基于气象因子的福建省森林火险预测模型

Model to predict forest fire occurrence in Fujian Province based on meteorological factors

福建省森林覆盖率位居全国第一,同时福建省也是林火高发区之一,但关于该地区林火预测模型和空间格局的研究还不完善。将2000-2004年林火数据分为60%(建模)和40%(校验)两部分进行建模和校验,得到5个中间模型,选择其中3个中间模型中显著的因子进入最后的全样本数据拟合,并使用2005年林火数据做独立校验。结果表明,日最高地表气温、日最低地表气温、平均风速、最大风速、20-20时降水量、日照时间、日最高气温、平均相对湿度、最小相对湿度共9个变量与林火发生显著相关。全模型拟合结果显示模型的预测准确率为72.3%。

Fujian Province is a high fire frequency region, however, the studies regarding the prediction model and spatial distribution of forest fire of Fujian are relatively rare. Due to the close relationship between climate factors and fire occurrence, in this study, the fire prediction model was established based on daily weather factors using logistic regression. Meanwhile the fire dataset during between 2000 and 2004 was divided into 60% （training） and 40% （validation） and made five repetitions to keep the stability of variable selection. Besides, fire dataset of 2005 was selected for further model validation. Therefore, five intermediate sub-models were created. The variables that were significant in at least three of five sub-models were selected to fit the final complete dataset and fire data of 2005 were applied to evaluate the prediction accuracy of model. The results showed that 9 variables such as daily maximum ground surface temperature, daily minimum ground surface temperature, and daily mean windspeed significantly impacted fire occurrence. In addition, the predicting accuracy of fire model was 72.3%.