| Taking Fangshan District in Beijing as research region, this article collects Fangshan District forest fire statistics, meteorological data, forest resources sub compartment data, district road map, DEM data and so forth during 1995-1999, and thoroughly analyses the relation between forest fire distribution and time, space, sub compartment land, climate, soil, plant and other sectors. It gives explorative research on building forest fire incidence forecasting model and classifying subpot forest fire based on geographically weighted logistic regression model. The following conclusions are presented through research:I. The relation between fire point and fire danger sectorFire has clear relationship with topographic factor. With the altitude increases, slope degree increases, the fire incidence reduces. Fire incidence in descending order distributes in flat followed by southeast, southwest, south, northeast, northwest, north, east, and west. Meteorological factor affects fire obviously. Fire incidence first increase and then decreases as the temperature keep up. The temperature to the highest incidence is 10.0℃-15.0℃.The relative humidity of 30%-40% easily cause forest fire, however, forest fire incidence decreases with relative humidity increase in the condition of relative humidity of 40%-100%. Forest fire incidence increases in waves after rainfall. Forest incidence peaks in the 100th-110th day, after rainfall, and increases with wind speed increasing which peaks when the wind speed during 4.1m/s-6m/s and decreases when the wind speed surpasses IOm/s.Ⅱ. Subpot fire forecasting model research and buildingForest fire incidence forecasting optimal model is obtained by comparing the nested logit model and evaluation index of regression coefficient, based on two logistic regression principle and multicollinearity diagnosis, with landscape, meteorology, plant, soil, social person and so on 23 factors as reference factor, analyzing the variable sample (subpot fire point and non-fire point in the proportion of 1:3.5 components)Ⅲ. Subpot fire incidence space forecasting model research and buildingThe article researches and builds Forest Fire Spatial Forecasting Model based on mixed Logistic GWR Model made by local and global variable after Logistic Regression development. Goodness-of-fit of the models speaking, indexes of mixed Logistic GWR Model are better than Logistic Regression Model. Lwr Quartile and Upr Quartile of road length, altitude, stock volume, soil thickness of mixed model surpass corresponding coefficient of global model B±1S.E. which fully present variable spatial instability. In the aspect of spatial analysis, spatial autocorrelation of mixed model improve a lot than global model which means that Fangshan District forest fire factor has features of spatial heterogeneity. GWR Logistic Regression Model would be better explain nonlocal and non-stationary fire factor which scientifically reveal the differences that different location fire factors influence forest fire incidence.IV. Fire factor spatial heterogeneity analysisAbove 4 factors have special instability which the range of differences are 0.505.4.108,2.739, 0.332, and significantly affect local-region. The influence of altitude on forest fire presents positive correlation -weak correlation -negative correlation from the east to west. The influence of soil thickness on fire presents inhibition which negative effect increases from center to both the east and the west sides. The influence of stock volume presents positive correlation, negative correlation from center to both the east and the west sides. The length from fire point to road presents negative correlation with forest fire in the most region of the center and west. Part of town north exist positive correlation between length and fire.Ⅴ. Subpot fire classificationThe article divides forest fire into level I to level V based on optimal forest fire incidence spatial forecasting probability model for Fire Department of Fangshan District reference. |
PDF Full Text Request