Evaluation of 3D-Var and 4D-Var data assimilation on simulation of heavy rainfall events over the Indian region

Abstract

The present study delineates the relative performance of 3D-Var and 4D-Var data assimilation (DA) techniques in the regional NCUM-R model to simulate three heavy rainfall events (HREs) over the Indian region. Four numerical experiments for three extreme rainfall cases were conducted by assimilating different combinations of observations from surface, aircraft, upper-air and satellite-derived Atmospheric Motion Vectors (AMVs) using 3D-Var and 4D-Var techniques. These experiments generated initial conditions (ICs) for the NCUM-R forecast model to simulate HREs. Key atmospheric variables, such as wind speed and direction, vertically integrated moisture transport (VIMT: kg.m⁻¹.s⁻¹), vertical profiles of relative humidity and temperature as well as various stability indices are analysed during the HREs. Forecast verification was performed using statistical skill scores and object-based methods from the METplus tool, comparing NCUM-R output against GPM rainfall data. The results demonstrate that the 4D-Var technique improves simulation accuracy compared to 3D-Var, particularly when assimilating satellite wind data. Incorporating satellite-derived AMVs improved the representation of rainfall intensity and spatial patterns, as well as other atmospheric variables. It is found that rainfall for Case-01, the VIMT was notably high along the eastern coast of India and southwest of BoB, with the 4DVS simulation better capturing moisture transport patterns compared to 3DVS and 3DV. The SWEAT index ranged from 205 to 250 J·kg⁻¹ in the morning, rising to 250–300 J·kg⁻¹ by noon, indicating increasing convective instability. On 18 March 2023 (Day-1), the K-index exceeded 30, signalling scattered thunderstorms, consistent with the IMD's reports of isolated to scattered rainfall on 19th and 20th March 2023. Similarly, it is found that satellite wind assimilation improved the statistical skill scores in predicting heavy precipitation in all three cases. Overall, the study suggested that the performance of the NCUM-R model integrated with the 4D-Var technique improved the model's forecast skill in the simulation of HREs.