A Guide to Tropical Cyclone Track and Intensity Forecast Models
By: Chris Hebert, TropicsWatch Supervisor, ImpactWeather, Inc.
Which is the "Best" Model to Use?
We're often asked that question. It's not an easy question to answer, as each model has its known strengths and weaknesses. Some of the factors which can affect the performance of the various models are:
* Tropical cyclone intensity and/or organization. Is the system very weak and disorganized or is it a strong, well-organized hurricane?
* Tropical cyclone location. Is it in the deep tropics (south of 20N latitude)? Or is it located farther north of 20N latitude in the subtropics?
* The time of year. Early season (June/July), mid season (August/September), or late season (October/November)?
* The "seasonal pattern".
With so many factors influencing model accuracy, it's difficult to say that one model will always be better than another. The tropical environment is constantly changing from day to day, week to week, even from one hurricane season to another. As the environment changes, development patterns change, and the best model to use will probably change.
Dynamic models employ the basic laws of physics as they apply to the atmosphere to predict the future course of the storm. Examples of dynamic models would be the GFDL (Geophysical Fluid Dynamics Lab), GFS (Global Forecast System - American model), UKMET (the United Kingdom Meteorological office model), ECMWF (European Center for Medium-range Weather Forecasting model), NOGAPS (Naval Operational Global Atmospheric Prediction System), or the newest hurricane model, the HWRF (Hurricane Weather Research and Forecasting model). These models work just as well in the deep tropics as in the subtropical latitudes. All of these models start with the six (6) basic equations concerning these physical laws as they apply to the atmosphere. Known as the "primitive equations" by meteorologists, these equations of 100 or more variables are anything but primitive. By knowing what is happening in the environment surrounding a tropical cyclone, these dynamic computer models attempt to predict how the atmosphere will behave over the coming days and weeks. If they can correctly predict these conditions, they can predict a tropcial cyclone's track with reasonable accuracy. Unfortunately, dynamic model predictions can only be as good as the data with which they are initialized. And, generally, data are quite sparse out across the tropical oceans. Modern reconnaissance aircraft have been a big help to model projections in recent years. The data obtained by high-altitude NOAA G-4 jets can help to better initialize the models, leading to more accurate forecasts. We've seen significant improvements in track forecasts once recon begins flying into the tropical cyclones over the past 3-5 years.
Statistical models do not employ physics. Some static models, like the BAMS/BAMM/BAMD or NHC98, are simple trajectory models. They look at the current wind flow in a shallow (BAMS), medium (BAMM), or deep-layer (BAMD) and use those steering currents to move a developing storm along. Other statistical models like CLIPER (CLImatology and PERsistence) are even more basic, merely looking at where a storm is now and comparing it to storms that were in the same location in years past then projecting a track based upon where previous storms have tracked. The static or statistical models work best in the very deep tropics, south of 20N latitude, where steering currents don't tend to change much. The BAM models, in particular, should not be trusted in the subtropics (north of 20N latitude) in most cases.
Model Accuracy -- Track Forecasts
In general, the dynamic models mentioned above have been steadily improving over the past 20-30 years as computer technology and sensors aboard reconnaissance aircraft have improved. Clearly, we are beginning to better understand the atmospheric features that help to steer hurricanes. Even the most accurate computer model is generally off by as much as 300 miles three days from landfall. Such a large error can mean the difference in hurricane-force wind, heavy rain, and a damaging storm surge or virtually no effects as the storm passes 300 miles away. By knowing the strengths and weaknesses of the better models, a hurricane forecaster can often produce a forecast that is at least a little more accurate than the best computer model guidance. In years to come, a new generation of satellites will provide detailed atmospheric sensing across the tropics, adding a tremendous amount of observational data to the model runs. In addition, new equipment aboard future reconnaissance aircraft will provide more detailed data within the tropical cyclone's environment. This should significantly increase model performance in the future. However, the launch dates of the new satellites keep getting pushed back, so it's unlikely that we'll see much benefit from the new satellite technology for perhaps another 5-10 years.
Model Accuracy - Intensity Forecasts
While track forecasts have benefited greatly from increased hurricane reconnaissance missions and improved technology, intensity forecasts have lagged far behind. Models perform much more poorly at predicting intensity because a hurricane is very sensitive to minor fluctuations in temperature, wind speed, and humidity around its core. These minor fluctuations in atmospheric conditions around the core of a hurricane are very difficult to measure with current instrumentation and they're very difficult to predict with current computer models. In general, the average error in the intensity forecast from computer models is about 15 mph per day. That can amount to a considerable error after 4-5 days. More often than not, computer models are too quick to develop a tropical disturbance, predicting much stronger winds than are actually observed. But just the opposite may occur for a well-developed tropical storm. In such a case, computer models may err on the low side in predicting intensity over a 5-day track.
SHIPS (Statistical Hurricane Intensity Prediction Scheme) is a statistical model that uses climatology, persistence, and other predictors. The primary predictors include the difference between the maximum possible intensity and the current intensity, the amount of horizontal wind shear present, persistence (the previous 12-hour intensity change), and an analysis of the upper-level wind and temperature field within about 60 miles of the center of the developing storm. Maximum potential intensity is estimated from an empirical relationship between sea surface temperature and intensity. In 2000, a new version of the model, called Decay SHIP (DSHP), was introduced. The DSHP is identical to the SHIPS model, except if the cyclone is forecast to cross land, the intensity is reduced accordingly.
The only dynamic models that do a fairly good job at predicting intensity are the GFDL and its replacement for 2007 -- the HWRF model. These two models actually do a very good job at initializing the core structure of a developing tropical cyclone. This allows them to produce a forecast that is often much more accurate than any of the statistical models like SHIPS or DSHP. But there is still much work to be done as far as predicting hurricane intensity goes. Hopefully, the next generation of satellites will eventually aid in predicting hurricane intensity.
Below is a table listing some of the more common computer models used by hurricane forecasters. We've added a short description of each model. Some models include a "TropicsWatch Tip ", just a little "inside information" on the use of the model. A complete listing of most of the NHC models that you'll see on your web page may be found here:
* NHC Tropical Models
Tropical Cyclone Models Guide
Model Type Description / Uses
GFS Global Forecast System (formerly AVN or MRF) Dynamic
This is the American model, formerly the MRF (Medium Range Forecast) or AVN model. It is run every 6 hours (0000, 0600, 1200, and 1800 UTC) and goes out to 16 days (384 hours).
TropicsWatch Tip: The GFS does tend to be a bit quick to develop tropical disturbances, and frequently it tends to turn systems to the north too quickly and possibly out to sea. But it is a good basic model for short-term forecasting (3-5 days). Beyond that time, there can be significant changes from one model run to another.
The "Interpolated" version of the American AVN (Aviation) model, which has been renamed the GFS model. As such, it is no better or worse than the GFS model upon which it is based.
GFDL Geophysical Fluid Dynamics Laboratory Dynamic
The GFDL is run every 6 hours (0000, 0600, 1200, and 1800 UTC) and forecasts out to 126 hours. It is only run when there is an active disturbance or named storm in the tropics. The GFDL model has a specialized method for initializing the center of the storm's circulation. The initial and boundary conditions are obtained form the GFS model.
TropicsWatch Tip: The GFDL model is most definitely one to consider when forecasting the path and intensity of a tropical cyclone. It was developed specifically to forecast hurricanes, and it generally does a fairly good job. In fact, the the GFDL model has been the best-performing model in 2003, 2004, and 2005.
An "Interpolated" version of the GFDL model which is available "early", before the full run of the GFDL is available.
HWRF Hurricane-Weather Research and Forecasting Dynamic
This is the model being designed to take the place of the GFDL in 2007. It has been undergoing extensive testing in 2004/2005 and will undergo final tests in 2006.
TropicsWatch Tip: We've heard good things about this new model, but it's yet to prove itself. The NHC has guaranteed that it won't replace the GFDL model unless it proves itself to be superior. It certainly has potential. Definitely worth careful consideration.
NOGAPS Naval Operational Global Atmospheric Prediction System Dynamic
NOGAPS was developed for use by the U.S. Navy. The NOGAPS model is run twice daily (0000 and 1200 UTC) and goes out to 144 hours.
TropicsWatch Tip: The NOGAPS model has not been very consistent in recent years. The NOGAPS tracks have tended to be a bit "left" (south or west) of the actual track for storms in 2004/2005, but that's no guarantee such a trend will continue in the future. Generally, NOGAPS has been in the "middle of the pack" of all the dynamic models in its forecasts for tropical cyclone movement. It's not the best model, but it's far from the worst.
An "Interpolated" version of the U.S. Navy NOGAPS model which is available "early", before the full run of the NOGAPS model is available. As such, it is no more accurate than the NOGAPS model upon which it's based.
The U.S. Navy version of the GFDL. It is initialized using the Navy NOGAPS model.
TropicsWatch Tip: The GFDN has also performed in the "middle of the pack" of models, similar to the Navy NOGAPS model upon which it is based.
An "Interpolated" version of the GFDN model which is available "early", before the full run of the GFDN model is available. As such, it is no more accurate than the GFDN model upon which it's based.
UKMET United Kingdom METeorological office Dynamic
The UKMET is the model of the United Kingdom METeorological office. It is run every 12 hours (0000 and 1200 UTC) and goes out to 120 hours.
TropicsWatch Tip: While not developed as a tropical model, it is a good dynamic model, in general. Definitely worth a look when considering the future track of a tropical cyclone.
This is what is called a "consensus" model, combining the forecast output from at least two of the following models -- GFDL, GFS, UKMET, NOGAPS, and GFDN. The hurricane forecasters at the NHC (National Hurricane Center) analyze the forecast plots from the 5 models listed above and decide if any of them appear to have serious errors. If so, then the model (or models) with the errors is removed from the consensus. Forecast positions are obtained by a simple addition of the latitudes/longitudes forecast and dividing by the number of models used.
TropicsWatch Tip: Consensus models can do a very good job of predicting tropical cyclone movement, better than any one single model. The CONU model performed very well in 2005, even better than the Florida Superensemble model.
A consensus model which adds the forecast latitudes and longitude points from the GFDL, UKMET, and NOGAPS then divides by 3 to get a single latitude/longitude pair for each forecast position.
TropicsWatch Tip: The GUNS model performed very well in 2005, on par with the CONU model.
A consensus model which adds the forecast latitudes and longitude points from the GFDL, UKMET, NOGAPS, and AVN (GFS) model then divides by 4 to get a single latitude/longitude pair for each forecast position.
TropicsWatch Tip: The GUNA model performed very well in 2005, on par with the CONU and GUNS models.
ECMWF European Center for Medium-range Weather Forecasting Dynamic
The ECMWF is run twice daily at 0000 and 1200 UTC and forecasts out to 10 days.
TropicsWatch Tip: Generally, it's a pretty good model for most of the year, but it may not be one of the best tropical models. Another negative is that it takes from 6-8 hours to initialize and run, so it is the last of all models to arrive. But definitely worth a look.
CMC Canadian Meteorological Center Dynamic
Generally referred to as the "Canadian model". The Canadian model was designed more for winter-type systems in northern latitudes rather than for tropical cyclones.
TropicsWatch Tip: We've noted that the Canadian model has a strong tendency for over-developing weak tropical disturbances. So watch out for many false alarms. However, it's not necessarily a bad model to use for existing named storms.
WRF Weather Research and Forecasting (Formerly NAM and ETA) Dynamic
This is the new American (U.S.) model that replaced the NAM (North American Model) on June 20th, 2006. It is not designed for use in forecasting tropical cyclones.
TropicsWatch Tip: Because this is not a tropical model, it tends to perform very poorly in predicting the development and movement of tropical cyclones. This is one model that it probably would be best to ignore as far as tropical cyclone predictions go.
LBAR Limited area BARotropic Dynamic
This model is a bit different from the other dynamic models. It is run every 6 hours (0000, 0600, 1200, and 1800 UTC) and goes out to 120 hours.
TropicsWatch Tip: By far, the LBAR is the poorest of the dynamic models at predicting the path of most tropical systems. Use this model with extreme caution. In fact, it would be best not to use this model for anything.
BAMS / BAMM / BAMD Beta and Advection Models Trajectory
The BAM follows a trajectory from the Aviation run of the American GFS model to provide a track forecast. This model incorporates a correction known as the "Beta Effect". This is used to account for the fact that the Coriolis force resulting from the rotation of the Earth is greater toward the poles, so the winds on the northern side of the storm in the Northern Hemisphere tropical cyclone are turned more than those on its southern side. If no other winds were steering a tropical cyclone, the "Beta Effect" would cause a westward-headed storm to drift toward the north in the Northern Hemisphere, and toward the south in the Southern Hemisphere. There are three (3) versions of the BAM...
BAMS - the BAM Shallow, this version averages winds from 5,000 to 10,000 feet (850 - 700 mb). The BAMS is best-used for tropical disturbances/tropical waves or tropical depressions. Such systems are very shallow in nature, moving with the lower-level winds.
BAMM - the BAM Medium, this version averages winds from 10,000 to 24,500 feet (850 - 400 mb). The BAMM is best-used for tropical storms of weak to moderate strength or disorganized hurricanes. Tropical storms are generally steered by the mid-level steering flow.
BAMD - the BAM Deep, this version averages winds from 24,500 to up to 47,000 feet (850 - 200 mb). The BAMD is best used for well-organized and/or stronger hurricanes. Such hurricanes are steered by the mid-level to upper-level wind flow.
Typically, the NHC will run the BAM models on a tropical disturbance just to get a general idea where this disturbance might track. Since the BAM is a very simple trajectory model, it is best used where the steering currents are less likely to change in the future. This would typically be in the deep tropics south of 20N latitude. Because the BAM models are not dynamic, they should be used with extreme caution outside of the deep tropics where steering currents may change with time.
None Don't be fooled by this "model", as it's really not a model at all. It's a simple extrapolated track that assumes the storm just keeps going at the same speed and direction. No physics at all are involved. Generally of no use except as a "what if" scenario.
NHCA98E / NHCA90E Climatology
Climatology-based "models". Not really models, they just compare the current storm position to previous storms and "guess" where it might go. Generally, these "models" are not used by hurricane forecasters when predicting where a tropical cyclone will go.
CLIPER CLImatology & PERsistence Climatology Persistence
CLImatology and PERsistence. This model employs no physics, it's not a real model. The predictors for CLIPER include the initial latitude and longitude of the storm, the components of the storm motion vector (or which direction it is moving), the day of the year, and the initial storm intensity. The CLIPER forecasts are used to normalize the output from the other forecast models and as a benchmark for tracking forecasting model skill. This is the simple type of model that most hurricane tracking software programs offer. It consists of a set of equations that separately predict future zonal (east-west) and meridional (north-south) movements of a tropical cyclone at 12-hr intervals out to 72 hours. The predictors include the current and previous 12-hr position, the current and 12-hr previous storm motion, the day of the year, and the maximum surface wind. The initial motion of the storm (persistence) is the most important predictor for this model. The skill of more complex forecast models is often compared to that of CLIPER. Any model that cannot demonstrate significant skill over CLIPER's combination of climatology and persistence is discarded.
Storm2k's very own reference library of links, terminology and basic storm information.
Moderator: S2k Moderators
1 post • Page 1 of 1
I hope this is okay to post here. I'm not a met, but I got this from a site that aims to explain the models to its non weather understanding clients. Ergo even I could understand it, and given the obsession with a lot of folks on the models when they really don't understand the basics of their predictions its a really good guide on how to read and take the information they give you. Its posted in full as you'd have to be client to hotlink and I'm sure my friends company wouldn't like it very much if I gave out the password.
Who is online
Users browsing this forum: No registered users and 1 guest