基于Logistic回归模型的大兴安岭地区林火发生概率预测研究

Prediction of forest fire probability in Daxing’an Mountains area based on Logistic regression model

为实现对大兴安岭地区林火发生概率的基本预测,根据该地区历史火灾资料、地形因子、气象因子和人为因子,基于Logistic回归模型建立该地区林火发生概率预测模型。研究结果表明:坡度、海拔、平均气压、平均气温、平均相对湿度、最小相对湿度是该地区林火发生的主要驱动因子;坡度、海拔、平均气压、平均气温与林火发生概率呈正相关,平均相对湿度和最小相对湿度与林火发生概率呈负相关;ROC曲线下面积值(AUC)为0.91,最佳临界值为0.425;建立的大兴安岭地区林火发生概率预测模型,建模样本总体准确率为82.4%,验证样本总体准确率为80.5%;该地区夏季林火发生概率明显高于春、秋季;林火发生概率较高的Ⅳ级和Ⅴ级火险区主要集中在夏季该地区东南部、西部以及春、秋季该地区东南部。研究结果可为该地区林火预测提供1种参考方法。

In order to realize the basic prediction on the occurrence probability of forest fire in Daxing’an Mountains area,according to the historical fire data,topographic factors,meteorological factors and human factors of this area,a prediction model on the occurrence probability of forest fire in this area was established based on Logistic regression model.The result showed that the slope,elevation,mean air pressure,mean air temperature,mean relative humidity and minimum relative humidity were the main driving factors of forest fire in this area.The slope,elevation,mean air pressure and mean air temperature were positively correlated with the occurrence probability of forest fire,while the mean relative humidity and minimum relative humidity were negatively correlated with the occurrence probability of forest fire.The AUC value under the ROC curve was about 0.91,and the optimal critical value was about 0.425.For the prediction model on the occurrence probability of forest fire in Daxing’an Mountains area,the overall accuracy of modelling samples was 82.4%,and the overall accuracy of validation samples was 80.5%.The probability of forest fire in this area was significantly higher in summer than those in spring and autumn.The grade IV and V risk areas with higher probability of forest fire were mainly concentrated in the southeast and west of the area in summer,and the southeast of the area in spring and autumn.The results can provide a reference method for the forest fire prediction in this area.