Hurricane Forecast Model Output, a product of the Mesoscale Research Group, UW-Milwaukee, Prof. Clark Evans.

Example Verification Plot

Example Track Verification Statistics Plot for Atlantic Hurricane Isaias (2020)


Starting in August 2020, we compute real-time verification statistics for the National Hurricane Center's official forecast and selected state-of-the-art model guidance for all classified storms in the Atlantic, East Pacific, and Central Pacific basins. Verification statistics provide a measure of what meteorologists refer to as forecast skill. A more-skillful forecast is one with lower forecast errors. So long as a skillful forecast is effectively communicated to those who would use it to take action (such as emergency managers to coordinate government response, or the general public to coordinate preparation activities), a more-skillful forecast is generally also a more valuable forecast.

We compute verification statistics for both track and intensity forecasts. The forecast skill metric for track is the position error in nautical miles, which you can multiply by 1.15 to get statute miles. This is computed as the great-circle distance between a storm's forecast and observed positions. The forecast skill metric for intensity is the absolute intensity error in kt, which you can multiply by 1.15 to get miles per hour. For forecast skill metrics that encompass multiple times, the position error and absolute intensity error are averaged over all times for which both observed and forecast data are available.

Four verification plots are provided for all classified storms. In the front page's drop-down menu for each storm, these are listed as "Track Error Statistics," "Intensity Error Statistics," "NHC Error Statistics," and "NHC Normalized Errors." The first two depict the mean position and mean absolute intensity error, respectively, for the National Hurricane Center's official forecast and selected state-of-the-art model guidance as a function of forecast hour (12, 24, 36, 48, 72, 96, and 120 hours). To provide context for the mean position and absolute intensity errors, each are also normalized by the error of a no-skill climatology and persistence forecast (OCD5). The no-skill forecast has small errors when a storm is easy to predict (in other words, it behaves like other storms in a similar location and time of year), whereas the no-skill forecast has large errors when a storm is hard to predict. The more positive the percentage improvement a forecast is relative to this no-skill forecast, the more skillful it is. Numbers atop the bars in these plots indicate the number of forecasts that entered into the calculation.

Conversely, the "NHC Error Statistics" and "NHC Normalized Errors" plots provide verification of each individual National Hurricane Center official forecast. These plots allow you to discern individual forecasts that have relatively small or large errors. The first of these plots presents the position and absolute intensity errors, whereas the second of these plots presents these errors normalized by the error of a no-skill climatology and persistence forecast (OCD5). The last five days' worth of forecasts for each storm are incomplete because the storm has not yet reached that time in reality; these times are shaded in light grey in these plots.

All verification statistics are calculated against the National Hurricane Center's working best-track, or preliminary archive, data. Official verification procedures use the finalized best-track or archive data, which is not available until after each hurricane season is complete. Thus, the final statistics you see here for each storm may not match those produced by the National Hurricane Center at the end of the season. Furthermore, no attempt is made here to compute track or intensity biases - i.e., in which direction (along or across the track, or in terms of cardinal directions) are track forecasts biased relative to observations, or in which way (more or less intense) are intensity forecasts biased relative to observations? These data may be added to the website at a later date.

Finally, past performance is no guarantee of future outcomes. The atmosphere is chaotic - literally - and thus a model that had small errors for one time and/or one storm will not necessarily have small errors for another time and/or another storm. These plots are provided for informational purposes only and should not be used to make life or death decisions.


Disclaimer: The data displayed here are informational only and should NOT be used for making life and death decisions. Always take the word of official sources - the National Hurricane Center and your local National Weather Service office - when preparing for any potential storm impact. If anything on these plots causes confusion, disregard the information in its entirety. The availability, timeliness, and reliability of these data are not guaranteed, and no liability is implied or expressed by your use of this website.


Site and Contents © 2007-2020, Clark Evans. Media use of these images is welcome with permission; please contact for details.