Evaluation of Multi-Physics Ensemble Prediction of Monsoon Rainfall Over Odisha, the Eastern Coast of India

Abstract

Selecting proper parameterization scheme combinations for a particular application is of great interest to Weather Research and Forecasting (WRF) model users. The goal of this research is to create an objective method for identifying a set of scheme combinations to form a Multi-Physics Ensemble (MPE) suitable for short-term precipitation forecasting over Odisha, India’s east coast state. In this study, five member ensembles for Cloud Microphysics (CMP) and Land Surface Model (LSM, conventional ensemble) are created, as well as an ensemble of the top five performing members (optimized ensemble) for 13 Monsoon Depressions (MD) and 8 Deep Depression (DD) cases. There are a total of 30 combinations (5 PBL * 5 CMP, 5 LSM with best PBL and CMP, and one with ISRO Land Use Land Cover data). WRF 4.1 is used to carry out simulations, which are initialized with ERA5 reanalysis data and have a 72-h lead time. Rainfall verification skill scores indicate that ensemble members perform significantly better than any deterministic model. Rainfall characteristics such as location, intensity, and time of occurrence are well predicted in ensemble members as measured by a higher correlation coefficient and a lower RMSE. Neighbourhood ensemble probability also demonstrates that ensemble members have a higher chance of detecting heavy to very heavy rainfall events with more spatial accuracy. The study also concludes that choice of parameterization also affects large-scale dynamical parameters (temperature, humidity, wind, hydrometeors) and thus associated rainfall. Ensemble members exhibited less bias in the composite analysis of large-scale parameters. Furthermore, a composite analysis of moisture budget components revealed that the convergence term is the most important component of moisture accumulation, resulting in rainfall during the monsoon low-pressure system. These findings indicate that the proposed method is an effective method for reducing bias in rainfall forecasts.