基于广义线性模型和最大熵模型的黑龙江省林火空间分布模拟

Modeling spatial patterns of forest fire in Heilongjiang Province using Generalized Linear Model and Maximum Entropy Model

林火分布模型是在较大区域上描述林火空间分布的强有力工具,并可以确定影响林火分布的控制因子。本研究基于黑龙江省1996—2006年的历史火烧记录数据,分别采用广义线性模型和最大熵模型分析了地形、人类活动和土地覆被类型等环境控制因子对黑龙江省林火空间分布的影响,并比较了模型预测精度、评价环境变量重要性及预测火点概率分布图等。结果表明:两个模型的预测精度达中等水平,而最大熵模型的预测精度要略高于广义线性模型。总体而言,与人类活动相关的变量是林火分布模型最佳的环境变量,地形变量次之。尽管两个模型在预测精度和环境变量重要性方面都有很大的相似性,但最大熵模型产生的火点概率图空间格局与广义线性模型产生的明显不同。本研究说明,为了更加精确地确定森林火灾发生的热点地区,应该采用不同模型进行比较,或者有选择性地进行组合以产生综合的预测结果,从而为森林防火工作提供更加合理高效的建议。

Forest fire distribution models are the powerful tools to map the spatial patterns of for- est fire in larger scale, and to quantify the relative importance of the major factors controlling for- est fire occurrence. Based on the forest fire ignition data in Heilongjiang Province in 1996-2006, and by using Generalized Linear Model （GLM） and Maximum Entropy Models （Maxent） , this paper analyzed the factors controlling the forest fire occurrence in the Province, including topog- raphy, human activity, and land vegetation type, and compared the modeling accuracy, variable importance, and ignition probability map. Both the GLM and the Maxent had intermediate pre- dictive performance, with the Maxent performed slightly better. Overall, the variables related to human activities were the most important predictors of forest fire ignition locations, followed by topographical variables. Despite the two models had similar modeling accuracy, the ignition prob- ability map generated by Maxent was noticeably different from that generated by GLM. It was suggested that to make a comparison of or to selectively assemble different type models to produce integrated prediction results would be more desirable to more accurately identify the hotspots of forest fire occurrence, and thus, to provide more reasonable and higher efficient comments for forest fire prevention.