NHC 2006 forecast review

The National Hurricane Center released it’s annual forecast verification report today. The report reviews the accuracy of the NHC’s track and intensity forecasts, as well as those of the forecast models, for both the Atlantic and eastern Pacific basins. Here’s some highlights for the Atlantic basin:

- The track forecasts for the time periods of 12-72 hours set new records for accuracy, breaking records set in 2004. The errors for 24, 48, and 72 hours were 50.8, 97.0, and 148.7 nautical miles, respectively.

- Of the models routinely available to the forecaster’s in time to be considered for their forecast, only the consenus models beat the NHC forecast across multiple time periods.  The Florida State Super-Ensemble was second-best among these models; it was slightly edged out by the GUNA consensus, which is a simple average of the GFDL, UKMET, NOGAPS, and GFS models.

- The report also includes performance of models that are not available to the NHC before forecast time. The model from the European Center for Medium Range Weather Forecasts, commonly referred to as “The European”, displayed superior performance for the 72-120 hour time periods (it’s error at 5 days was a mere 162.7 nautical miles, while the GUNA’s  was 227.1).

- As far as overall accuracy goes, the story was the same as in previous years for the NHC’s intensity forecasts, the average error was in line with that of the past five years.  While there was an overall tendency to under-forecast storms in 2005, the reverse was true in 2006.

- As far as performance relative to their statistical baseline model goes, however, the story was (more) discouraging for the forecasters.  The forecast generated by the simple SHIFOR model was better than the NHC’s from 24 hours onwards. At the extended periods, the NHC forecast was 30-45% worse than that of SHIFOR.

- The forecast models were similarly embarassed. On average, after 24 hours, none of the state of the art models beat out the model that was designed in 1979.