Probabilistic rainfall threshold of landslides in Data-Scarce mountainous Areas: A case study of the Bailong River Basin, China
Probabilistic rainfall threshold of landslides in Data-Scarce mountainous Areas: A case study of the Bailong River Basin, China
复制标题
资料稀缺山区滑坡的降雨阈值研究--以白龙江流域为例
DOI：
10.1016/j.catena.2022.106190
复制
发表时间：
2022-06
期刊：
影响因子：
6.2
通讯作者：
Wanyu Jiang;Guan Chen;Xing-min Meng;Jiacheng Jin;Yan Zhao;Linxin Lin;Yajun Li;Yi Zhang
中科院分区：
文献类型：
--
作者：
Wanyu Jiang;Guan Chen;Xing-min Meng;Jiacheng Jin;Yan Zhao;Linxin Lin;Yajun Li;Yi Zhang
The damage caused by rainfall-induced landslides has increased globally. The development of urbanisation has led to the expansion of residential areas in mountainous areas, and human activities have accelerated slope instability. However, the limited data records in mountainous zones have resulted in low-precision landslide rainfall thresholds. Establishing a landslide early warning model in areas with scarce data is an unresolved problem. This study uses the Bailong River Basin in western China as an example. By downscaling Tropical Rainfall Measuring Mission (TRMM) data to a daily resolution of 1 km and conditionally merging it with ground station rainfall, we obtain high-precision rainfall data to establish rainfall threshold curves based on Bayes' theorem and the frequency method. Subsequently, a landslide susceptibility map based on deep learning was integrated to construct a regional landslide early warning model. The results are as follows: First, compared with downscaling rainfall and rain gauge interpolation data, the daily rainfall data with a resolution of 1 km that combines the two has better overall performance and higher accuracy. Second, we generated the event-rainfall-and-duration threshold curves of 5%, 20%, and 50% probability by combining the frequency method and Bayesian theorem, and show that with a probability of landslide occurrence of 50%, while the intercept of Bayes' theorem is greater than that of the frequency method, the overall trend is consistent. Finally, by using a typical mass rainfall event to test the performance, the results show that the early warning capability of the model integrated with the rainfall threshold and landslide susceptibility map is more accurate than that of using a separate rainfall threshold. The outcomes of this research are expected to provide efficient support for early warning and risk management of geological disasters in mountainous areas with scarce data.
