The objective of this study is to assess the impact of using different initialization techniques and cloud microphysics of a numerical atmospheric model to improve the forecasting of Indian summer monsoon rainfall (ISMR). A total of six intense precipitation events over the Ganges–Brahmaputra–Meghna and Indus River basins were tested to identify the most suitable combination of parameterization and initialization techniques. The global forecast system (GFS)‐based numerical weather prediction (NWP) forecast fields were dynamically downscaled by the mesoscale model of weather research and forecasting (WRF). The performance of four types of initial conditions with three cloud microphysics was assessed using a model resolution of up to 9 km. A main conclusion is that the model initialized using hot start in the study involves more uncertainty, probably due to poor‐quality data assimilation, and it cannot exceed the performance of cold‐start initialization. The study findings provide evidence that the finer resolution initial condition is promising in higher resolution models. In the case of cloud microphysics, the performance of WRF single moment 5 class (WSM5) was sufficient for South Asian monsoon systems within this scale of the model resolution. The findings provide a general guideline for flood forecasters for the WRF model set‐up for forecasting the ISMR from publicly available GFS‐based NWP forecast fields.