文献类型：期刊文献

中文题名：基于贝叶斯-零膨胀负二项模型的森林火灾发生预测研究

英文题名：Predicting forest fire occurrence based on zero-inflated negative binomial model using Bayesian method

作者：肖云丹[1] 纪平[1]

第一作者：肖云丹

机构：[1]中国林业科学研究院资源信息研究所,北京100091

年份：2021

卷号：41

期号：5

起止页码：49-56

中文期刊名：中南林业科技大学学报

外文期刊名：Journal of Central South University of Forestry & Technology

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD_E2021_2022】；

基金：中央级公益性科研院所基本科研业务费专项(CAFYBB2017QA010);国家科技基础条件数据平台项目(2005DKA32200)。

语种：中文

中文关键词：贝叶斯估计;森林火灾;气象因子;负二项回归模型;零膨胀负二项模型

外文关键词：Bayesian method;forest fire;meteorological variables;negative binomial model;zero-inflated negative binomial model

分类号：S762.2

摘要：【目的】随着全球气候变化,极端天气频繁出现,发生森林火灾的可能性也显著提高。在过去的森林火灾发生预测研究中并没有考虑林火发生与气象因子关系模型的不确定性。然而,在全球气候变化背景下,要充分考虑模型的不确定性,科学分析气象因子与森林火灾发生次数的关系,以更好地预测森林火灾的发生,为管理部门的预防工作提供参考依据。【方法】以贵州黔南布依族苗族自治州为研究区域,通过研究森林火灾的发生次数和火险天气的气象变量,基于有效分析模型不确定性且提高模型预测可靠性的贝叶斯法,采用零膨胀负二项和负二项回归两种模型模拟森林火灾发生次数,研究分析发生森林火灾次数与气象因子之间的关系以及评价模型的不确定性。【结果】由零膨胀负二项模型模拟的贝叶斯估计模型可知,森林火灾发生数随着月最小相对湿度的增大而减小,随着月最大风速的增大而增加。贝叶斯法(均方根误差RMSE=5.5569)比传统法(RMSE=5.5776)在火灾发生数拟合方面精度更高。【结论】零膨胀负二项模型由于建模过程中能够很好地对火灾零次发生数据进行分析拟合,模拟精度要比负二项模型好。此外,利用贝叶斯法估计森林火灾发生模型能够很好地提高模型预测的可靠性,并且更精确地评价森林火灾发生模型的不确定性。
【Objective】With the global climate change,the frequency of extreme weather events is increasing and the possibility of forest fire has increased significantly.In the past studies,the uncertainty of the relationship model between forest fire occurrence and climate factors was not considered.Considering the uncertainty of the relationship model between forest fire occurrence and climatic factors can provide important reference for management departments to do prevention work under the background of climate change.【Method】Based on the forest fire occurrence and meteorological variables in spring fireproofing period in Qiannan area,negative binomial model and zero-inflated negative binomial model were established respectively to predict the forest fire occurrences by Bayesian method.The relationship between the frequency of forest fires and climate factors was studied and the uncertainty of the evaluation model was analyzed.【Result】According to the model,the forest fires occurrence decreases with the increase of the minimum relative humidity monthly.On the contrary,it increases with the increase of the maximum wind speed monthly.The results show that the accuracy of the Bayesian method(RMSE=5.5569)is higher than that of traditional method(RMSE=5.5776).【Conclusion】The results showed that ZINB model was better than NB model.Additionally,the model reliability using Bayesian method was better than using classical method.The advantage of Bayesian method is prior information and sample information.Moreover,the Bayesian method could well improve the model prediction reliability and evaluate the uncertainty of forest fire occurrence model more accurately.