基于BP神经网络的南丹县滑坡易发性及耦合致灾机制研究

Study on Landslide Susceptibility and Coupled Disaster-Causing Mechanism in Nandan County Based on BP Neural Network

摘要: 针对当前滑坡研究中易发性评价与致灾机制协同不足、难以形成“评价–机理–防控”逻辑闭环的问题，本文以广西南丹县为研究区，选取高程、坡度、地形起伏度、岩性、距断层距离、距河流距离、距道路距离、归一化植被指数、年均降雨量9个评价因子，基于BP神经网络模型开展滑坡易发性高精度评价。从三个方面解析了耦合致灾机制，并识别了关键主控因子。结果表明：南丹县滑坡极高、高易发区面积占比分别为20.6%、27.4%，集中于中低山丘陵过渡带、断裂带沿线、河流岸坡及矿产开采区；坡度 > 25˚、软弱夹层岩性、距断层 < 2 km、年均降雨量 > 1200 mm、距道路 < 1 km为滑坡高发组合条件，五因子叠加区滑坡密度达0.121个/km²；岩性、坡度、年均降雨量、距断层距离、距道路距离为全局主控因子，累计贡献达93.6%。本文研究形成了风险识别–机理解析–靶向防控的一体化思路，可为南丹县滑坡精准防控、矿山生态修复与国土空间规划提供科学支撑。

Abstract: This study focuses on the poor integration of landslide susceptibility assessment and disaster-causing mechanism, which hinders the formation of a complete “assessment‑mechanism‑prevention” system. Taking Nandan County, Guangxi as the study area, a BP neural network was used to evaluate landslide susceptibility using nine factors: elevation, slope, relief, lithology, distance to faults, rivers, roads, NDVI, and annual rainfall. The coupled disaster mechanism was analyzed from three aspects, and key controlling factors were identified. Results show that extremely high‑ and high‑susceptibility zones account for 20.6% and 27.4%, mainly distributed in mountain‑hill transitions, faults, rivers and mining areas. The high‑risk combination includes slope > 25˚, weak lithology, distance to faults < 2 km, rainfall > 1200 mm and distance to roads < 1 km, with a landslide density of 0.121/km². Lithology, slope, rainfall, distance to faults and roads contribute 93.6% cumulatively. This research establishes an integrated framework and supports landslide prevention, mine restoration and spatial planning in Nandan County.