Abstract. Reliable estimates of extreme rainfall events are necessary for an accurate prediction of floods. Most of the global rainfall products are available at a coarse resolution, rendering them less desirable for extreme rainfall analysis. Therefore, regional mesoscale models such as the Advanced Research version of the Weather Research and Forecasting (WRF-ARW) model, are often used to provide rainfall estimates at fine grid spacing. Modelling heavy rainfall events is an enduring challenge, as such events depend on multiscale interactions, and the model configurations such as grid spacing, physical parameterization and initialization. With this background, the WRF-ARW model is implemented in this study to investigate the impact of different processes on extreme rainfall simulation, by considering a representative event that occurred during 15–18 June 2013 over the Ganges basin in India, which is located at the foothills of the Himalayas. This event is simulated with ensembles involving four different microphysics (MP), two cumulus (CU) parameterizations, two planetary boundary layer (PBL), and two land surface physics options; and different resolutions (grid spacing) within the WRF model. The simulated rainfall is evaluated against the observations from 18 rain gauges and the Tropical Rainfall Measuring Mission Multi-Satellite Precipitation Analysis (TMPA) 3B42RT version 7 data. From the analysis, it is noted that the selection of MP scheme influences the spatial pattern of rainfall, while the choice of PBL and CU parameterizations influence the magnitude of rainfall in the model simulations. Further, WRF run with Goddard MP, Mellor–Yamada–Janjic PBL and Betts–Miller–Janjic' CU scheme is found to perform best in simulating this heavy rain event. The model performance improved through incorporation of detailed land surface processes involving prognostic soil moisture evolution in Noah scheme as compared to the simple Slab model. To analyze the effect of model grid spacing, two sets of downscaling ratios – (i) 1 : 3, Global to Regional (G2R) scale; and (ii) 1 : 9, Global to Convection-permitting scale (G2C) are employed. Results indicate that higher downscaling ratio (G2C) causes higher variability and consequently, large errors in the simulations. Therefore, G2R is opted as a suitable choice for simulating heavy rainfall event in the present case study. Further, the WRF simulated rainfall is found to exhibit least bias when compared with that of the Coordinated Regional Climate Downscaling Experiment (CORDEX) data and the NCEP FiNaL (FNL) reanalysis data.