An evaluation of digital filtering and 4DVar data assimilation in the WRF model towards the simulation of tropical cyclones

Abstract

Numerical Weather Prediction (NWP) models generate enormous noises due to imbalances in the mass and wind fields at the initial stage, such as bogus gravity waves. Therefore, Initialization techniques have been developed to eliminate spurious gravity waves from the analysis. In this work, a digital filter initialization (DFI) method within the weather research and forecasting (WRF) model, known as twice diabatic filter initialization, is used to reduce the noises from the model’s initial and boundary conditions. The conventional observations and satellite radiances have been assimilated using the 4DVar method in the WRF model. The performance of the optimized model framework has been evaluated for the simulation of four very severe cyclonic storms, namely Lehar (2013), Vardah (2016), Titli (2018), and Bulbul (2019) over the North Indian Ocean. A successful 6-hourly cycling data assimilation produced superior initial and boundary conditions, and the free forecasts, based on the final cycling assimilation, showed a better simulation for selected cyclones. The DFI with 4DVar substantially enhanced the simulation of cyclone characteristics, including track, intensity, and induced rainfall. The along-track and cross-track errors in the cyclones’ simulations were reduced by approximately 20–25% in the 4DVar_DFI experiments. Thus, the results provide reasonable confidence in using DFI with the 4DVar method for the NIO cyclone’s real-time prediction.