Overall, the accuracy of 24-hour precipitation occurrence ranges from approximately 30% to 50% in the ensemble, with around two thirds of WRF simulations underestimating the precipitation frequency. The WRF simulations struggle with predicting precipitation intensity and timing on shorter timescales. While near-surface temperatures and vertical profiles are generally well represented, deviations in low-level moisture are quite large. Moreover, the WRF simulations show substantial sensitivity to the different initial/boundary conditions and PBL scheme, with the sensitivity to the microphysics scheme and aerosol loading playing a secondary role. Results also suggest that the 1-km ensemble mean generally demonstrates superior performance than the original coarser-resolution operational forecasts in terms of 24-h accumulated precipitation, near-surface thermodynamics, and atmospheric vertical profiles. However, identifying the underlying physical processes responsible for the model–observation discrepancies is difficult given the range of environmental conditions spanning the field campaign period. Therefore, it is of importance to conduct further individual case studies to understand the impact of physical processes and parameterizations on simulated and forecasted convective clouds and precipitation.

