河南省火灾影响因素的空间分析

科学揭示火灾及其影响因素间的空间关系可为防火管理提供决策支持和有益启示。以往研究多在“空间平稳”的框架下进行火灾影响因素分析,但火灾和其可量化的影响因素往往自身均表现为“空间异质”,基于非空间的全局模型模拟可能会得出误导性甚至错误的结论。地理加权回归(GWR)可解释火灾及其影响因素间空间关系的局部变异。本文选取影响火灾分布的高程、坡度、居民地可达性、道路可达性、地表温度、归一化差植被指数和全球植被湿度指数作为解释变量,以是否火烧作为二元因变量,应用logistic GWR对河南省2002-2012年火季(9-10月)火灾的影响因素进行探索性分析。以多时态空间抽样取得训练样本,利用GWR 4.0软件开发一个logistic GWR火烧概率模型,从可靠性和区分能力两方面对模型性能分别进行内部检验和独立检验,以确保火灾影响因素分析的可靠和合理性。结果表明:①坡度、居民地可达性、温度、植被长势和植被湿度对河南省火灾的影响呈现显著空间变化,高程、道路可达性的影响空间变化不显著,低海拔、道路可达性差的区域更易发生火灾。②温度和植被长势对火灾影响省内全局显著,坡度、居民地可达性和植被湿度对火灾影响在省内仅部分区域显著。③河南省可划分为7种类型区,不同类型区的火灾影响因素相对重要性存在差异,应因地制宜制定防火策略和确定防火重点。④logistic GWR模型可用于分析火灾影响因素的局部空间变异,作为火险研究的一种有效工具。

Spatial analysis of fire-influencing factors in Henan Province

The spatial relationships between fire events and fire-influencing factors have important implications for fire managers and scientifically revealing these relationships is therefore significant for management purposes. The spatial stationary relationships between fire events and fire-influencing factors considered by previous fire risk studies contradict the fact that fire events and their quantifiable influencing factors are always characterized by spatial heterogeneity. If the intrinsically non-stationary relationships between fire events and fire-influencing factors are modeled by some stationary models, misleading and even erroneous conclusions can be drawn, which hamper fire prevention operations. In this study, logistic geographically weighted regression (LGWR) that can account for local variations of spatial relationships between fire events and fire-influencing factors was employed to analyze the influences of different fire-influencing factors on fire events in the high risk season (September and October) from 2002 to 2012 in Henan Province. The independent variables of the model include altitude (Al), slope (Sl), distance to the nearest village (Dv), distance to the nearest path (Dp), land surface temperature (LST), Normalized Difference Vegetation Index (NDVI), and global vegetation moisture index (GVMI); and the binary dependent variable is monthly fire presence, with 1 representing presence and 0 representing absence. A training subset derived from spatial random sampling was created and potential multicollinearity among the independent variables was excluded, and then a LGWR fire probability model was developed using the GWR 4.0 software. The reliability and discrimination capacity of the developed fire probability spatial model was evaluated using a testing subset and an independent validation subset and the results show good model performance. The model was used for fire-influencing factor analysis in the next step. After delineating and overlaying the significant areas of the non-stationary fire-influencing factors, seven fire prevention regions were identified in Henan Province. The relative importance of the non-stationary fire-influencing factors was evaluated by comparing the absolute values of their estimated coefficients spatially. The results indicate that: I) The influences of Sl, Dv, LST, NDVI and GVMI on fire events present significant spatial variability, whereas the influences of Al and Dp exhibit insignificant spatial variability in Henan Province. II) The influences of LST and NDVI on fire events are significant globally in Henan Province, whereas the influences of Sl, Dv and GVMI are only significant locally. The sites most strongly influenced by LST are mainly Nanyang, Zhumadian, Xinyang and their contiguous areas. The sites most strongly influenced by vegetation cover (NDVI) are primarily Zhoukou, Xinyang, Luohe, Xuchang, Zhumadian and Shangqiu. In Xinyang and southeast Zhumadian, fire events are most strongly influenced by Sl, while in Luoyang this factor is Dv, and in Zhoukou and the adjacent area of Luoyang, Nanyang and Pingdingshan, it is GVMI. III) This study demonstrates the usefulness of LGWR for exploring local variations of fire-influencing factors and for examining the validity of a global fire probability model. The practical implication of spatial analysis of fire-influencing factors resulted from LGWR is that different fire prevention policies and emphases should be formulated for each of the seven fire prevention regions. Because of such heterogeneity, fire prevention policies need to take into consideration local conditions.