Author(s): Sijia Hao; Qiang Ma; Philippe Gourbesville; Guomin Lyu; Wenchuan Wang; Changjun Liu

Linked Author(s): Qiang Ma, Philippe Gourbesville, Changjun Liu

Keywords: Machine learning; Parameter regionalization; Ungauged catchment; Flash floods; Henan Province

Abstract: Caused by extreme rainfall, flash floods in mountainous catchments often lead to serious losses of life and property. Henan province is located at the central part of China which frequently suffered flash flood disasters due to its typical topographic and meteorological characteristics. Currently, the local decision makers and managers in the province have urgently presented the requests of having an accurate model simulation and forecast of flash flood disasters. However, as the flash flood disaster often occurred in mountainous catchments which have few gauging stations existed, it is quite challenging to obtain precise simulation results with many unknown parameters in the model simulation. In order to overcome this obstacle and to improve the local defense capacity of flash flood disasters, this paper has proposed a machine learning approach for parameter regionalization which is able to estimate the model parameters from gauged catchments in this province. With the data obtained from deterministic distributed hydrological simulations of 19 catchments in Henan province, a Xgboost model has been set up to detect the similarity between the targeted and the referenced catchments, and then to determine the most similar catchment for the target to give values for its model parameters. The results shown that: (1) the characteristics of the top soil layer and the short-term rainfall intensity are the main criterion to determine the similarity between two catchments in Henan province. (2) Compare to other classical approaches for parameter regionalization such as the “shortest distance” approach, the Xgboost model is able to effectively improve the accuracy of flash flood simulation in Henan province. (3) The machine learning approach presented in this paper could be optionally applied in other area that has ungauged mountainous catchments frequently suffered flash flood disasters.

DOI: https://doi.org/10.3850/IAHR-39WC2521711920221308

Year: 2022