Abstract. The Loess Plateau is the most erosion-prone area in China, while under large-scale ecological restoration runoff and sediments continue to decrease. This study examined the runoff generation mechanism at the catchment scale to understand the change in runoff generation. Six baseflow used to separation method were tested and the nonparametric simple smoothing method was seperating base flow. With the event runoff separation procedure, 340 rainfall–runoff events are selected in five typical catchments affected by significant human intervention in the Loess Plateau. Runoff characteristics, such as the event runoff coefficient, time scale, rise time, and peak discharge are studied on monthly and long-term scales. In catchments of Jialuhe, Chabagou and Gushanchuan with poor vegetation runoff response is strongly decided by rainfall intensity and is produced by Horton overland flow (HOF). While the mountainous catchments of Jingle and Zulihe runoff response is controlled by rainfall volume. The relation between runoff event characteristics and rainfall is complicated in Loess Plateau, where rainfall and underlying surface is significantly changing. The monthly of event characteristics is mostly controlled by rainfall characteristics. Long-term runoff coefficient experiences decreasing trend, while time scale trend is increasing. Land use changes lead to increasing catchment wetness display mostly strong reason in event characteristic response. According to our proposed framework for classifying dominant runoff generation patterns considering of hydrograph response time, discharge source, and flow paths, HOF runoff is still the dominant mechanism, but gradually shifts to Dunne overland flow (DOF) and combination runoff. We speculate that the reduction in runoff in the Yellow River is likely to be the dominant runoff mechanism changing.
Multi-time-scale input approaches for hourly-scale rainfall–runoff modeling based on recurrent neural networks
