林业工程学报2017，Vol.2Issue(1)：135-142,8.DOI:10.13360/j.issn.2096-1359.2017.01.024
广义线性模型在林火发生预报中的应用
Research on generalized linear models applied to forest fire forecast
摘要
Abstract
This paper firstly introduces the application of generalized linear models in forest fire forecast at home and a⁃broad. Then the formulae, parametric estimation methods and hypothesis test methods of six forest⁃fire⁃forecasting models are introduced, including linear regression models, Logistic generalized linear models, Poisson generalized linear models, negative binomial generalized linear models, Zero⁃inflated generalized linear models and Hurdle gener⁃alized linear models. Logistic generalized linear models are mainly used for predicting the probability of forest fire oc⁃currence and the other models are used for estimating the number of forest fireoccurrence. According to the characteris⁃tics of the forest fire frequency data and the previous relevant studies, it can be concluded that Poisson generalized lin⁃ear models, negative binomial generalized linear models, Zero⁃inflated generalized linear models and Hurdle general⁃ized linear models are more suited for estimating the number of forest fire than linear regression models. When the va⁃riance of the forest fire frequency is close to the expected value, Poisson generalized linear models or Zero⁃inflated Poisson generalized linear models can be successfully applied. If the variance of the forest fire frequency is significant⁃ly higher than the expected value, negative binomial generalized linear models and Zero⁃inflated negative binomial generalized linear models can better estimate the number of forest fire. Finally, suggestions are given on the applica⁃tion of the generalized linear models in forest fire forecast:firstly, more independent variables ( the characteristics of forest fuel, terrain, human activities and other factors) should be added to established generalized linear models;sec⁃ondly, the application for forest fire forecast in landscapes layer should be increased;thirdly, the modeling approaches of generalized linear models should be expanded ( generalized linear mixed model and generalized additive models can be utilized) .关键词
广义线性模型/泊松回归模型/负二项分布回归模型/零膨胀模型/栅栏模型Key words
generalized linear models/Poisson models/negative binomial models/Zero-inflated models/hurdle models分类
农业科技引用本文复制引用
张洁,赵浩彦,张民侠,李晨阳,陈戈萍..广义线性模型在林火发生预报中的应用[J].林业工程学报,2017,2(1):135-142,8.基金项目
中央高校基本科研业务费专项资金项目（ LGQN201402，LGYB201616）；南京森林警察学院教研教改项目（ YB14118）。 （）
