Main content area

Rainfall threshold for initiation of channelized debris flows in a small catchment based on in-site measurement

Wei, Zhenlei, Shang, Yuequan, Zhao, Yu, Pan, Pan, Jiang, Yuanjun
Engineering geology 2017 v.217 pp. 23-34
engineering, landscapes, least squares, mass movement, neural networks, rain, runoff, support vector machines, threshold models, watersheds
The peak runoff discharge plays an important role in triggering channelized debris flows, especially for small catchments, due to the features of a small area and short channels. In small catchments, intense rainfall can trigger a flood within a short period of time. But currently, there are few rainfall threshold models that address the runoff from short-duration (<2h) rainfall. In this context, 37 short–duration rainfall events were analyzed, and field measurements were made for a partially-modified landscape in order to compare the variation in the amount of runoff. The grey relational analysis allowed us to rank the influence of the various factors that affected variations in the runoff. The results indicated that the intensity of rainfall was the most important factor, having a greater effect than other factors, such as the depth of the rainfall and its duration. Then, we trained a runoff prediction model based on these factors and verified its accuracy using the moving least squares (MLS) method, the genetic back-propagation neural network (GABP) method and the genetic support vector machine (GASVM) method. This analysis indicated that the MLS method had the best predictive capability. The assessment of the debris-flow hazard based on the predicted runoff and intensity-duration-frequency (IDF) curves is discussed in the study area. It was indicated that even a rainfall event of a two-year return period was dangerous for this specific case.