大兴安岭塔河地区林火发生的优势预测模型选择  

Selection of advantage prediction model for forest fire occurrence in Tahe,Daxing'an Mountain

作　　者：秦凯伦[1] 郭福涛[2] 邸雪颖[1] 孙龙[1] 宋禹辉[2] 吴瑶 潘建峰[4] 

机构地区：[1]东北林业大学林学院,哈尔滨150040 [2]福建农林大学,福州350002 [3]黑龙江省林业科学研究所,哈尔滨150081 [4]江西农业大学,南昌330045

出　　处：《应用生态学报》2014年第3期731-737,共7页Chinese Journal of Applied Ecology

基　　金："十二五"农村领域国家科技计划项目(2011BAD08B01-03);中央高校基本科研业务费专项资金项目资助

摘　　要：选取在经济学和社会科学领域广泛应用的零膨胀模型(zero-inflated models)和栅栏模型(Hurdle models)对大兴安岭地区林火发生进行模拟,应用赤池准则(AIC)、似然比检验(LR)和模型残差平方和(SSR)对两类共4个回归模型——零膨胀泊松模型(ZIP)、零膨胀负二项模型(ZINB)、栅栏泊松模型(PH)、栅栏负二项模型(NBH)进行拟合分析,最终选取适合此林火发生特性的预测模型.模型的AIC和SSR值表明,ZINB模型对当地林火数据的拟合度最高.运用LR检验对嵌套模型(ZINB与ZIP,NBH与PH)进行检验,结果显示:ZINB和NBH均优于各自的嵌入模型,说明负二项(NB)模型对数据结构中的过度离散现象可以很好地模拟和解释.根据研究区林火实际发生规律和两类不同模型的应用假设条件判断,零膨胀模型更适合塔河地区的林火特性.This study chose zero-inflated model and Hurdle model that have been widely used in economic and social fields to model the fire occurrence in Tahe, Daxing＇ an Mountain. The AIC, LR and SSR were used to compare the models including zero-inflated Poisson model （ZIP） , zero- inflated negative binomial model （ZINB）, Poisson-Hurdle model （PH） and negative Binomial Hurdle （NBH） （ two types, four models in total） so as to determine a better-fit model to predict the local fire occurrence. The results illustrated that ZINB model was superior over the other three mod- els （ZIP, PH and NBH） based on the result of AIC and SSR tests. LR test revealed that the nega- tive binomial distribution was suitable to both the ＂count＂ portion of zero-inflated model and hurdle model. Furthermore, this paper concluded that the zero-inflated model could better fit the fire fea- ture of the study area according to the hypotheses of the two types of models.

关 键 词：林火发生 模型拟合度 栅栏模型 泊松分布 负二项分布 零膨胀模型 

分 类 号：S762.2[农业科学—森林保护学]