This article is from the Storms FAQ, by Chris Landsea firstname.lastname@example.org with numerous contributions by others.
(Track model information contributed by Sim Aberson)
A variety of hurricane track forecast models are run operationally
for the Atlantic hurricane basin:
(1) The basic model that is used as a "no-skill" forecast to compare
other models against is CLIPER (CLImatology and PERsistence), a multiple
regression model that best utilizes the persistence of the motion and
also incorporates climatological track information (Neumann 1972, Merrill
1980). Surprisingly, CLIPER was difficult to beat with numerical model
forecasts until the 1980s.
(2) A statistical-dynamical model, NHC90 (McAdie 1991), uses geopotential
height predictors from the Aviation model to produce a track forecast four
times per day. The primary synoptic time NHC90 forecasts (00 and 12
UTC) are based upon 12 h old Aviation runs. A special version of NHC90,
NHC90-LATE, is run at primary synoptic times with the current Aviation
run, and is available a number of hours after NHC90. Both versions of
NHC90 have been run operationally since 1990.
(3) The Beta and Advection Model, BAM, follows a trajectory in the
pressure-weighted vertically-averaged horizontal wind from the Aviation
model beginning at the current storm location, with a correction that
accounts for the beta effect (Marks 1992). Three versions of this model,
one with a shallow-layer (BAMS), one with a medium-layer (BAMM), and one
with a deep-layer (BAMD), are run. BAMS runs using the 850-700 mb layer,
BAMM with the 850-400 mb layer, and BAMD with the 850-200 mb layer. The
deep-layer version was run operationally for primary synoptic times in
1989; all three versions have been run four times per day since 1990.
(4) A nested barotropic hurricane track forecast model (VICBAR) has been
run four times daily since 1989. The 0000 and 1200 UTC runs are based
upon current NCEP analyses, the others upon six hour old data (Aberson
and DeMaria 1994). Another barotropic model, LBAR, for Limited-Area
Barotropic Model, is also being run operationally every 6 h based upon
six hour old data, so is available for earlier use by the NHC forecasters.
(5) A triply-nested movable mesh primitive equation model developed at
the Geophysical Fluid Dynamics Laboratory (Bender et al 1993), known as the
GFDL model, has provided forecasts since the 1992 hurricane season.
(6) The NCEP Aviation and MRF models (Lord 1993) have been used for
track forecasting since the 1992 hurricane season. These are global
(7) The United Kingdom Meterological Office's global model (UKMET) is
utilized for forecasting the track of tropical cyclones around the
world (Radford 1994). The National Hurricane Center starting receiving
these operationally during 1996.
(8) The United States Navy Operational Global Atmospheric Prediction
Systems (NOGAPS) is also a global numerical model that shows skill in
forecasting tropical cyclone track (Fiorino et al. 1993). This model was
also first received operationally at the National Hurricane Center
Despite the variety of hurricane track forecast models, there are
only a few models that forecast intensity change for the Atlantic
(1) Similar to the CLIPER track model, SHIFOR (Statistical Hurricane
Intensity Forecast model) is used as a "no-skill" intensity change
forecast. It is a multiple regression statistical model that best
utilizes the persistence of the intensity trends and also incorporates
climatological intensity change information (Jarvinen and Neumann 1979).
Surprisingly, no other intensity models provide better forecasts on average
(2) A statistical-synoptic model, SHIPS (Statistical Hurricane
Intensity Prediction Scheme), has been available the National Hurricane
Center since the mid-1990s (DeMaria and Kaplan 1994). It takes current
information on the synoptic scale on the sea surface temperatures,
vertical shear, etc. with an optimal combination of the trends in
the cyclone intensity. For the first time in 1996, SHIPS outperformed
SHIFOR (by having lower absolute wind speed errors) from the 24 hour to
72 hour forecasts, though the differences were small.
(3) The GFDL model, described above in the track forecasting models,
also issues forecasts of intensity change for the National Hurricane
Center. However, to date, these have yet to show any skill (i.e. GFDL
errors are larger than those from SHIFOR).