Improvement in hurricane track forecasts has been well documented over the past three decades. Figure B-1 shows the reduction of the 72 hour official track error of the TPC/NHC over this period from 400 nm to less than 200 nm. The improved skill closely follows the continuous advancement of operational numerical models and their enhanced forecast capabilities.
The documentation of model track skill originated with CLIPER in the 1970s, a statistical model based on climatology and persistence, which became the benchmark for track skill for all future model track forecasts (Neuman 1972). An improvement of track skill continued in the 1980s with the use of a hybrid model that combined statistical techniques with background fields from NOAA’s global model (e.g., the NHC-83, NHC-90 [Neuman and McAdie 1991]), the development of a barotropic model (VICBAR) for operations, a one level advection model (BAMS), and a quasi-lagrangian dynamical model (QLM) that ran at NCEP in the late 1980s and early 1990s. These early models all helped define the downward linear trend into the early-mid 1990s to reduce the forecast errors at all forecast times. The description and performance characteristics of these early track models are described in DeMaria (1997); McAdie and Lawrence (2000).
As shown in Figure B-1, a pronounced acceleration in track forecast skill occurred in the mid-1990s with increased use of global models (e.g., the Global Forecast System (GFS; formerly the MRF/AVN) run at the National Centers for Environmental Prediction (NCEP); the Navy Operational Global Atmospheric Prediction System (NOGAPS) run at Fleet Numerical Meteorology and Oceanography Center (FNMOC), and the United Kingdom Meteorological Office global model (UKMO)). The increase in forecast skill is tied to the availability of global observations (e.g., satellites), an advancement of global modeling numerical techniques that could maximize the usefulness of these data, and increased sophistication in representing model physics to provide routine high quality global analyses.
Since the early 1990s, the horizontal resolution of the NOGAPS global spectral model has tripled while its vertical resolution has nearly doubled. For example, from 1989 to1994, the NOGAPS model resolution was T79L18 (~165 km horizontal resolution, 18 vertical levels), using the Arakawa-Schubert convective parameterization scheme (Arakawa-Schubert 1974) ; from 1994 to 2000, the NOGAPS model resolution was either T159L18 or T159L24 (by 1998), using the Arakawa-Schubert convective parameterization scheme. From 2000 to 2002, the NOGAPS model resolution was T159L24, using the Emanuel convective parameterization scheme (Emanuel 1991, Emanuel and Zivkovic-Rothman 1999). Finally, from 2002 to the present, the resolution of the NOGAPS model has been T239L30, using the Emanuel convective parameterization scheme. Similar upgrades to other global models also occurred.
There were other important upgrades to the global models, and a few examples follow. To represent the hurricane scale, a bogus vortex was developed by Lord (1993) and incorporated in the NCEP global model. The direct use of satellite radiances replaced the use of retrievals in the GFS data assimilation system at NCEP in 1995 (Derber and Wu 1998). The assimilation of high density multispectral GOES-8 winds (Velden et al. 1997) into NOGAPS was initiated at FNMOC in 1996 (Goerss et al. 1998).
Although global model forecasts were advancing to provide better track forecasts, after the devastation of Hurricane Andrew in 1992 in South Florida and a clear inability of the global models to forecast the high regime catastrophic landfall winds, a resurgence in developing high resolution dynamical hurricane models became a focus not only for improving track forecasts but also offered promise for providing the higher resolution forecast models needed to address hurricane intensity forecasts. Research for more than two decades since the late 1970s, led by the pioneering effort of Yoshio Kurihara at NOAA’s Geophysical a Dynamics Laboratory (GFDL), led to the development of a movable nested grid hurricane model (Kurihara and Bender 1980). The TPC/NHC requested that the high resolution, nested, movable GFDL model be evaluated to assess its performance in a semi-operational mode at the TPC/NHC in 1993. The seminal GFDL forecasts of Hurricane Emily and the recurvature of this storm off the outer banks of North Carolina in 1993 led to a pioneering collaboration between NOAA research and NOAA operations.
For the 1994 hurricane season, the GFDL model was monitored for operational performance. Due to its promising performance in providing higher track skill for the Atlantic and East Pacific basins than all other operational models, the GFDL model was transitioned into NCEP operations for the 1995 hurricane season. A version of the GFDL that is run at FNMOC, the GFDN, became operational in May 1996 (Rennick 1999). Although many transitional modeling and code obstacles existed at the time to transition a research model into operations, the joint efforts of the NCEP Environmental Modeling Center (EMC) and GFDL became a defining collaboration that has endured to the present.
With continuous yearly upgrades to the GFDL model, which were aligned with the upgrades to the NCEP global model, the GFDL model became the top track performance model and the mainstay for hurricane forecast guidance at TPC/NHC (Kurihara et al, 1998). In carrying out the joint vision for operational performance standards, the close collaboration between EMC and GFDL is considered one the most successful collaborations within NOAA and perhaps within the U.S. modeling community between research and operations.
As skillful track forecasts became more consistently deliverable to TPC/NHC, CPHC, and JTWC forecasters during the 1990s, particularly with the operational implementation of the GFDL and GFDN models, more attention became focused on improving intensity forecasts from dynamical models within the hurricane community. An aspect of addressing the intensity issue required the coupling of the atmosphere with the ocean. To meet this requirement, the University of Rhode Island (URI) offered an ocean model that could be readily coupled to the GFDL model (Bender et al. 2001). The coupled GFDL model was run on 163 forecasts during the 1995–98 seasons (Bender and Ginnis 2000). The coupling of the atmosphere with the ocean improved intensity forecasts, with the mean absolute error in the forecast of central pressure reduced by about 26 percent compared to the operational (non-coupled) GFDL model.
The coupled GFDL model became operational in 2001. It provided an upgrade to the GFDL system for hurricane scenarios where changes in sea surface temperatures (SSTs) were important to intensity changes. This effort formed unique three-way collaborations between operational hurricane modeling at EMC, NOAA research, and academia. With support through the USWRP, this close working collaboration has continued to the present.
To date, the coupled GFDL model has become the benchmark for performance against which the future operational NCEP hurricane model—the Hurricane Weather and Research Forecast system (HWRF)—will be measured for track forecast skill and forecast consistency.