【摘要】 为研究秦岭地区2003—2018年林火发生与林火驱动因子之间的关系，对林火发生概率进行预测，使用2003—2018年MODIS卫星火点数据，基于逻辑斯蒂回归、随机森林和支持向量机建立秦岭地区林火发生预测模型，并进行精确度评价，选择拟合结果较好的随机森林林火发生预测模型绘制林火发生风险等级图。结果表明，1)随机森林对林火发生预测拟合结果较好，精确度为87.04%,AUC值为0.949。2)秦岭地区林火发生驱动因子重要性排序为NDVI、月平均风速、海拔、月平均气压、风速、气压、月平均降水、坡度、月平均空气比湿、空气比湿、连续无降水日和降水量。3)秦岭地区东部和南部为林火易发区域，建议在这些地区加大森林火灾防范宣传，增加瞭望台等基础设施，减少人为火灾的发生，做到林火早发现、早扑灭。更多还原

【Abstract】 This study aims to investigate the relationship between forest fire occurrences and their driving factors in the Qinling Mountains from 2003 to 2018,and to predict the probability of forest fires.Based on MODIS satellite fire point data from 2003 to 2018,logistic regression, random forest, and support vector machine models were developed to predict forest fire occurrences in the Qinling Mountains and their accuracy was evaluated.The random forest model, which provided better fitting results, was chosen to create the forest fire risk level map.The results showed that: 1) The random forest model had better fitting results for predicting forest fire occurrences, with an accuracy of 87.04% and an AUC value of 0.949.2) The importance of the driving factors for forest fire occurrences in the Qinling Mountains, in descending order, were NDVI,average monthly wind speed, elevation, average monthly atmospheric pressure, wind speed, atmospheric pressure, average monthly precipitation, slope, average monthly air humidity, air humidity, consecutive days without precipitation, and precipitation amount.3) The eastern and southern parts of the Qinling Mountains were identified as areas prone to forest fires.It is recommended to increase forest fire prevention publicity in these counties to reduce human-caused fires, and to enhance infrastructure such as lookout towers to ensure early detection and early suppression of forest fires.更多还原