云南省森林火灾发生驱动因子分析及火险区域划分

Driving Factors Analysis and Fire Risk Regional Division of Forest Fire in Yunnan Province

森林火灾的有效预测是减少火灾所造成损失的重要手段之一。云南省森林资源丰富且植被覆盖率高,易发生森林火灾,加之平均海拔较高,扑救山火难度大、危险系数高。如能有效构建该区域森林火灾发生预测模型,将为我国西南林区森林火灾预防与管理提供科学参考。提取了云南省1999-2019年森林火点数据进行分析,利用ArcGIS 10.7软件,随机建立与火点形成1:1数据随机点,使数据符合二项分布,对月平均最高气温、月平均降水量、日照时长、海拔等24类驱动因子进行空间信息提取,并用SPSS 25.0软件通过Logistic回归和ROC曲线模型检验方法,分析影响云南省森林火灾发生的主要驱动因子,构建了该地区森林火灾发生预测模型,并验证了该模型适用性,根据R0C曲线坐标确定了最佳阈值,划分了该地区的高火险区域。月平均气压、月平均相对湿度、月平均最高气温、日照时数、极大风速、归一化植被指数值(NDVI)等6类驱动因子,与云南省森林火灾的发生存在显著关系,基于Logistic回归构建的森林火灾发生预测模型准确率可达到82.3%。模型AUC值为0.894,通过计算得出云南省火险划分最佳阈值为0.645。构建的森林火灾发生预测模型具有较高的预测准确率,可为该地区森林火灾发生预测提供参考。云南省的高火险区主要集中在云南中部和北部地区,应进一步加强森林火灾预防和管理。

The effective prediction of forest fire is one of the important means to reduce the loss caused by fire.Yunnan province is rich in forest resources and high in vegetation coverage,which is prone to forest fires.In addition,the average altitude is high,so it is difficult to put out mountain fires and the risk factor is high.If we can effectively build the prediction model of forest fire in this region,it will provide a scientific reference for the prevention and management of forest fire in Southwest China.The forest fire point data of Yunnan province from 1999 to 2019 was extracted and analyzed.The random points forming 1:1 data with the fire points were randomly established by using ArcGIS 10.7 software to make the data conform to the binomial distribution.The spatial information of 24 driving factors such as monthly average maximum temperature,monthly average precipitation,sunshine duration and sea level were extracted.SPSS 25.0 software was used to test the logistic regression and ROC curve model.The main driving factors affecting the occurrence of forest fires in Yunnan province were analyzed,a prediction model of forest fires in this region was constructed,the applicability of the model were verified,the optimal threshold according to the ROC curve coordinates was determined,and the high fire risk areas in this region were divided.Six driving factors,such as monthly average air pressure,monthly average relative humidity,monthly average maximum temperature,sunshine hours,extreme wind speed and normalized vegetation index(NDVI),were significantly related to the occurrence of forest fires in Yunnan province.The accuracy of the prediction model based on logistic regression could reach 82.3%.The AUC value of the model is 0.894,and the best threshold of fire risk classification in Yunnan province is 0.645.The forest fire prediction model has a high prediction accuracy,which can provide a reference for the prediction of forest fire in this area.The high fire risk areas in Yunnan province are mainly concentrated in the central and northern regions of Yunnan province,so forest fire prevention and management should be further strengthened.