中文摘要：

为分析不同空间抽样方式对逻辑火险建模的影响,以我国东北为研究区,设计3种全局抽样方案和1种分区抽样方案,分别取得训练样本并开发了逻辑火险模型,对模型的因子和因子相对重要性做比较分析.基于4种抽样方案的模拟结果,利用内部检验子集和独立检验子集,从可靠性和区分能力检验模型性能并进行比较.结果表明：①全局模型间差异较小,表明不同的全局抽样方式对建模的影响有限,总体来看,空间平衡抽样略优于空间系统抽样,空间系统抽样略优于空间随机抽样;②分区模型间差异显著,表明不同子区的火灾事件与火环境因子间关系明显不同;③分区建模的性能总体上优于全局建模,但分区建模仍受限于逻辑模型自身的非空间、全局特征;④NDVI是全局尺度分析结果中最重要的火环境因子,高程因子的影响不显著.本研究可为东北地区的火险研究和采用逻辑模型的相关研究提供参考.

英文摘要：

In order to analyze the influences of different spatial sampling programs on logistic fire danger modeling,the northeast China was selected as the study area and three global sampling methods and one zonal sampling pattern were employed to create six training subsets,and then the six training subsets were respectively used to develop six logistic fire danger models.In terms of the included fire-influencing factors and the relative importance of the included factors or types,the six developed models were analyzed and compared.For the four modeling results,deriving from four corresponding spatial sampling programs,evaluated by the inner testing and independent validation,better reliability and discrimination capacity of the four modeling were concluded.In addition,the reliability and discrimination capacity of the four modeling were also compared.The results or conclusions are as follows：i） There exists little difference among the three global models,indicating limited influence of different global sampling programs on logistic fire danger modeling.In summary,the spatially balanced sampling is slightly better than the spatially systematic sampling and the spatially systematic sampling is slightly better than the spatially random sampling.ii） Obvious differences exist among three zonal models,showing that the relationships between fire events and fire-influencing factors are different for different subregions.iii） On the whole,the performance is better for zonal modeling than for global modeling,but the performance of zonal modeling is still limited by the characteristics of being non-spatial and global of the binary logistic model.iv） It can be seen from the results of global scale analysis that NDVI is the most important fire-influencing factor in Northeast China whereas altitude can be ignored.The current study can support fire danger studies of Northeast China and related studies which employ binary logistic regression model.