【摘要】 [目的]对区域性泥石流敏感性进行分析,为吉林省洮南市泥石流灾害预测研究提出一种高效快捷的分析模型。[方法]针对现行大多数概率统计模型预测率较低的不足,利用人工智能算法中效果明显的随机森林算法,以吉林省洮南市西北部山区为研究区域,选用高程、坡度、坡向、平面曲率、剖面曲率、河流、归一化差分植被指数、地形湿度指数、土地利用及岩性10个评价因子构建了频率比和随机森林泥石流敏感性评价模型进行对比验证。模型准确性的验证方法采用受试者特征曲线(ROC曲线)及累积频率曲线下面积(area under curve,AUC)。[结果]随机森林对研究区泥石流敏感性进行分析,并通过GIS将敏感性图分为5个敏感性区域,位于高敏感性区以上的灾害点占82.3%。验证模型成功率及预测率分别为88.4%与90.4%,相较于频率比的成功率及预测率(86.4%和75.1%)效果良好。[结论]在洮南市北部进行泥石流敏感性分析中,采用随机森林方法进行建模,并利用频率比方法进行对比,结果显示随机森林法结果可靠准确。更多还原

【Abstract】 [Objective] The susceptibility of regional debris flow was analyzed and an efficient and rapid analysis model proposed for the debris flow disaster prediction research Taonan area of Jilin Province. [Methods] Owing to the current shortcomings of most probabilistic statistical models that have low prediction rates, the random forest algorithm with obvious effect built on an artificial intelligence algorithm was used to study the northwestern mountainous area of Taonan City, Jilin Province. The elevation, slope, aspect, plane curvature, profile curvature, river, normalized difference vegetation index, topographic humidity index, land use, and lithology were selected. A random forest debris flow susceptibility assessment model for the study area was constructed using these factors. The frequency ratio method was used to model and compare with the random forest model. The model effect was verificated using a receiver operating characteristic curve and an area under the curve. [Results] Random forests were used to analyze the sensitivity of debris flow in the study area and were divided into five sensitivity areas by GIS. The disaster points above the high sensitivity area accounted for 82.3%. The success rate and prediction rate of the verification model were 88.4% and 90.4%, respectively, which was better than the success rate and prediction rate of frequency ratio(86.4% and 75.1%). [Conclusion] The sensitivity analysis of debris flow in the Northern Taonan City was performed using the random forest method and compared with the frequency ratio method. The random forest results were reliable and accurate.更多还原