The vaunted European model did not have the best predictions for Hurricane Florence, as many meteorologists suggested. The American Global Forecast System (GFS) model was actually the most accurate, according to a National Weather Service analysis of model performance.
The analysis, obtained by the Capital Weather Gang, shows both the operational version of the GFS and the new experimental version of the model, known as the FV3, produced smaller errors than the European model in forecasts for the storm’s location at most time frames.
In fact, of the three models, the operational GFS was most accurate, the GFS FV3 (which stands for Finite Volume Cubed-Sphere dynamical core) ranked second, and the European model came in dead last. Especially two and three days before Florence struck, the European model’s forecasts for the storm’s location tended to be too far south.
The GFS (both the operational and FV3) also produced superior forecasts to the European model for the storm’s intensity at all time steps.
These results help make the point that, while the European model is in general the top-performing model, results may differ storm to storm. The European model, run at the European Center for Medium-range Weather Forecasting in Reading, United Kingdom, outperformed the operational GFS in forecasting Hurricane Lane - which dumped record amounts of rain on Hawaii in August.
While the GFS’s performance topped the European model during Florence, it’s important to note the National Weather Service analysis only extended out five days before the storm made landfall.
The analysis did not include statistics for model performance out to nine days before the storm struck which is, arguably, when the European model was at its finest. The European model was the first model to suggest Florence would make landfall in the Carolinas, while the American model incorrectly predicted it would stay out to sea.
While these various models oscillated back and forth in their projections for the storm, it’s worth pointing out that the National Hurricane Center’s forecasts were both steady and extremely accurate. Its landfall forecast for Florence, made five days out, was within an astonishing two miles of being correct.
Andrew Freedman, science writer at Axios, described the forecast as “insanely” accurate, writing: “The forecast provided residents of the Carolinas with 5 days of lead time to prepare for the storm, and alerted governors and emergency managers to start moving assets into position to respond to the storm."
The Hurricane Center, based in Miami, nailed the track forecast, and the National Weather Service’s Weather Prediction Center, based in College Park, Md., produced remarkably accurate forecasts for the storm’s tremendous rainfall.
This is clearly a case where human forecasters were able to improve on model forecasts by examining the whole universe of models and bringing in their own expertise.