To enhance the precision of short-term regional weather forecasts with the Rapid-refresh Multi-scale Analysis and Prediction System-Urban subsystem (RMAPS-Urban), an urbanized land data assimilation system RMAPS-LDAS with a spatial resolution of 1 km is developed by coupling the Noah land surface model with a multi-layer urban model, namely, the building effect parameterization and building energy model (BEP–BEM), enabling improved representation of land surface elements for both natural and urban underlying surfaces. RMAPS-LDAS runs offline utilizing merged meteorological data from high-density in-situ observations and a gridded meteorological dataset generated via thin-plate smoothing spline models. The structure of the coupled Noah and BEP–BEM model is modified to enhance computational and storage efficiency, making RMAPS-LDAS operationally applicable for regional weather forecasts. The urban layer (roof, wall, and road) temperatures require a spin-up period of no more than seven days, significantly shorter than that needed for soil states. Evaluation over the Beijing–Tianjin–Hebei region show that RMAPS-LDAS reduces the root-mean-square error (RMSE) and bias of simulated soil temperature by 47.8% and 92.3%, respectively, compared with ECMWF soil temperature forecasts. The system reduces RMSE and absolute bias in soil moisture simulations by 6.5% and 18.2%, respectively. Comparisons with observed road and building temperatures confirm that the BEP–BEM model realistically captures the thermal evolution of urban surfaces. Coupling RMAPS-LDAS with RMAPS-Urban led to a 9.6% reduction in RMSE for 2-m specific humidity forecasts, although a slight increase in air temperature RMSE (0.54%) was observed. However, for urban areas, the coupling reduced air temperature RMSE by 4.9%. Overall, the impact on wind speed prediction remains relatively minor.

The runoff calculation scheme in the Weather Research and Forecasting (WRF) model is based on an infiltration-excess surface runoff scheme, which likely leads to an overestimation of soil moisture and an underestimation of surface runoff when heavy rainfall occurs in areas with complex terrain. To overcome this defect, we considered the overland flow process on the grid scale of the WRF model for the first time by coupling a two-dimensional diffusion wave equation into the WRF Noah land surface model (LSM), called the WRF_Overland Flow (WRF_OLF) model. The new WRF model was then utilized to simulate the extreme rainfall that occurred during 19–22 July 2021 near the city of Zhengzhou in central China, which led to an extreme flood event. The results showed that the new WRF model simulated well the convergence and accumulation of overland flow in low-lying areas, changing the distributions of surface runoff and soil moisture and thereby influencing the exchanges of heat and water vapor between the surface and the atmosphere. The local change in non-adiabatic heating at the surface contributed to a decrease in surface pressure and then affected the development of the weather systems associated with the heavy rainfall event. Relative to a remarkable underestimation of rainfall in the original WRF simulation, the maximum rainfall intensity and the cumulative rainfall in the simulation with the new WRF configuration increased by 54.7% and 49.5%, respectively, bringing them closer to their observations. Concurrently, the new WRF model increased the skill for flood prediction. The results of this study provide new insights into the mechanisms of interaction between the land surface and the atmosphere and their roles in helping to predict heavy rainfall and associated flooding in areas of complex topography.