Best Performing Models

This product contains the best performing models, in terms of forecasting the track of a storm, as determined by our system in real time using specific criteria, including how long a storm has been active. This criteria is not based on any official method. This calculation is storm specific and is a non-homogeneous comparison of early and/or late cycle models. There is only a small subset of the best performing models ranked in the mapping. Intensity error is not factored into the ranking unless there is a tie between multiple models' position error.

Only models that were released in the NHC's "aid_public" directory or the ECMWF's site, when available, are ever included in this product. This means that some models, including ones frequently used, are not included in this ranking.

This feature attempts to pick up on how well a model is performing recently for a specific storm. When there are multiple storms the model data from one storm may be from a different date than the model data for another storm. Even the model data for a specific storm can have different dates among the best performing models. If a best performing model did not have model data released at the time of the most recently released models, the latest available model run from that model will appear if it was released in the 14 hours prior to the time when the best performing models were calculated. When model data is unavailable for a model, there will be nothing mapped for that ranked model. There would be one less model that appears on the map.

Each storm has different criteria going into the model error calculation. Click on a model line for additional information on the criteria used for ranking the models for each specific storm.

Position error is averaged for up to 4 cases, depending on how long it has been since models were first available, for the forecast position we are calculating error for. As we get further from the time models were first released, we can compare more model runs and compare a later forecast hour's position. The table at the bottom of this page shows how the best performing models calculation is done depending on how many hours have passed since models were first available.

A model can only be considered in the best performing models calculation if the model has at least one case in which model error data can be calculated for the particular forecast hour position. For models that come out every 12 hours, there would be at least half as many runs to compare.

All of the models are ranked, from the model having the least amount of position error for the specific forecast hour position being the top ranked model and the model with the greatest position error ranked last. If there is a tie between multiple models the intensity error at the same forecast hour used for position error will be used as a tie breaker, with the model having the least amount of intensity error being ranked best out of the multiple models tied. Intensity error is not used in the best performing models calculation unless there is one or more ties.

Models excluded from the calculation will be removed from the ranking and the top 5 models from the final list are displayed as the best performing models.

Hours Since Models First Available	Forecast Hour To Calculate Position Error For, Number of Cases to Average and Details About Each Case
0 Hours	Default model displayed: TABM
6 Hours	Default model displayed: TABM
12 Hours	Determine error for model's 12hr forecast position, for up to only 1 run. Position error: Distance from model's 12hr forecast position made 12hrs ago (when models first available) to current storm position.
18 Hours	Determine average error for model's 12hr forecast position, for up to 2 runs for models that come out every 6 hours. Position error to average: Distance from model's 12hr forecast position made 18hrs ago (when models first available) to storm position 6hrs ago. Distance from model's 12hr forecast position made 12hrs ago to current storm position.
24 Hours 1 Day	Determine average error for model's 12hr forecast position, for up to 3 runs for models that come out every 6 hours. Position error to average: Distance from model's 12hr forecast position made 24hrs ago (when models first available) to storm position 12hrs ago. Distance from model's 12hr forecast position made 18hrs ago to storm position 6hrs ago. Distance from model's 12hr forecast position made 12hrs ago to current storm position.
30 Hours 1.25 Days	Determine average error for model's 24hr forecast position, for up to 2 runs for models that come out every 6 hours. Position error to average: Distance from model's 24hr forecast position made 30hrs ago (when models first available) to storm position 6hrs ago. Distance from model's 24hr forecast position made 24hrs ago to current storm position.
36 Hours 1.5 Days	Determine average error for model's 24hr forecast position, for up to 3 runs for models that come out every 6 hours. Position error to average: Distance from model's 24hr forecast position made 36hrs ago (when models first available) to storm position 12hrs ago. Distance from model's 24hr forecast position made 30hrs ago to storm position 6hrs ago. Distance from model's 24hr forecast position made 24hrs ago to current storm position.
42 Hours 1.75 Days	Determine average error for model's 24hr forecast position, for up to 4 runs for models that come out every 6 hours. Position error to average: Distance from model's 24hr forecast position made 42hrs ago (when models first available) to storm position 18hrs ago. Distance from model's 24hr forecast position made 36hrs ago to storm position 12hrs ago. Distance from model's 24hr forecast position made 30hrs ago to storm position 6hrs ago. Distance from model's 24hr forecast position made 24hrs ago to current storm position.
48 Hours 2 Days	Determine average error for model's 36hr forecast position, for up to 3 runs for models that come out every 6 hours. Position error to average: Distance from model's 36hr forecast position made 48hrs ago (when models first available) to storm position 12hrs ago. Distance from model's 36hr forecast position made 42hrs ago to storm position 6hrs ago. Distance from model's 36hr forecast position made 36hrs ago to current storm position.
54 Hours 2.25 Days	Determine average error for model's 36hr forecast position, for up to 4 runs for models that come out every 6 hours. Position error to average: Distance from model's 36hr forecast position made 54hrs ago (when models first available) to storm position 18hrs ago. Distance from model's 36hr forecast position made 48hrs ago to storm position 12hrs ago. Distance from model's 36hr forecast position made 42hrs ago to storm position 6hrs ago. Distance from model's 36hr forecast position made 36hrs ago to current storm position.
60 Hours 2.5 Days	Determine average error for model's 48hr forecast position, for up to 3 runs for models that come out every 6 hours. Position error to average: Distance from model's 48hr forecast position made 60hrs ago (when models first available) to storm position 12hrs ago. Distance from model's 48hr forecast position made 54hrs ago to storm position 6hrs ago. Distance from model's 48hr forecast position made 48hrs ago to current storm position.
66+ Hours 2.75+ Days	Determine average error for model's 48hr forecast position, for up to 4 runs for models that come out every 6 hours. Position error to average: Distance from model's 48hr forecast position made 66hrs ago to storm position 18hrs ago. Distance from model's 48hr forecast position made 60hrs ago to storm position 12hrs ago. Distance from model's 48hr forecast position made 54hrs ago to storm position 6hrs ago. Distance from model's 48hr forecast position made 48hrs ago to current storm position. This is then used throughout the duration of the storm.

The following models are excluded from this calculation: For a listing of most model names, click here.