Study On The Features Of Forest Fire In Gansu Province And Fire Dangerous Forecasting Methods

By using the forest fire data, related meteorological data, NCEP/NCAR daily reanalyzed data and remote sensing data, the forest climate features, temporal and spatial characteristics of forest fire, the effect of meteorological facts in forest area on forest fire and the drought before fatal forest fire and its development are analyzed. And then a forecasting and warning system for forest fire of Gansu province is set up based on the Support Vector Machines (SVM) and the regressive method. The results show that:1. In Gansu forest areas, there are great differences in spatial distribution of temperature, which gradually decreases from southeast to northwest and also from basin or valley to plateau or mountain. Temperature in July is highest and in January is lowest. Form 1988 to 2003, temperature in Gansu forests steadily increases.2. Also for precipitation, there are great differences in spatial distribution in Gansu forest areas, which decreases from southeast to northwest. Annual average of precipitation in Kangnan forest area, southeast part of Gansu province is the maximum, and that in Qilian mountain forest, west part of Gansu is the minimum, particularly in the middle of Qilian.3. The drought in each forest area in Gansu province is more and more severe from 1998 to 2003, during which the year of 1991, 1997, 2001, 2002 are the most dry, particularly 1997.4. Forest fire occurs frequently in Gansu , about 15 times per year, mostly in winter and spring, and mainly concentrates in such forests as Maxian mountain, Xiaolong mountain, Bai Longjiang river, Ziwu mountain and Qilian mountain.5. There are five synoptic styles mainly influencing forest fires, which are high-pressure control style, multiwave in latitudinal circumfluence style, straight west wind jet style, subtropical high-pressure control style and cold-trough eastern moving style.6. Soil moisture and vegetation humidity can be reflected by remote sensing datai which is one of the most effective factors for forecasting the happening of forest fires. Remote sensing can also be used to continuously monitor and track the process of fire development, estimate the location, area and temperature of forest fires.7. There are negative correlations between forest fire frequency and NDVI and average relative humidity. Compositive fire danger index which comprises forest vegetation index and air humidity, integrating the vegetable moisture as well as air dryness in forest area, is one of an objective and feasible index for forest fore forecasting, and can reveal combustibility of forest.8. There are positive correlation between compositive fire danger index and drought index. It has scientific foundation that introduce drought index into fire danger forecasting model.9. Close relationships exist between forest fire danger and moisture, temperature, precipitation and drought according to the regressive equation. Support Vector Machines (SVM) method can improve forest fire forecasting accuracy .It is feasible to introduce SVM into fire danger forecasting model.