Abstract: A landslide dam is formed by natural landslide and is extremely vulnerable to dam failure due to flow scour and seepage, thereby posing a severe threat to human lives and property downstream. Rapid and accurate prediction of the peak flow rate through the breach is the most important task in the rescue work of landslide dam breaching. In previous studies, a series of landslide dam peak flow prediction models were developed based on the idea of the database-statistical regression-breaching model, but none of them took into account the effects of dam erosion characteristics or particle composition on the breaching flow. With continuous expansion of the data, it is difficult yet for these prediction models to achieve the expected results. This study is based on a database of 2060 cases of landslide dam failure in China and abroad, and takes into account the landslide dam geometrical characteristics, erosive characteristics and particle composition, through extracting seven parameters－weir height (Hd), weir width (Wd), weir volume (Vd), reservoir capacity (Vl), failure depth (Hw), average width of failure (Wb), and median particle size (d50). We classify the degree of dam erosion into five categories, and establish a peak flow prediction model of landslide dam failure that includes a factor of erosion characteristics. Then, we validate the model by selecting 120 cases with complete data from the database, and achieve a root mean square error (RMSE) of 9581 m3/s and a correlation coefficient (R2) of 0.965, with a comparison with the typical parameter models available in China and abroad. The results show that considering the landslide dam erosion characteristics and the d50 parameter improves the model significantly in predicting the peak breaching flow, and our model is satisfactory in accuracy and applicable to the peak flow predictions aimed at rescue and relief work. It would also help the routing of such outburst floods and the division of the affected areas downstream.

Key words: landslide dam, erosion characteristics, outburst flow, parametric model