Analysis and Verification of GraphCast Model Forecasts with Mid-Latitude Cyclone Cases

The usage of machine learning (ML) and artificial intelligence (AI) in the atmospheric sciences has propelled a diversity of avenues in research that continues to expand rapidly. One of these avenues relates to Research to Operations (R2O). Currently, the performance of AIWP models has been studied with forecasting extreme weather events such as tropical cyclones and atmospheric rivers, where these models perform well with predicting track but less well for intensity. Although current AIWP model performance exceeds that of some operational models as measured by large-scale metrics (e.g., RMSE), careful study of details regarding weather extremes is needed. Additionally, most AIWP models have been trained using reanalysis datasets such as the ERA5. Analyzing the impacts of new training data for fine tuning current models is needed. In this study, a list of mid-latitude cyclone cases from 2022-2023 were selected and analyzed using forecasts of the GraphCast model initialized with Global Forecast System (GFS) data by the Cooperative Institute for Research in the Atmosphere (CIRA). Forecasts were taken from 10 days before an event to the day of the event and compared to ERA5 Reanalysis data to identify the difference of location, timing, and intensity of atmospheric variables relevant to the life cycle of a mid-latitude cyclone. Additionally, some station data or other observations will be compared with these forecasts to further verify the model.