Assessment of Pangu-Weather Machine Learning and NCEP GFS Dynamical Model over South Asia

This study aims to evaluate the Pangu-Weather machine learning (ML) model against the operational numerical weather prediction (NWP) model of the National Centres for Environmental Prediction Global Forecast System (NCEP GFS). The initialisation and integration of the Pangu-Weather ML model, utilising NCEP GFS and Global Data Assimilation System (GDAS) analysis, and evaluating its performance alongside NCEP GFS operational forecasts is one of the major scientific problems addressed in this study. The European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5) dataset is used to validate the 240-hour predictions made by the NCEP GFS and Pangu-Weather ML models for the South Asia region between May 1 and 15, 2024. Based on the results, surface temperature, pressure, and wind speed may all be predicted more accurately. Improvements in temperature and moisture forecasts are the primary areas where these beneficial effects are observed at various levels of the atmosphere, becoming more pronounced with longer forecast periods. The results of this study open new avenues for investigating how machine learning can be combined with more conventional dynamical models, in addition to demonstrating how effectively the Pangu-Weather ML model improves forecasting accuracy. This synergy holds the promise of advancing meteorological forecasting capabilities, particularly in regions characterised by high temperature and humidity variability.