多源数据支持下的三峡库区滑坡灾害空间智能预测

Spatial Intelligent Prediction of Landslide Hazard Based on Multi-source Data in Three Gorges Reservoir Area

针对传统空间分析技术不易发掘多源、海量滑坡数据中隐藏的模式、趋势和关系等问题,以三峡库区为研究对象,通过多源数据融合提取滑坡孕灾环境和影响因素信息,进而利用动态构建算法建立BP神经网络模型,定量预测滑坡空间易发性,生成滑坡易发性分区图。采用成功率曲线分析预测结果,三层BP神经网络的预测精度达到89.75%,预测结果与野外调查实际情况吻合较好。

Landslide hazard is influenced by many temporal and spatial factors. Traditional spatial analytical techniques cannot easily discover new and unexpected patterns, trends, and relationships that can be hidden deep within very large diverse geographic datasets. Focusing on Three Gorges Reservoir Area, environmental and triggering factors for landslide occur- rences were extracted from multi-source data. Then, quantitative landslide susceptibility in- dices were calculated using the trained three-layered BP neural network, and the landslide susceptibility maps were generated. Finally, success rate curve was used to verify the results of landslide susceptibility mapping, and the results showed the best accuracy of 89. 75%. The validation showed sufficient agreement between the prediction results and existing land- slide. Therefore, the proposed model is an efficient method for landslide intelligent predic- tion, and can provide a significant reference for landslide hazard prediction and assessment.