大兴安岭塔河地区林火发生的优势预测模型选择

应用生态学报

大兴安岭塔河地区林火发生的优势预测模型选择

秦凯伦¹,郭福涛^2**,邸雪颖¹,孙龙¹,宋禹辉²,吴瑶³,潘建峰⁴

(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)

出版日期:2014-03-18 发布日期:2014-03-18

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

QIN Kai-lun¹, GUO Fu-tao², DI Xue-ying¹, SUN Long¹, SONG Yu-hui², WU Yao³, PAN Jian-feng^4

(¹College of Forestry, Northeast Forestry University, Harbin 150040, China; ^2Fujian Agriculture and Forestry University, Fuzhou 350002, China; ³Heilongjiang Academy of Forestry, Harbin 150081, China; ⁴Jiangxi Agricultural University, Nanchang 330045, China)

Online:2014-03-18 Published:2014-03-18

摘要/Abstract

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.

Abstract: 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 models (ZIP, PH and NBH) based on the result of AIC and SSR tests. LR test revealed that the negative 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 feature of the study area according to the hypotheses of the two types of models.

秦凯伦1,郭福涛2**,邸雪颖1,孙龙1,宋禹辉2,吴瑶3,潘建峰4. 大兴安岭塔河地区林火发生的优势预测模型选择[J]. 应用生态学报.

QIN Kai-lun1, GUO Fu-tao2, DI Xue-ying1, SUN Long1, SONG Yu-hui2, WU Yao3, PAN Jian-feng4. Selection of advantage prediction model for forest fire occurrence in Tahe, Daxing’an Mountain.[J]. Chinese Journal of Applied Ecology.