Note: The NWS provides links to certain external Web pages because they may contain related information of interest to you. These links do not constitute an endorsement by the NWS of any information, products or services on those external Web sites. SR/SSD-95-36
7-15-1995
Latest Revision 7-27-2006
Technical Attachment

AN OVERVIEW OF NHC PREDICTION MODELS

Bernard N. Meisner
Scientific Services Division
National Weather Service Southern Region

Introduction

Storm Track Guidance Models
HURRAN | CLIPER | NHC98 | BAM | LBAR | GHM (GFDL) | GFS (formerly AVN) | NOGAPS | GUNS, GUNA and CONU Ensembles
Relative Skill of the Statistical | Numerical Guidance Models

Storm Intensity Guidance Models
SHIFOR | SHIPS | GHM (GFDL)
Relative Skill of the Intensity Guidance Models

Forecast Verification

Storm Surge Guidance Model
SLOSH

Acknowledgements

Additional References

Introduction

Meteorologists at the National Hurricane Center (NHC) have a variety of prediction models available to provide guidance for their forecasts of tropical cyclone tracks and intensity. The intent of this paper is to provide a brief overview of each of the models. Forecasters may find this information helpful when considering NHC discussions which mention the performance of individual models. A primary reference is provided after the summary of each model for readers who desire more information. NOTE: All thumbnail graphics in this Web document are linked to larger version of the graphics. Just click the thumbnail to view the larger version.

As noted by Neumann (1979), models for the prediction of tropical cyclone motion and intensity may be classified as either statistical or dynamical. Statistical models rely on what has happened-the climatology of past storms, for example. Dynamical models can be classified as either barotropic or baroclinic. Statistical-dynamical models are an intermediate class that incorporate numerically forecast data into a statistical prediction framework, similar to the Model Output Statistics used to provide guidance for specific parameters such as temperature and probability of precipitation.

Storm Track Guidance Models

HURRAN - The Hurricane Analog Technique

The HURRAN climatological model was developed at the National Hurricane Center in 1969. By identifying previous storms with characteristics in common with a current storm, HURRAN attempted to determine the most likely track of the current storm.

The technique first searched the database containing the tracks of all Atlantic tropical storms and hurricanes for analogs to the current storm. The analogs were those storms with similar location, speed and direction of motion, that occured at about the same time of year Map showing tracks of analog storms. The tracks of the analog storms were shifted to pass through the location of the current storm Map showing tracks adjusted to current storm's location. Finally, probability ellipses were calculated based on the adjusted tracks of the analog storms Map showing probability ellipses for 12, 24 and 36 hour forecast storm positions..

The HURRAN model predicted the tracks of "well-behaved" storms (those south of 25oN, before recurvature) well. Since HURRAN lacked any synoptic input it did not perform well in predicting tracks of storms just before or after their recurvature. While it provided useful information to the hurricane specialists at the time, the skill of HURRAN is so poor that the model no longer used by the National Hurricane Center.

Reference: Hope, John R. and Charles J. Neumann, 1970: An operational technqiue for relating the movements of existing tropical cyclones to past tracks. Monthly Weather Review, 98(12), Dec. 1970, pp 925-933.

CLIPER/CLIPER5 - A Combination of Climatology and Persistence

CLIPER is a statistical track prediction model based on climatology and persistence. 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 hrs. Equations were developed in 2001 to extend the CLIPER forecasts to Day 5 (120 hours) in preparation for the release of official forecasts out to Day 5 in 2003. 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 original CLIPER equations were developed based on historical storm track data for all storms in the North Atlantic Ocean, Carribean Sea and Gulf of Mexico that persisted for at least five days during the period 1931-1970.

Here are some examples of CLIPER's forecasts: Sample_Tracks

Storm "A," Hurricane Edith (1971), has a well-behaved track toward the west-northwest and was well forecast by CLIPER. Storm "B," Hurricane Ginger (1971) tracked to the southwest; however, CLIPER forecast a northwest track. Storm "C," a stationary storm, is gradually forecast to accelerate to the northeast by CLIPER. The two forecast tracks of Storm "D" show how CLIPER forecasts more intense storms to move northward more rapidly than weaker storms. Storm "E," a storm initially moving towards the southeast, is forecast by CLIPER to return to a more typical northwest track.

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. CLIPER can also account for relatively easy or difficult seasons. A candidate prediction model, tested during a difficult year, with somewhat large absolute track errors, might be retained for future use if its errors relative to CLIPER were small.

Here is a plot of CLIPER's skill for the 1996-2002 Atlantic hurricane seasons: Plot of the CLIPER model errors by year. Note that CLIPER's errors for the 2000 season were much less at most verification times than those for the other years, indicating that the 2000 season was a relatively easy one compared to the others.

In preparation for the official release in 2003 of the Day 4 and Day 5 forecasts by the National Hurricane Center, CLIPER was substantially updated to CLIPER5. The new version of the model is based on data from 1956-1995, the prediction equations for all verifying times were changed, and new predictors were added. CLIPER5 has much smaller forecast biases, but similar forecast errors, than the original CLIPER.

References:
Aberson, S. D., 1998: Five-day tropical cyclone track forecasts in the North Atlantic basin. Wea. Forecasting, 13, 1005Ė1015.
Neumann, C. J., 1972: An alternate to the HURRAN tropical cyclone forecast system. NOAA Tech. Memo. NWS SR-62, 22 pp.

NHC98 - A Statistical-Dynamical Hurricane Track Prediction Model

The NHC98 model is the latest in a series of mixed statistical-dynamic track prediction models. Earlier versions included the statistical models NHC67 and NHC72, and the statistical-dynamical models NHC73, NHC83 and NHC90.

In NHC98, storms are stratified based on their latitude and their current motion, with different equations used for westward and eastward-moving storms. This stratification is used to account for the observation that storms within the easterlies tend to move to the right of the steering flow, while storms within the westerlies tend to move to the left of the steering flow. Map showing North and South Zones used in NHC98

South Zone equations are used for storms south of 15oN, and for storms between 15oN and 25oN that are moving to the west or northwest. North Zone equations are used for storms north of 25oN, and for storms between 15oN and 25oN that are moving to the north or northeast.

The NHC98 model produces a forecast track that is a combination of three independent track estimates. The first estimated track is that produced by CLIPER.

The second estimated track is predicted using observed deep layer mean geopotential heights from the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS). Similar to CLIPER, the storm motion is separated into two components. One set of equations is used to predict the along-track movement of the storm at 12-hr intervals out to 120 hr. Another set of equations is used to predict the across-track movement of the storms. Deep-layer mean geopotential heights at two or three locations in the vicinity of the storm are used to represent the mean flow in which the storm is imbedded. Separate equations are used for each time period, but the geopotential height predictors are consistent from one time period to the next in order to avoid abrupt shifts in the predicted storm locations. These two diagrams show the grid points used for the first four time periods for the along track NHC90_Chart and cross track NHC90_Chart equations for the South Zone equations. The grids points used in the equations were determined by overlaying the grid shown on historic storm tracks and correlating the deep-layer mean heights at each grid point with the actual storm motion. The grid points that best correlated with storm motion are included in the prediction equations.

The third estimated track is computed similar to the second, except forecast deep-layer mean geopotential heights from the NCEP GFS are used to produce the forecast track. This sample shows the grid points used for the 36 hr, North Zone equations. Map showing the grid points used in the 36 hours North Zone NHC98 equations (Starting with the NHC98 version, the circulation of the tropical cyclone is removed from the numerical analysis before the deep layer mean geopotential heights are determined.)

NHC98 combines the three track estimates (that from CLIPER, that based on the current geopotential height analysis, and that based on the forecast geopotential heights) into an optimum track forecast. NHC90_Chart

NHC98 is run four times per day. The primary synoptic time NHC98 forecasts (0000 and 1200 UTC) are based on the previous (1800 and 0600 UTC) runs of the NCEP Global Forecast System. A special version, NHC98-LATE, is run at the primary synoptic times using forecasts from the current GFS model run and is available several hours after NHC98.

Reference: Neumann, C. J. and C. J. McAdie, 1991: A revised National Hurricane Center NHC83 Model (NHC90). NOAA Tech. Memo. NWSNHC-44,35 pp.

BAM - The Beta and Advection Model

The Beta and Advection Model is a baroclinic-dynamical track prediction model. It produces a forecast track by following a trajectory in the vertically averaged horizontal wind starting at the current storm location out to 120 hours. The trajectory is corrected to account for the variation of the Coriolis force with latitude, the so-called Beta effect. (Beta is the Greek letter frequently used in meteorological equations to represent the change in the Coriolis parameter with latitude.)

The figure shows how the conservation of absolute vorticity results in the formation of anticyclonic relative vorticity in the northeast quadrant of the storm, and the formation of cyclonic relative vorticity in the southwest quadrant of the storm: Diagram of absolute vorticity advection and relative vorticity formation in the vicinity of a tropical cyclone.. The result adds a component of motion to the northwest to the storm's trajectory.

Three versions of the BAM model are run with shallow (850-700 mb), medium (850-400 mb), and deep (850-200 mb) layers. All three versions of the model are run operationally four times per day.

Reference: Marks, D. G., 1992: The beta and advection model for hurricane track forecasting. NOAA Tech. Memo. NWS NMC- 70, 89 pp.

LBAR - A Nested Barotropic Hurricane Track Forecast Model

LBAR is a dynamical track prediction model. LBAR is the NHC's implementation of the GFDL VICBAR model. (VICBAR stands for Vic Ooyama's Barotropic model.) The model is initialized with deep layer mean winds and geopotential heights from a mass-weighted average of the 850, 700, 500, 400, 300 and 200 mb levels. Analyses are produced on three nested grids: (1) a fixed synoptic domain 27.5 S to 67.5 N, 10 E to 140 W; (2) a storm environment 50 degree latitude-longitude box centered on the current storm location; and (3) a vortex domain circle of 7.4 degree (about 800 km) radius centered on the current storm location. Domain_Map

The synoptic scale analysis is obtained directly from the NCEP global spectral model analysis. The storm environment domain analysis is produced with a two-dimensional spectral application of finite element representation, using all available data (rawindsondes, cloud drift winds, aircraft observations, etc.), with the NCEP global model analysis used as a low level background field. The vortex domain analysis consists of synthetic observations representing storm circulation and current storm motion. The vortex is prescribed to be the same size and intensity in all directions (axisymmetric), with winds increasing linearly from the center to the radius of maximum winds. Wind speeds beyond the radius of maximum winds are prescribed to decrease exponetially to the edge of the storm. In the event of multiple tropical cyclones, synthetic vorticies are included for each storm.Rankine_Vortex

The simplicity of barotropic models means they can be run quickly on inexpensive computers. In the LBAR prediction model, the shallow water equations are solved on a series of nested grid meshes on a Mercator projection. The inner meshes move to remain centered on the storm, while the outer mesh is fixed geographically. Time-dependent boundary conditions from the NCEP Global Forecast System model run are applied outward from a transition zone between 1500 and 2500 km. LBAR runs on a 6-hr forecast cycle and produces forecasts out to 120 hr.

Strengths: LBAR runs quickly (the hurricane specialists can view the output of the 1200 UTC LBAR run before they have to complete their 1500 UTC package). LBAR performs best early in the hurricane season (before fronts penetrate into the subtropics) and on storms that move primarily westward and only move slowly northward. LBAR outperforms all the statistical track guidance models, and its skill in the 12-36 hr time frame is comparable to that of the more complex baroclinic models.

Weaknesses: LBAR does not perform well whenever there is significant vertical wind shear, or when there are multiple, interacting storms. Interacting_Storms

Reference: DeMaria, M. S., S. D. Aberson, K. V. Ooyama and S. J. Lord, 1992: A nested spectral model for hurricane track forecasting. Mon. Wea. Rev., 120, 1628-1643.

Relative Skill of the Statistical Guidance Models

These graphs shows the relative skill of the statistical guidance models and LBAR for the 1996-1998 Atlantic seasons, Skill Chart for 1996-98 Atlantic seasons the 2000 season, Skill Chart for 2000 Atlantic season the 2001 season, Skill Chart for 2001 Atlantic season the 2002 season, Skill Chart for 2002 Atlantic season the 2003 season, Skill Chart for 2003 Atlantic season the 2004 season Skill Chart for 2004 Atlantic season and the 2005 season Skill Chart for 2005 Atlantic season.
Note that the statistical models have errors about 20-40% less than those of CLIPER/CLIPER5 and that the skill of the BAM and LBAR models varies widely from year to year.

GHM - The GFDL Multiply-Nested Moveable Mesh Hurricane Model

The GHM is a dynamical baroclinic track prediction model. The model also produces experimental forecasts of hurricane intensity, precipitation, and wind swath maps that show the distribution of predicted maximum surface and boundary layer winds. The GHM was developed by NOAA's Geophysical Fluid Dynamics Laboratory at Princeton University. The GHM is a triply nested, moveable mesh primitive equation model formulated in latitude, longitude, with the terrain following sigma vertical coordinates.Graphical representation of physical processes included in the GHM model

The grid configuration of the GHM was most recently modified prior to the 2005 Atlantic hurricane season. Under the restored three-nest grid configuration the innermost grid covers five degrees in latitude and longitude with a grid spacing of 1/12 of a degree (about nine kilometers). The middle grid covers 11 degrees with a grid spacing of 1/6 degree. The outermost grid extends over nearly a fifth of the globe (75o in latitude and longitude. Model_Grids In 2003 the vertical resolution was increased from 18 to 42 vertical levels, and the track skill improved substantially for all verification times.

The storm is centered in the middle of the finest grid at the start of an integration. Lateral boundary conditions are obtained from the NCEP Global Forecast System. There is two-way interaction among the grids, i.e., features that form during an integration on the innermost grid are passed to the other grids, and vice-versa. The inner and middle grids move along with the storm. Multiple storms with a basin can be handled by a single run of the model by introducing separate moveable inner and middle grids for each storm.

Numerical experiments have shown that typical initial position errors have only minor effects on track forecast errors beyond 12 hr. However, errors in the initial storm motion have an effect on the forecast error out to 72 hr and beyond. The GHM model attempts to resolve those errors by replacing the poorly resolved vortex in the coarse resolution analysis of the NCEP model (too large, too weak, wrong location) with a more realistic vortex constructed to match the high resolution GHM model. Vortex_Replacement

The specified vortex has structural consistency--there is a smooth connection of the environmental field from the storm area to the surrounding domain. The vortex resembles the corresponding real storm and is compatible with the grid resolution, computational schemes, and physics of the prediction model. The use of the specified vortex eliminates the initial adjustment and false spin-up of the model to the coarsely analyzed vortex in the NCEP global model and results in improved initial track prediction. Storm asymmetries are represented in the current version of the GHM model.

The GHM forecasts are available about five hours after the primary and intermediate synoptic times (0000, 0600, 1200 and 1800 UTC). To overcome this shortcoming, the Tropical Prediction Center has developed an interpolation technique to transpose the forecast from the previous run to the current storm position. This procedure is used for all the "late" models (i.e., those that depend on the GFS model for their lateral boundary conditions).

In 2001, the GHM model had the benefit of running from an updated and improved version of NCEP's global model (see below). In addition, the version of the GHM used for Atlantic basin storms was coupled with a version of the Princeton Ocean Model. The uncoupled version of the GHM model will still be used for Eastern Pacific storms. The GHM models for both basins use the recently-upgraded GFS model for their initial and boundary conditions.

The new, coupled version of the model allows the sea surface temperature to evolve throughout a forecast integration. The evolution of the sea surface temperature can have a strong impact on the intensity of a storm. Previous uncoupled versions of the hurricane model essentially left the sea surface temperature constant throughout the duration of a forecast. Results have indicated that the ocean coupling has a positive influence on the skill of intensity forecasts Forecast intensity skill comparison graph. The new coupled model should also have better skill in track forecasts Forecast intensity track comparison graph

Reference: Bender, Morris A., Timothy P. Marchok and Robert E. Tuleya, 2002: Draft changes to the GFDL hurricane forecast system for 2002. NWS Technical Procedures Bulletin No. 492. NWS Office of Meteorology, Silver Spring, MD, 16 pp.
Tuleya, Robert, M. Bender, Y. Kurihara and S. Lord, 1995: The GFDL Multiply-Nested Moveable Mesh Hurricane System. NWS Technical Procedures Bulletin No. 424. NWS Office of Meteorology, Silver Spring, MD, 22 pp.
Kurihara, Yoshio, et al., 1995: Improvement in the GFLD Hurricane Prediction System. Monthly Weather Review, 123(9), Sept. 1995, pp 2791-2801.

GFS - NCEP's Global Forecast System

The numerical model used for NCEP's global data assimilation system and for the aviation (AVN) and medium-range forecasts (MRF) is the global spectral model, now known as the Global Forecast System (GFS). As one might guess from its name, the "aviation model" was not specifically developed to predict hurricane motion or intensity. Rather, one of the primary uses of the GFS is to produce forecasts for aviation guidance worldwide. The GFS model is run four times each day at the primary and intermediate synoptic times (0000, 0600, 1200 and 1800 UTC) with a wait of 2.75 hr for data arrival. Forecasts are made out to Day 16.

The GFS is a baroclinic-dynamical model. Like the GHM, the model is a primitive equation model which predicts winds, temperature, surface pressure, humidity, and precipitation. The prediction equations include the divergence and vorticity equations, the hydrostatic equation, the thermodynamic equation, a mass continuity equation, and a conservation equation for water vapor.

The GFS differs from the GFDL Hurricane Model (GHM) model in that it has a global domain, and the fields within the model are represented by a set of mathematical (sine and cosine) functions rather than values at discreet grid points. The forecast equations are solved for the coefficients of the mathematical functions.

The GFS currently is configured to handle 382 triagular waves across the globe (comparable resolution to a grid point model with a grid spacing of 37 km) and has 64 vertical levels. For integrations between 7 1/2 and 16 days the horizontal resolution is reduced to 190 triangular waves.

In July, 2000, a numerical scheme was implemented to change how tropical cyclone vorticies are initialized in the global spectral model (Technical Procedure Bulletin). In the past, bogus observations based on the National Hurricane Center's estimates of storm location, intensity and size were input to the model's analysis scheme. This has been replaced by a procedure that relocates the vortex in the "first guess" field (the forecast from the previous run of the model) to the correct location.

The relocation procedure takes the model guess field and moves the hurricane vortex to the correct location before the model's analysis is completed. The steps can be briefly summarized as:

  1. Use a spectral filter to separate the total wind field into Basic and Disturbance fields. (Long waves predominate in the Basic wind field while short waves predominate in the Disturbance wind field.)
  2. Locate the hurricane vortex center in the Disturbance wind field,
  3. Separate the hurricane model's vortex from the non-hurricane component in the Disturbance wind field,
  4. Combine the Basic wind field and non-hurricane component of the Disturbance wind field into the Environmental wind field.
  5. Move the extracted hurricane vortex to the NHC official position.
  6. If the vortex is too weak in the guess field, add bogus observations to the model analysis.
  7. The data assimilation scheme uses the revised guess field and all available observations to produce the final analysis for input to the forecast model.
This animated GIF Vortex replacement procedure flow chart depicts the vortex relocation procedure in a flow-chart format.

The vortex is not relocated if the center of the hurricane is over a major land mass or if the topography in the filtered domain is greater than 500 m in elevation. When the procedure was tested in retrospect on the 1999 hurricane season forecasts, the average track forecasts improved by approximately 30% over that of the operational AVN model. The skill of the AVN track forecasts during the 2000 Atlantic season were significantly better than those of previous years. Skill of AVN track forecasts for 1995-2000

In May, 2001, momentum mixing was included in GFS model's cumulus parameterization scheme. Tests showed this reduces tropical storm false alarm forecasts. Additional changes made to the model at that time were expected to improve the skill in tropical circulation forecasts at all time ranges.

In July, 2006, the vertical coordinate of the GFS was changed from terrain following sigma to a hybrid sigma-pressure coordinate system. The hybrid coordinate is being adopting by modeling centers around the globe and provides improved performance in upper troposphere and stratosphere. Retrospective runs made of some 2005 hurricanes suggest the forecast track skill will not change, while forecast strength may be better than before for hurricanes. Reference: Kanamitsu, M., 1989: Description of the NMC global data assimilation and forecast system. Wea. Forecasting, 4, 335-342.

NOGAPS - The Navy's Operational Global Atmospheric Prediction System

The NOGAPS model was not designed specifically to predict the motion of tropical cyclones. Rather it is the Navy's operational global atmospheric prediction system. The NOGAPS model is run four times daily every day of the year, producing forecasts out to 144 hours.

In 2002 the resolution of the NOGAPS model was increased from T150L24 to T239L30 (239 waves, which is about 55 km horizontal resolution with 30 vertical levels). In September 2003, the NOGAPS optimal interpolation data assimilation system was upgraded to the NRL Atmospheric Variational Data Assimilation System, a three-dimension variational system. Additionally, in November 2003, NOGAPS implemented the use of terrain fields from USGS Global Land One-kilometer Base Elevation database and changed the gravity wave drag scheme. The result of these changes has been an average increase in verification scores and fewer bad forecasts. The accompanying table provides a summary and relevant references for the NOGAPS model.

The NOGAPS model initializes the tropical cyclone using synthetic soundings based on the National Hurricane Center's estimates of storm location and intensity. These soundings are automatically inserted into the model in the vicinity of the tropical storm. The artificial atmospheric soundings are constructed at the storm center and at radii of two, four and six degrees from the storm The winds are derived from a specified vortex which has a radius of maximum winds of 50 km. Due to the horizontal spatial resolution of the NOGAPS model the maximum wind speeds inserted into the model are 60-80% of those observed.

The GUNS, GUNA and CONU Ensembles - An Average of the GFDL, UKMET Office, NOGAPS, GFS Models

GUNS. James Goerss of the Naval Research Laboratory in Monterey, California, demonstrated that a simple consensus of the GFDL, UKMET and NOGAPS models was about 20% more accurate at 24, 48 and 72 hrs than the best of individual models. The National Hurricane Center confirmed his results and dubbed the ensemble "GUNS," using the initials of the three models. Relative skill of the GFDL, UKMET Office, NOGAPS models and the GUNS ensemble track forecasts 1995-1998

Consensus forecasts, on average, are often more accurate than the forecasts from individual models, and the spread of an ensemble has potential use as a measure of confidence in the forecast. The following two diagrams show the track guidance and verification for two time periods during the lifetime of Hurricane Georges (1998). The first diagram shows the track guidance available at 1800 UTC on September 22nd. The spread of the guidance tracks was relatively small and the 72 hr forecast of the GUNS ensemble (the center of the cyan triangle) was very close to the actual storm track (indicated by the large red dot). Plot of guidance model tracks, official forecast and actual storm track for the 72 hour period starting 1800 UTC September 22, 1998

The second diagram shows the track guidance available about four days later, at 1200 UTC on September 26th. In contrast to the first diagram, the spread of the guidance tracks was relatively large and the 72 hr forecast of the GUNS ensemble (the center of the cyan triangle) was far from the actual storm track (indicated by the large red dot). Plot of guidance model tracks, official forecast and actual storm track for the 72 hour period starting 1800 UTC September 22, 1998

GUNA. Once the "Aviation" run of the NCEP's global spectral model (now called the GFS - Global Forecast System) was improved in 1998 it became one of the best track guidance models. As a result, Jim Goerss added the Aviation model to the GUNS ensemble, renaming it GUNA for GFDL, UKMET, NOGAPS and AVN. Relative skill of the GFDL, UKMET Office, NOGAPS and AVN models and the GUNA ensemble track forecasts 2003

CONU. More recently Jim has created an ensemble called "CONU" (pronounced "CON-you"). This "consensus" ensemble is computed when track forecasts from at least two of the five models (the NCEP GFS and GFDL models, the Navy's version of the GFDL model and NOGAPS, and the UK Met office model) are available. The CONU tries to take advantage of the ensemble forecasts even if one or a few of the members is not available. Relative availability of GUNA and CONU ensembles 200102003 The skill of CONU is comparable to that of the GUNA ensemble. Relative skill of the Official, GUNA and CONU track forecasts 2003

Reference: Goerss, James S., 2000: Tropical cyclone track forecasts using an ensemble of dynamical models, Mon. Wea. Rev., April 2000, pp 1187-1193

Relative Skill of the Numerical Guidance Models

These graphs shows the relative skill of the numerical guidance models for the 2000 Atlantic season, Skill_Chart the 2001 Atlantic season, Skill_Chart the 2002 Atlantic season. Skill_Chart the 2003 Atlantic season, Skill_Chart the 2004 Atlantic season, Skill_Chart and the 2005 Atlantic season. Skill_Chart. (The charts for 2001 and later years include data out to 120 hours since in 2001 the Tropical Prediction Center began testing their techniques prior to the release of official Day 4 and Day 5 forecasts in 2003.) Note that, except for the GUNA ensemble, the skills for the earliest time periods are comparable to, or even less than, those of the statistical models. However, the skill of the numerical models greatly exceeds those of the statistical models at the longer lead times.

Intensity Guidance Models

The National Hurricane Center now produces an experimental storm intensity graphic. The Wind Speed Probability Table Wind_table shows the probability that the maximum one-minute wind speed of the tropical cyclone will be within any of eight intensity ranges during the next 72 hours. It is based on the outcomes of similar NHC wind speed forecasts during the period 1988-1997. The data base excludes unnamed tropical depressions. NA indicates data are not available. TF indicates too few (< 10) similar forecasts during 1988-1997 to yield reliable results. Note this product refers to the entire storm, without regard to any particular geographic location.

In 2006 the NHC began operational distribution of wind speed probability graphics. Note these products refer to particular geographic locations. As noted on the NHC Web site:

"These graphics are based on the official National Hurricane Center track, intensity, and wind radii forecasts, and on NHC forecast error statistics for those forecast variables during recent years. Each graphic provides cumulative probabilities that wind speeds of at least storm force (34 knots), gale force (50 knots) or hurricane force (64 knots) will occur during cumulative time periods at each specific point on the map. The cumulative periods begin at the start of the forecast period and extend through the entire five day forecast period at cumulative 12-hour intervals.

"The previously provided strike probability product (discontinued after 2005) conveyed the chances of a "close" approach of the center of the cyclone. However, these new probability products are about the weather. That is, these cumulative wind speed probabilities provide the chances that wind speeds of at least 74 mph will occur at individual locations. The cumulative probabilities can answer the question, "What are the chances that sustained winds of tropical storm or hurricane force will occur at any specific location?" This can also help one answer the question, "Do I need to take certain actions to prepare?" A companion product, the wind speed probability text product, will also be issued and updated with each advisory package. That product is recommended to more easily assess when winds of each threshold are most likely to start at any specific location, helping to answer the question, "How long do I have to prepare?" Overall, these probabilities provide users with information that can enhance their ability to make preparedness decisions specific to their own situations.

"It is important for users to realize that wind speed probabilities that might seem relatively small at their location might still be quite significant, since they indicate that there is a chance that a damaging or even extreme event could occur that warrants preparations to protect lives and property."

Links to wind speed table and probability graphics for any current storms may be found on the NHC Web page, and our NWS Southern Region Tropical Weather Web page.

Statistical Hurricane Intensity Forecast (SHIFOR) Model

As is evident by its name, SHIFOR is a statistical intensity prediction model. Similar to the CLIPER track guidance model, the SHIFOR model uses several climatological and persistence parameters to predict the future intensity of the storm at 12-hr periods out to 72 hr. The predictor variables include: (1) Julian day; (2) current storm intensity; (3) intensity change during past 12 hr; (4) initial storm latitude and longitude; and (5) zonal (east-west) and meridional (north-south) component of the storm motion. The original SHIFOR equations were developed using data from all historic storms during the period 1900-1972 that were at least 30 nautical miles from land. The new SHIFOR5 equations are based on the historic tropical cyclones from 1967-1999 that, at a minimum, intensified into tropical storms.

Various combinations of the primary parameters were also offered to the multiple regression procedure. In order to maintain continuity, parameters that were selected for inclusion in the prediction equation at one particular forecast time were given preference in the selection for the subsequent forecast time.

Between five and eight predictor terms are included in each equation; most of which are second order products of the seven primary predictors listed above. The most important terms are the current intensity, the 12 hr intensity change, the Julian day and the current storm position. Unlike CLIPER, different predictors may appear in the equations for each lead time. Equations based on storms from throughout the Atlantic, Carribean and Gulf of Mexico are used, since they out-performed those stratified by basin.

Reference: Jarvinen, B. R., and C. J. Neumann, 1979: Statistical forecasts of tropical cyclone intensity. NOAA Tech. Memo. NWS NHC-1 0, 22 pp.
Knaff, John A., Mark DeMaria, Charles R. Sampson and James M. Gross, 2003: Statistical, five-day tropical cyclone intensity forecasts derived from climatology and persistence. Wea. Forecasting, 18, 80-92.

Statistical Hurricane Intensity Prediction Scheme (SHIPS) Model

The SHIPS model is a statistical-dynamic intensity prediction model. This model was developed using standard multiple regression techniques with climatological, persistence, and numerical model forecasts as predictors. Estimates of future storm intensity are made for 12-hr periods out to 120 hr.

The SHIPS equations were initially developed using data from 49 storms during the period 1982-1992 that were at least 30 nautical miles from land. (The collection of synoptic data for LBAR began in 1989, as did the archive of operational intensity forecasts. Data for selected storms during 1982-88 were available and included in the SHIPS developmental data set.) The SHIPS equations are typically updated each year.

Major changes have been the development of DSHIPS (Decay SHIPS) in 2000 to account for the decay of storms over land, the extension of the forecasts to Day 5 in 2001, the replacement of the simple dry-adiabitic prediction model in 2001 with output from the operational global model (the GFS) for the evaluation of the environmental predictors, and the inclusion of satellite-derived parameters (specifically, cloud top temperatures and oceanic heat content) in 2004.

Unlike earlier versions, the most recent versions of SHIPS have significant skill over climatology, at least out to Day 3.

The primary predictors are:

  1. Current storm intensity;
  2. Day of the year;
  3. Persistence (intensity change in previous 12 hrs);
  4. East-west compontent of storm motion;
  5. Divergence of the wind at 200 mb;
  6. Intensification potential (the difference between the current storm intensity and an estimate of the Maximum Possible Storm Intensity determined from the sea surface temperature);
  7. The vertical shear of the horizontal wind in the 850-200 mb layer;
  8. Average 200 mb temperature;
  9. Average 850 mb vorticity;
  10. Average 500-300 mb layer relative humidity;
  11. Cloud top temperature as measured by the GOES satellite infrared imager channel and
  12. Oceanic heat content inferred from altimetry measurements from polar orbiting satellites.
Vertical wind shear is evaluated for the 850-200 mb layer because most satellite cloud track winds are assigned to those levels. The 500-300 mb relative humidity attempts to estimate the impact of any Saharan Dust Layer propagating across the Atlantic basin, which tends to inhibit storm intensitifaction. Terms 1-5 are evaluated at the initial storm location. All other terms are averaged along the forecast storm track.

Research has shown that the sea surface temperature (SST) alone does not provide a good indication of whether a storm will intensify. (See, for example the SST/Intensity relationships of recent Atlantic tropical cyclones.) However, SST does provide an upper limit to storm intensity. In SHIPS, the Maximum Possible Storm Intensity (MPI) is related to the SST by the equation:

MPI = 55.6 kt + 108.5 kt exp[0.1813 * {SST - 30.0oC)]

Since the SHIPS equations were developed using data from storms that were over water, the SHIPS intensity forecasts are not valid for storms near the coast. 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 DSHIPS model had the smallest errors at all forecast periods during the 2000 Atlantic season.

References: DeMaria, M. and J. Kaplan, 1999: An updated statistical hurricane intensity prediction scheme (SHIPS) for the Atlantic and Eastern North Pacific Basins. Wea. Forecasting, 14, 326-337.
DeMaria, M. et al., 2005: Futher improvements to the Statistical Hurricane Intensity Prediction Scheme (SHIPS). Wea. Forecasting, 20, 531-543.

Relative Skill of the Intensity Guidance Models

These graphs shows the skill of the intensity guidance models for the past two seasons Graph of 2004 hurricane intensity skill relative to SHIFOR5 | Graph of 2005 hurricane intensity skill relative to SHIFOR5 Note that the skill for intensity is typically much less than that for storm track, and has not improved substantially in recent years.

Forecast Model Verification

The NWS/NCEP Environmental Modeling Center has produced maps of the spatial variation of the forecast track and intensity errors for Atlantic Storms for the year 1995-1999. Error maps are available for the official forecasts and most forecast models for each of the standard forecast times.

The error plots indicate:

The plots also confirm the left of track bias in the official forecasts, especially at the longer forecast periods.

Storm Surge Guidance Model

SLOSH - The Sea Lake and Overland Surges from Hurricanes Model

When hurricane warnings contain the range of expected peak storm surge heights within the hurricane-warning area, the surge information is often based on the SLOSH model. The dynamical SLOSH model computes the water height over a geographical area or basin. Computations have been run for a number of basins covering most of the Atlantic and Gulf Coasts of the U.S. SLOSH_Basins and the offshore islands SLOSH_Basins

The typical SLOSH grid contains over 500 points located on lines extending radially from a common basin center. The distance between grid points ranges from 0.5 km near the center (where surge water heights are of more interest), to 7.7 km in the deep water at the edge of the grid. Bathymetric and topographic map data are used to determine a water depth or terrain height for each grid point. SLOSH_Basins

The model consists of a set of equations derived from the Newtonian equations of motion and the continuity equation applied to a rotating fluid with a free surface. The equations are integrated from the sea floor to the sea surface. The coastline is represented as a physical boundary within the model domain. Subgrid-scale water features (cuts, chokes, sills and channels), and vertical obstructions (levees, roads, spoil banks, etc.) can be parameterized within the model. Astronomical tides, rainfall, river flow, and wind-driven waves have not been incorporated into the model.

The primary use of the SLOSH model is to define flood-prone areas for evacuation planning. The flood areas are determined by compositing the model surge values from 200-300 hypothetical hurricanes. Separate composite flood maps are produced for each of the five Saffir-Simpson hurricane categories.

Some sample SLOSH maps are available for East-Central Florida and Alabama. A sample animation of a SLOSH model run for Hurricane Hugo is available on the Techniques Development Laboratories Web site.

The SLOSH model can also be run using forecast track and intensity data for an actual storm as it makes landfall. The model is highly responsive to the point of landfall, however. For such operational predictions, the SLOSH model has only limited utility. However, since the North and South Carolina WFOs indicated real-time SLOSH output allowed them to provide more specific storm surge forecasts to their customers during Hurricane Floyd (1999), the Hurricane Floyd Service Assessment includes a recommendation that the NWS Techniques Development Laboratory provide real-time SLOSH output to WFOs when a hurricane is within 12 hours of landfall.

Reference: Jarvinen B. J. and C. J. Neumann, 1985: An evaluation of the SLOSH storm surge model. Bull. Amer. Meteor. Soc., 66, 1408-1411.

Acknowledgment. Special thanks to Jiann-Gwo Jiing, Richard Pasch, James Franklin, Naomi Surgi and the staff of NCEPís Tropical Prediction Center for information and materials used in this presentation.

Additional References

Neumann, C. J., 1979: A guide to Atlantic and Eastern Pacific models for the prediction of tropical cyclone motion. NOAA Tech. Memo. NWS NHC-1 1, 26pp.

Sheets, R. C., 1990: The National Hurricane Center -- past, present and future. Wea. Forecasting, 5, 185-232.