**5. Discussion**

The forecasting accuracy of the flood peaks of the lumped model seems to be more demanding in terms of the accuracy of the input rainfall than the two other coupling systems, since the flood peaks of the lumped model were more moderate when unassimilated rainfall was poorly simulated such as in Event 4, where less than a quarter of the observed flood peaks were simulated before assimilation. Although the results of the corresponding indices were better following assimilation, this characteristic greatly increased the uncertainty of the atmospheric-hydrological coupling process, which can easily lead to errors when studying flash floods since it is difficult to guarantee the accuracy of the rainfall forecasting.

Although data assimilation improves the precipitation input required by WRF-Hydro, it is still insufficient for complex model systems due to the need to input more meteorological variables. To improve the performance of the coupled system in the prediction of hydrological processes, research on WRF-Hydro also includes the assimilation of soil moisture variables [59] and real-time flood assimilation. Indeed, the coupling system of the WRF-Hydro model has a stronger basis in physical processes than the former two coupled systems; however, the complexity of parameter estimation that emerges from the model also poses a greater challenge, which is a typical problem in many complex physics-based models such as the variable infiltration capacity model and the community land model (CLM) [60]. In addition to data assimilation, more precise expressions of regional rainfall-runoff mechanisms also need to be further explored. The hydrodynamic parameters developed for local areas may not necessarily be applicable to mesoscale areas; therefore, although the model structure is feasible, the parameters of WRF-Hydro in terms of soil properties are not fully calibrated for northern China. There is an urgen<sup>t</sup> need to find easily available parameters and expression equations that reflect the spatial heterogeneity of local infiltration processes across the region as an alternative to model application.
