基于RF-FR模型的滑坡易发性评价——以略阳县为例

滑坡易发性评价是指导区域滑坡初步预警、预报的重要手段。为提高县域滑坡易发性评价的准确性，以随机森林模型（RF）、频率比模型（FR）为基础模型，结合2种模型的优越性，建立随机森林-频率比模型（RF-FR），进行滑坡易发性评价。以略阳县域为研究区，选取高程、坡向、坡度、地层、地表粗糙度、距断层的距离、曲率、距道路的距离、地形湿度指数、距河流的距离及降雨量等14项影响因子建立数据库，采用Spearman方法对各因子相关性进行分析，剔除地形起伏度等3项相关性较高的评价因子，并基于滑坡相对点密度（LRPD）进行评价因子分析。结果表明：①滑坡灾害点与线状因子的距离呈负相关，即距离越近，灾害点越多。②FR、RF、RF-FR模型预测率分别为84.3%、90.1%、95.0%，RF-FR模型较FR、RF模型预测精度分别提高了10.7%、4.9%。③RF-FR模型的滑坡灾害点在高、极高易发区的比例比FR、RF模型分别提高了15.89%、5.29%。

Assessment of Landslide Susceptibility Based on RF-FR Model: Taking Lueyang County as an Example

The assessment of landslide susceptibility is an important means to guide the preliminary early warning and forecast of regional landslide. In order to improve the accuracy of landslide susceptibility evaluation in county area, random forest model (RF) and frequency ratio model (FR) were used as the basic models, and RF-FR was established to evaluate landslide susceptibility combined with the advantages of the two models. In Lueyang county domain for the study area, selection of elevation, slope direction, slope, formation, surface roughness, the distance from the fault, curvature and the distance from the road, terrain humidity index, the distance from the river, a database of 14 factors such as rainfall, by adopting the method of Spearman correlation analysis of each factor, eliminate topographic relief degree of three high correlation of evaluation factors, and evaluate the relative point density (LRPD) based on landslide factor analysis. The results show that:①there is a negative correlation between the distance between landslide disaster points and linear factors, that is, the closer the distance is, the more disaster points are. ②The prediction rates of FR, RF and RF-FR models are 84.3%, 90. 1% and 95.0%, respectively. Compared with FR and RF models, the prediction accuracy of RF-FR model is 10.7% and 4.9% higher than that of FR and RF models. ③The proportion of landslide disaster points in high and extremely high-risk areas of 4RFmurFR model is 15.89% and 5.29% higher than that of FR and RF model, respectively.