GFS-Powered Machine Learning Weather Prediction: A Comparative Study on Training GraphCast with NOAA’s GDAS Data for Global Weather Forecasts

Our study represents the efforts to advance the state-of-the-art by devising a methodology for parallel training of GraphCast using Global Data Assimilation System (GDAS) data obtained from NCEP’s current operational Global Forecast System (GFS version 16). GDAS provides real-time initial conditions to make the experimental real-time MLWP global forecasts possible. Our study includes a framework that includes model training, validation, and testing processes, along with a performance comparison of GraphCast. In addition to this comparative analysis, we examine the benefits and drawbacks of GraphCast's forecasting ability using GDAS data and suggest possible ways to improve subsequent iterations of this research.