Model settings change how extreme rainfall is simulated

Cumulus parameterization had a much larger influence on rainfall results than cloud microphysics.

A study evaluated how different settings in the Weather Research and Forecasting (WRF) model affected a rainfall event from 5-7 July 2026. The event occurred over the Konkan coast and Western Ghats in Maharashtra, India.

The study used a 3 by 3 factorial design with nine simulations. It combined WSM6, WDM6, and Thompson microphysics with Kain-Fritsch, Tiedtke, and Grell-Freitas cumulus schemes.

Cumulus parameterization is a way to simulate convection. Cloud microphysics is a scheme used in the simulations. Rainfall was compared against IMD 0.25-degree gridded observations, IMERG, and ERA5.

Cumulus parameterization accounted for 97.3% of the variation in mean rainfall. It accounted for 88.2% in RMSE and 84.7% in spatial correlation. Microphysics contributed only 2.0% to mean rainfall variation.

Kain-Fritsch had the highest spatial correlations between 0.533 and 0.581 but overestimated rainfall. Tiedtke had the smallest mean biases and RMSE, but showed a 28-34% rainfall deficit over the Konkan-Ghats belt.

All nine simulations displaced the rainfall maximum by approximately 140-195 km northward. A single-event, single-member experimental design does not establish a universally optimal physics configuration.

Key facts

  • Cumulus parameterization accounted for 97.3% of across-experiment variation in mean rainfall.
  • All nine simulations displaced the event-total rainfall maximum by 140-195 km northward.
  • Tiedtke achieved the smallest mean biases and RMSE.

All sections

Edition