Data assimilation enhanced WRF simulation of the 2018 Kerala flood: sensitivity to initial conditions, parameterization schemes and extreme rainfall prediction skill

The frequent flood events and the impacts of climate change in Kerala highlight the urgent need to enhance the simulation capabilities of the Weather Research and Forecasting (WRF) model. This study focuses on simulating the 2018 Kerala flood using the Advanced Research WRF (WRF-ARW). Two different initial condition datasets, the National Centre for Environmental Prediction Final Analysis (NCEP-FNL) and the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ECMWF-ERA5), are employed to assess their impact on forecast accuracy. Simulations are conducted using three different parameterization scheme combinations, and the results are evaluated against India Meteorological Department (IMD) observations to determine the best-performing configuration. The optimal combination of initial conditions and parameterizations is further refined through data assimilation using the three Dimensional Ensemble Variational Assimilation (3DEnVAR), leading to improved model accuracy. The analysis indicates that the WSM6 microphysics, Grell 3D cumulus, and YSU planetary boundary layer schemes consistently provide the best performance across both initialization datasets (NCEP-FNL and ERA5) and during data assimilation. Spatial comparisons show that simulations using NCEP-FNL data are slightly more accurate than those using ERA5. The study also evaluates the model’s ability to predict very heavy and extreme rainfall events by identifying rainfall peaks and quantitatively assessing the Structural Similarity Index Measure (SSIM), Normalized Mutual Information (NMI), and correlation. The WSM6–Grell 3D–YSU combination captures peak rainfall events well on 15 and 16 August 2018 and produces strong similarity metrics, demonstrating reliable and consistent performance across both domains.