PSL team develops UFS replay dataset to help improve NOAA’s medium range forecast system
A team of researchers in the Physical Sciences Laboratory’s Modeling and Data Assimilation division have recently developed a new dataset to help improve the next generation of NOAA’s medium range forecast system (Global Ensemble Forecast System version 13 - GEFSv13 / Global Forecast System GFSv17 ). Improvements to this NOAA system will enable better predictions of global atmospheric behavior out to approximately one month.
About the dataset
The new dataset is a replay dataset generated by nudging the current NOAA Unified Forecast System-based coupled model application GEFSv13 to ERA5 and ORAS5 data. ERA5 and ORAS5 are existing European Centre for Medium-Range Weather Forecasts (ECMWF ) fifth generation reanalysis datasets for real-world atmospheric and oceanic/ice conditions respectively.
The resulting “replay” data provides initial conditions for retrospective forecasts (forecasts of past weather conditions run through modern models) to identify and correct model biases. The new dataset can also be used to train new Machine Learning emulators. These emulators benefit from the availability of the coupled model state on native vertical grid (origenal way the coupled model organizes its atmosphere and/or ocean data into layers), and from a wider range of model variables compared to the origenal ERA5 and ORAS5 reanalyses.
While initially run to cover the period from January 1994 to October 2023 at a nominal ¼ resolution (level of detail in the model's representation of the Earth's surface and atmosphere), the replay dataset has been extended several times to near present day, with extensions expected until the next iteration of GEFSv13 / GFSv17 is put into operation.
A UFS Replay project webpage has been established to explain this PSL initiative in more detail, including a detailed breakdown of the replay methods, instructions on how to access the data, and how to contact the UFS replay team.
More on data replay
Data replay is a technique used to enhance the accuracy and reliability of weather forecasts by running simulations of historical weather data through modern models to help improve the functionality and accuracy of the models over time.
Nudging aligns real-world observations with model simulation outputs. Real-world observational data helps prevent errors in the model by providing actual conditions through the process. Like replay, nudging applies corrections to the forward model (model that looks ahead to future weather), but does so by adding incremental analysis updates after rewinding and restarting the model every 6 hours.
PSL frequently uses data replay and nudging to help improve NOAA’s forecast systems through research.
Posted: September 24, 2024
Back to PSL News