井冈山森林火险等级划分及预报模型构建

Grades Classification and Prediction Model Construction of Forest Fire Alarm in Jinggang Mountain

森林火灾严重威胁生态安全和国民经济,由于不同林地的气候以及可燃物存在差异,森林火险具有明显的区域性特征。因此,在考虑气象因子的基础上,将可燃物含水率引入小区域的森林火险指数的计算,对建立更精确的小区域森林火险等级标准和预报模型具有重要意义。本研究系统地分析了2013—2016年井冈山地区森林可燃物含水率与气象因子的分布频率及因子间的相互关系。通过主成分分析方法对所有因子进行降维处理,获得火险因子得分方程,并计算出2013—2016年井冈山地区逐日森林火险指数,进而构建火险等级划分和森林火险等级预报模型。结果表明:井冈山森林火险等级划分为5类,分别为低(火险值≤0.024)、较低(0.024<火险值≤0.067)、高(0.067<火险值≤0.167)、较高(0.167<火险值≤0.232)、极高(火险值>0.232)。基于BP神经网络模型构建了井冈山森林火险等级预报模型,预测精度可达96.4%。并利用2013—2017年卫星监测到的井冈山地区热源点数据对模型进行检验,预报准确率高达92.3%,表明该火险等级标准和预报模型能够满足井冈山地区日常防火业务需求。

Forest fire seriously threatens the ecosystem safety and economy in China, which is characterized by regional features due to the regional variation of climate and wildland fuels. The current standard grades and prediction models of forest fire danger focuses only on meteorological factors, thus introducing the moisture of wildland fuel to the model is meaningful. The distribution frequency of the meteorological factors and the wildland fuel moisture contents in Jinggang mountain during 2013-2016 were analyzed, and their relationships were also discussed. Based on dimension reduction of those factors using the principal component analysis, the scoring equation and values of forest fire danger in Jinggang mountain were obtained, and the values were classified into five grades as follows: low (value ≤ 0.024), relatively low (0.024 < value ≤ 0.067), high (0.067 < value ≤ 0.167), relatively high (0.167 < value ≤ 0.232), very high (value > 0.232). Furthermore, the prediction model of forest fire danger grade was built using the Back Propagation (BP) Neural Network, and the prediction accuracy reached 96.4%. This prediction model can directly output the grade after inputting the value of each factor, the prediction results of 13 hotspots showed that this model and forest fire danger grade can meet the needs of daily prevention of forest fire in Jinggang mountain.