给森林资源和人民生命财产安全造成严重威胁。如能有效预测森林火灾的发生规律,采取有效措施加以预防,将会大大减少森林火灾所造成的损失。本研究总结了几类国内外森林火灾发生预测模型和发生次数预测模型,对比各类模型的优缺点,分析了森林火灾预测模型的发展前景,为国内相关模型的研究和开发提供参考。结果发现:(1)在森林火灾发生预测模型中,机器学习预测方法比广义线性回归模型的拟合效果更好,考虑到空间异质性的模型以及多因子模型预测精度更高,在进行森林火灾预测研究时可优先选择GWLR和RF两种模型;(2)在森林火灾次数预测模型中,GM(1,1)模型预测拟合效果较好,但对数据要求更高,ZINB和Hurdle模型整体上要比Poisson回归模型和NB回归模型预测拟合效果好。中国目前关于森林火灾预测模型的研究主要集中在东北、西南和东南等地区,且没有可广泛适用的模型,还需进一步深入研究。

With the intensification of global greenhouse effect,climate warming and human activities,global forest fires have occurred frequently in recent years,posing a serious threat to forest resources and the safety of people s lives and property.If the occurrence pattern of forest fire can be effectively predicted,so that the sufficient measures can be taken,then the loss will be greatly reduced.This study summarizes several types of forest fire occurrence prediction models and occurrence frequency prediction models domestic and abroad,compares the advantages and disadvantages of various models,and analyzes the development prospect of forest fire prediction models,in order to provide reference for the research and development of relevant models in China.The results show that,(1)in the forest fire prediction model,the fitting effect of machine learning prediction method is better than that of generalized linear regression model.The prediction accuracy of the model considering spatial heterogeneity and multi factor model is higher.GWLR and RF models can be selected first in the research of forest fire prediction;(2)Among the forest fire frequency prediction models,GM(1,1)model has better prediction fitting effect,however it has higher requirements for data.ZINB and hurdle models have better prediction fitting effect than Poisson regression model and Nb regression model generally.Research on forest fire prediction model in China nowadays is mainly concentrated in Northeast,southwest and southeast regions,and there is no widely applicable model,which needs further research.