论文部分内容阅读
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.
The zero-inflated models and Hurdle models, which are widely used in economics and social science, were used to simulate the occurrence of forest fires in the Greater Xing’an Mountains. The Aichi-Chi-square (AIC), likelihood ratio test (LR) (ZINB), Poisson model (PH), negative binomial of the fence Model (NBH) were selected to fit the prediction model of forest fire occurrence characteristics.The AIC and SSR values of the model indicate that ZINB model has the highest fitting degree to the local forest fire data.Using LR test to test the nested model ZINB and ZIP, NBH and PH). The results show that both ZINB and NBH are superior to their respective embedding models, indicating that the negative binomial (NB) model can well simulate and explain the overdispersion in the data structure. According to The actual occurrence of forest fires in the study area and the application of two different types of model assumptions to determine the zero-expansion model is more suitable for the fire characteristics of the Tahe area.