Relative Impact of Assimilation of Multi-Source Observations Using 3d-Var on Simulation of Extreme Rainfall Events Over Karnataka, India

This study explores the impact of assimilating diverse observational data on forecasting extreme rainfall events (EREs) using a three dimensional variational (3D-Var) assimilation approach. It focuses on 38 EREs across three meteorological divisions in Karnataka, India, using a high-resolution (03-km) Weather Research and Forecasting (WRF) model with three nested domains. Five distinct experiments were conducted, including a Control experiment without assimilation, and subsequent experiments integrating observations from various sources like atmospheric profiles from Atmospheric InfraRed Sounder (AIRS) and Moderate resolution Imaging Spectroradiometer (MODIS) satellites and radiosondes, ocean surface wind observations from Advanced SCATterometer (ASCAT), Special Sensor Microwave Imager (SSMI), and WindSAT satellites and buoys, ground observations from Karnataka State Natural Disaster Monitoring Centre (KSNDMC), as well as a combined assimilation experiment with all available observations. The accuracy of rainfall forecasts is evaluated by comparing model outputs with high-resolution telemetric rain-gauge (TRG; 6480 stations) data and other meteorological parameters against telemetric weather station (TWS; 860 stations) data from KSNDMC. Results consistently indicate underprediction of rainfall in the intricate topographical region of the Western Ghats (WG) across all experiments, contrasting with overprediction along the coastal areas of Karnataka. Assimilation experiments show positive improvements over control experiment in predicting rainfall. The experiment involving Ocean Winds showcased a substantial 40% reduction in rainfall overprediction (above 2 mm threshold). Both Ocean Winds and Station Data experiments notably enhanced rainfall prediction accuracy over most of the regions in Karnataka, with Ocean Winds exhibiting the highest improvement (53%), closely followed by Station Data (50%). Importantly, assimilating Ocean Winds and Station Data aided in reducing overprediction, while assimilating Satellite Profiles reduced underprediction in the interior part of Karnataka but increased overprediction over the coastal region compared to the control experiment. Prediction of frequency of occurrence of rainfall is considerably enhanced along the coastline due to 3D-Var assimilation. Skill scores computed indicated that largest improvement due to assimilation using Ocean Winds and Station Data when compared to control experiment. The diurnal variability in basic meteorological parameters is also better simulated in experiment which assimilated Ocean Winds and Station Data. The results underscore the crucial role assimilation of ocean winds and insitu observations in improving forecast accuracy by better temperature and moisture distribution over land during the monsoon season.