• Aberson, S. D., and M. DeMaria, 1994: Verification of a nested barotropic hurricane track forecast model (VICBAR). Mon. Wea. Rev., 122 , 28042815.

    • Search Google Scholar
    • Export Citation
  • Bender, M. A., and I. Ginis, 2000: Real-case simulations of hurricane–ocean interaction using a high-resolution coupled model: Effects on hurricane intensity. Mon. Wea. Rev., 128 , 917945.

    • Search Google Scholar
    • Export Citation
  • Bender, M. A., R. J. Ross, R. E. Tuleya, and Y. Kurihara, 1993: Improvements in tropical cyclone track and intensity forecasts using the GFDL initialization system. Mon. Wea. Rev., 121 , 20462061.

    • Search Google Scholar
    • Export Citation
  • Bengtsson, L., M. Botzet, and M. Esch, 1995: Hurricane-type vortices in a general circulation model. Tellus, 47 , 175196.

  • Businger, J. A., J. C. Wyngard, Y. Izumi, and E. F. Bradley, 1971: Flux profile relationship in the atmospheric surface layer. J. Atmos. Sci., 28 , 181189.

    • Search Google Scholar
    • Export Citation
  • Chen, Q-S., L-E. Bai, and D. H. Bromwich, 1997: A harmonic-Fourier spectral limited-area model with an external wind lateral boundary condition. Mon. Wea. Rev., 125 , 143167.

    • Search Google Scholar
    • Export Citation
  • Cocke, S., 1998: Case study of Erin using the FSU Nested Regional Spectral Model. Mon. Wea. Rev., 126 , 13371346.

  • DeMaria, M., and J. Kaplan, 1994: Sea surface temperature and the maximum intensity of Atlantic tropical cyclones. J. Climate, 7 , 13241334.

    • Search Google Scholar
    • Export Citation
  • DeMaria, M., and J. Kaplan, 1997: An operational evaluation of a statistical hurricane intensity prediction scheme (SHIPS). Preprints, 22d Conf. on Hurricanes and Tropical Meteorology, Ft. Collins, CO, Amer. Meteor. Soc., 280–281.

    • Search Google Scholar
    • Export Citation
  • Goerss, J. S., 2000: Tropical cyclone track forecasts using an ensemble of dynamical models. Mon. Wea. Rev., 128 , 11871193.

  • Goerss, J. S., and R. Jeffries, 1994: Assimilation of synthetic tropical cyclone observations into the Navy Operational Global Atmospheric Prediction System. Wea. Forecasting, 9 , 557576.

    • Search Google Scholar
    • Export Citation
  • Harshvardan, and T. G. Corsetti, 1984: Longwave parameterization for the UCLA/GLAS GCM. NASA Tech. Memo. 86072, 52 pp.

  • Heming, J. T., 1997: UK Meteorological Office forecast performance during the unusual Atlantic hurricane season of 1995. Preprints, 22d Conf. on Hurricanes and Tropical Meteorology, Ft. Collins, CO, Amer. Meteor. Soc., 511–512.

    • Search Google Scholar
    • Export Citation
  • Hogan, T., and T. Rosmond, 1991: The description of the Navy Operational Global Atmospheric Prediction System's spectral forecast model. Mon. Wea. Rev., 119 , 17861815.

    • Search Google Scholar
    • Export Citation
  • Holland, G. J., 1983: Tropical cyclone motion: Environmental interaction plus a beta effect. J. Atmos. Sci., 40 , 328342.

  • Horsfall, F. M., M. DeMaria, and J. M. Gross, 1997: Optimal use of large-scale boundary and initial fields for limited-area hurricane forecast models. Preprints, 22d Conf. on Hurricanes and Tropical Meteorology, Ft. Collins, CO, Amer. Meteor. Soc., 571–572.

    • Search Google Scholar
    • Export Citation
  • Hoyer, J. M., 1987: The ECMWF spectral limited-area model. Proc. ECMWF Workshop on Techniques for Horizontal Discretization in Numerical Weather Prediction Models, Shinfield Park, Reading, United Kingdom, ECMWF, 343–359.

    • Search Google Scholar
    • Export Citation
  • Jarvinen, B. R., and C. J. Neumann, 1979: Statistical forecasts of tropical cyclone intensity. NOAA Tech. Memo. NWS NHC-10, 22 pp.

  • Juang, H. H-M., and M. Kanamitsu, 1994: The NMC Nested Regional Spectral Model. Mon. Wea. Rev., 122 , 326.

  • Kanamitsu, M., 1975: On numerical prediction over a global tropical belt. Rep. 75-1, Department of Meteorology, Florida State University, 282 pp.

    • Search Google Scholar
    • Export Citation
  • Kanamitsu, M., K. Tada, K. Kudo, N. Sato, and S. Isa, 1983: Description of the JMA operational spectral model. J. Meteor. Soc. Japan, 61 , 812828.

    • Search Google Scholar
    • Export Citation
  • Kitade, T., 1983: Nonlinear normal mode initialization with physics. Mon. Wea. Rev., 111 , 21942213.

  • Krishnamurti, T. N., and L. Bounoua, 1996: An Introduction to Numerical Weather Prediction Techniques,. CRC Press, 293 pp.

  • Krishnamurti, T. N., S. Low-Nam, and R. Pasch, 1983: Cumulus parameterization and rainfall rates II. Mon. Wea. Rev., 111 , 815828.

  • Krishnamurti, T. N., D. Oosterhof, and N. Dignon, 1989: Hurricane prediction with a high-resolution global model. Mon. Wea. Rev., 117 , 631669.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., J. Xue, H. S. Bedi, K. Ingles, and D. Oosterhof, 1991a: Physical initialization for numerical weather prediction over the tropics. Tellus, 43AB , 5381.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., K. S. Yap, and D. K. Oosterhof, 1991b: Sensitivity of tropical storm forecast to radiative destabilization. Mon. Wea. Rev., 119 , 21762205.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., H. S. Bedi, G. D. Rohaly, D. K. Oosterhof, R. C. Torres, E. Williford, and N. Surgi, 1996: Physical initialization. J. Atmos. Oceans, 35 , 369398.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., W. Han, and D. K. Oosterhof, 1998: Sensitivity of hurricane intensity forecasts to physical initialization. Meteor. Atmos. Phys., 65 , 171181.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., C. M. Kishtawal, T. LaRow, D. Bachiochi, Z. Zhang, C. E. Williford, S. Gadgil, and S. Surendran, 1999: Improved skills for weather and seasonal climate forecasts from multimodel superensemble. Science, 285 , 15481550.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., C. M. Kishtawal, D. W. Shin, and C. E. Williford, 2000a: Multimodel superensemble forecasts for weather and seasonal climate. J. Climate, 13 , 41964216.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., C. M. Kishtawal, T. LaRow, D. Bachiochi, Z. Zhang, C. E. Williford, S. Gadgil, and S. Surendran, 2000b: Improving tropical precipitation forecasts from a multianalysis superensemble. J. Climate, 13 , 42174227.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., and Coauthors. 2001: Real-time multianalysis–multimodel superensemble forecasts of precipitation using TRMM and SSM/I products. Mon. Wea. Rev., 129 , 28612883.

    • Search Google Scholar
    • Export Citation
  • Kuo, H. L., 1965: On formation and intensification of tropical cyclones through latent heat release by cumulus convection. J. Atmos. Sci., 22 , 4063.

    • Search Google Scholar
    • Export Citation
  • Kuo, H. L., 1974: Further studies of the parameterization of the influence of cumulus convection on large-scale flow. J. Atmos. Sci., 31 , 12321240.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., and M. A. Bender, 1982: Structure and analysis of the eye of a numerically simulated tropical cyclone. J. Meteor. Soc. Japan, 60 , 381395.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., G. J. Tripoli, and M. A. Bender, 1979: Design of a movable nested-mesh primitive equation model. Mon. Wea. Rev., 107 , 239249.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., M. A. Bender, and R. T. Ross, 1993: An initialization scheme of hurricane models by vortex specification. Mon. Wea. Rev., 121 , 20302045.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., M. A. Bender, R. E. Tuleya, and R. J. Ross, 1995: Improvements in the GFDL hurricane prediction system. Mon. Wea. Rev., 123 , 27912801.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., R. E. Tuleya, and M. A. Bender, 1998: The GFDL hurricane prediction system and its performance in the 1995 hurricane season. Mon. Wea. Rev., 126 , 13061322.

    • Search Google Scholar
    • Export Citation
  • Lacis, A. A., and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth's atmosphere. J. Atmos. Sci., 31 , 118133.

    • Search Google Scholar
    • Export Citation
  • Lawrence, M. B., L. A. Avila, J. L. Beven, J. L. Franklin, J. L. Guiney, and R. J. Pasch, 2001: Atlantic hurricane season of 1999. Mon. Wea. Rev., 129 , 30573084.

    • Search Google Scholar
    • Export Citation
  • Lord, S. J., 1991: A bogussing system for vortex circulations in the National Meteorological Center global forecast model. Preprints, 19th Conf. on Hurricanes and Tropical Meteorology, Miami, FL, Amer. Meteor. Soc., 328–330.

    • Search Google Scholar
    • Export Citation
  • Lord, S. J., H. E. Willoughby, and J. M. Piotrowicz, 1984: Role of a parameterized ice-phase microphysics in an axisymmetric, nonhydrostatic tropical cyclone model. J. Atmos. Sci., 41 , 28362848.

    • Search Google Scholar
    • Export Citation
  • Louis, J. F., 1979: A parametric model of vertical eddy fluxes in the atmosphere. Bound.-Layer Meteor., 17 , 187202.

  • Mackey, B. P., and T. N. Krishnamurti, 2001: Ensemble forecast of a typhoon flood event. Wea. Forecasting, 16 , 399415.

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

  • Neumann, C. J., 1972: An alternate to the HURRAN tropical cyclone forecast system. NOAA Tech. Memo. NWS SR-62, 22 pp.

  • Neumann, C. J., and C. J. McAdie, 1991: A revised National Hurricane Center NHC83 model (NHC90). NOAA Tech. Memo. NWS NHC-44, 35 pp.

  • Pasch, R. J., L. A. Avila, and J. L. Guiney, 2001: Atlantic hurricane season of 1998. Mon. Wea. Rev., 129 , 30853099.

  • Reynolds, R. W., and T. M. Smith, 1993: An improved real-time global sea surface temperature analysis. J. Climate, 6 , 114119.

  • Rotunno, R., and K. A. Emanuel, 1987: An air–sea interaction theory for tropical cyclones. Part II: Evolutionary study using a nonhydrostatic axisymmetric numerical model. J. Atmos. Sci., 44 , 542561.

    • Search Google Scholar
    • Export Citation
  • Surgi, N., H-L. Pan, and S. J. Lord, 1998: Improvement of the NCEP global model over the Tropics: An evaluation of model performance during the 1995 hurricane season. Mon. Wea. Rev., 126 , 12871305.

    • Search Google Scholar
    • Export Citation
  • Thu, T. V., and T. N. Krishnamurti, 1992: Vortex initialization for typhoon track prediction. J. Meteor. Atmos. Phys., 47 , 117126.

  • Tiedke, M., 1984: The sensitivity of the time-mean large-scale flow to cumulus convection in the ECMWF model. Proc. ECMWF Workshop on Convection in Large-Scale Numerical Models, Reading, Berkshire, United Kingdom, ECMWF, 297–316.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., S. Tibaldi, and A. J. Simmons, 1983: Reduction of systematic forecast errors in the ECMWF model through the introduction of envelope orography. Quart. J. Roy. Meteor. Soc., 109 , 683718.

    • Search Google Scholar
    • Export Citation
  • Wang, Y., 2001: An explicit simulation of tropical cyclones with a triply nested movable mesh primitive equation model: TCM3. Part I: Model description and control experiment. Mon. Wea. Rev., 129 , 13701394.

    • Search Google Scholar
    • Export Citation
  • Williford, C. E., R. J. Correa-Torres, and T. N. Krishnamurti, 1998: Tropical cyclone forecasts made with the FSU Global Spectral Model. Mon. Wea. Rev., 126 , 13321336.

    • Search Google Scholar
    • Export Citation
  • Zhang, D-L., and E. Altshuler, 1999: The effects of dissipative heating on hurricane intensity. Mon. Wea. Rev., 127 , 30323038.

  • Zhang, Z., and T. N. Krishnamurti, 1999: A perturbation method for hurricane ensemble predictions. Mon. Wea. Rev., 127 , 447469.

  • View in gallery

    FSU superensemble production time line

  • View in gallery

    (a) The 1998 Atlantic tropical system cross-validation-based track errors, hours 12–72, including FSU superensemble and ensemble mean forecasts; (b) the 1998 Atlantic tropical system cross-validation-based intensity errors, hours 12–72, including FSU superensemble and ensemble mean forecasts

  • View in gallery

    (a) Hurricane Georges 72-h track forecast, starting 1200 UTC 22 Sep 1998; (b) Hurricane Georges 72-h track forecast, starting 1200 UTC 23 Sep 1998

  • View in gallery

    (a) Hurricane Bonnie 72-h intensity forecast, starting 1200 UTC 20 Aug 1998; (b) Hurricane Bonnie 72-h intensity forecast, starting 1200 UTC 25 Aug 1998

  • View in gallery

    (a) Hurricane Georges 72-h intensity forecast, starting 1200 UTC 21 Sep 1998; (b) Hurricane Georges 72-h intensity forecast, starting 1200 UTC 22 Sep 1998

  • View in gallery

    (a) The 1999 Atlantic tropical system track errors, hours 12–72, including FSU superensemble and ensemble mean forecasts; (b) the 1999 Atlantic tropical system intensity errors, hours 12–72, including FSU superensemble and ensemble mean forecasts

  • View in gallery

    (a) The 1999 Atlantic individual storm track errors, hour 48, including FSU superensemble and ensemble mean forecasts; (b) the 1999 Atlantic individual storm track errors, hour 72, including FSU superensemble and ensemble mean forecasts

  • View in gallery

    (a) Hurricane Floyd 72-h track forecast, starting 1200 UTC 13 Sep 1999; (b) Hurricane Floyd 72-h track forecast, starting 1200 UTC 14 Sep 1999

  • View in gallery

    Hurricane Irene 72-h track forecast, starting 1200 UTC 15 Oct 1999

  • View in gallery

    (a) Hurricane Lenny 72-h track forecast, starting 1200 UTC 15 Nov 1999; (b) Hurricane Lenny 72-h track forecast, starting 1200 UTC 16 Nov 1999

  • View in gallery

    Hurricane Floyd 72-h intensity forecast, starting 1200 UTC 13 Sep 1999

  • View in gallery

    Hurricane Irene 48-h intensity forecast, starting 1200 UTC 15 Oct 1999

  • View in gallery

    (a) Hurricane Lenny 72-h intensity forecast, starting 1200 UTC 15 Nov 1999; (b) Hurricane Lenny 36-h intensity forecast, Starting 1200 UTC 17 Nov 1999

  • View in gallery

    The 2000 Atlantic tropical system track errors, hours 12–72, including FSU superensemble and ensemble mean forecasts

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 134 62 6
PDF Downloads 81 36 3

Real-Time Multimodel Superensemble Forecasts of Atlantic Tropical Systems of 1999

View More View Less
  • 1 Department of Meteorology, The Florida State University, Tallahassee, Florida
  • | 2 T. J. Watson Laboratory, IBM, New York, New York
  • | 3 Department of Meteorology, The Florida State University, Tallahassee, Florida
Full access

Abstract

In this paper, Atlantic hurricane forecasts for the year 1999 are addressed. The methodology for these forecasts is called the multimodel superensemble. This statistical method makes use of the real-time forecasts provided by a number of operational and research models to construct the superensemble forecasts. This method divides the forecast time line into two phases: a training phase and a forecast combining phase. The training phase includes an inventory of past applicable hurricane forecasts, each by the multimodels. The model biases of position and intensity errors of past forecasts are summarized via a simple linear multiple regression of these forecasts against the best-observed estimates of position and intensity. These statistics are next passed on to future forecasts of the multimodels in order to forecast the hurricanes of 1999. This method was first tested for the hurricanes of 1998 with considerable success, with some of those results summarized here. Those statistics were refined for the 1999 Atlantic hurricane season. Overall, the main result of the seasonal summary is that the position and intensity errors for the multimodel superensemble are generally less than those of all of the participating models during 1–5-day real-time forecasts. Some of the major storms of the 1999 season, such as Dennis, Floyd, Irene, and Lenny, were extremely well handled by this superensemble approach. The message of this study is that the proposed approach may be a viable way to construct improved real-time forecasts of hurricane positions and intensity.

Corresponding author address: C. Eric Williford, Weather Predict, Inc., 3200 Atlantic Ave., Suite 114, Raleigh, NC 27604. Email: cew@weatherpredict.com

Abstract

In this paper, Atlantic hurricane forecasts for the year 1999 are addressed. The methodology for these forecasts is called the multimodel superensemble. This statistical method makes use of the real-time forecasts provided by a number of operational and research models to construct the superensemble forecasts. This method divides the forecast time line into two phases: a training phase and a forecast combining phase. The training phase includes an inventory of past applicable hurricane forecasts, each by the multimodels. The model biases of position and intensity errors of past forecasts are summarized via a simple linear multiple regression of these forecasts against the best-observed estimates of position and intensity. These statistics are next passed on to future forecasts of the multimodels in order to forecast the hurricanes of 1999. This method was first tested for the hurricanes of 1998 with considerable success, with some of those results summarized here. Those statistics were refined for the 1999 Atlantic hurricane season. Overall, the main result of the seasonal summary is that the position and intensity errors for the multimodel superensemble are generally less than those of all of the participating models during 1–5-day real-time forecasts. Some of the major storms of the 1999 season, such as Dennis, Floyd, Irene, and Lenny, were extremely well handled by this superensemble approach. The message of this study is that the proposed approach may be a viable way to construct improved real-time forecasts of hurricane positions and intensity.

Corresponding author address: C. Eric Williford, Weather Predict, Inc., 3200 Atlantic Ave., Suite 114, Raleigh, NC 27604. Email: cew@weatherpredict.com

1. Introduction

Current research thrusts on hurricane forecast modeling have used both high-resolution global and regional numerical models. Much progress has emerged in such modeling in recent years as noted in numerous recent publications (Bender et al. 1993; Kurihara et al. 1979, 1982, 1993, 1995, 1998; Lord et al. 1984; Rotunno and Emanuel 1987; Zhang and Altshuler 1999). Previous studies using The Florida State University Global Spectral Model (FSU GSM) have shown the potential for improved prediction of tropical systems as seen in Krishnamurti et al. (1989, 1991a), Williford et al. (1998), and Cocke (1998). Bengtsson et al. (1995) have also addressed, in detail, the hurricane forecast issues from global models.

Our goal is to produce high-resolution, improved tropical forecasts out to 5–6 days into the future and develop innovative techniques for using a suite of multimodel forecast products to better understand and predict the ongoing tropical development. This can be of great use to state and government agencies, especially when the products are tailored for a specific environment (e.g., landfalling wind swaths, strike probability ellipses for specific coastlines).

The FSU real-time forecasting efforts are based on a number of numerical weather prediction models and also from modeling techniques developed at FSU. The FSU GSM is a high-resolution numerical weather prediction model that includes a physical initialization component (Krishnamurti et al. 1991a) and tropical ensemble forecasting (Zhang and Krishnamurti 1999; Mackey and Krishnamurti 2001). The GSM's performance has been tested on a large sample of past storms over the Atlantic and Pacific Oceans. Using a physical initialization procedure within a high-resolution FSU GSM, Krishnamurti et al. (1998) and Williford et al. (1998) have shown great success in predicting hurricane tracks. In trial runs Cocke (1998) has noted that a high-resolution regional spectral model outperforms the FSU GSM, which is run at a lower resolution, and the regional model has demonstrated some improved skills in predicting tropical cyclone intensities.

Operationally for the public, a charter of the National Hurricane Center (NHC) in Miami is to provide forecasts in real time for the Atlantic, Caribbean, and Gulf of Mexico hurricanes and other tropical systems. The experienced forecasters at NHC receive real-time forecasts from a number of leading operational modeling groups from the United States and other countries and produce what is called the NHC “official” or forecaster–consensus forecast. The suite of models utilized by the National Hurricane Center includes in-house dynamical models such as the Limited Area Sine Transform Barotropic Model (LBAR; Chen et al. 1997) and Vic Ooyama's Spectral Barotropic Model (VICBAR; Aberson and DeMaria 1994), trajectory models (Beta Advection Model: shallow, medium, and deep; Marks 1992; Holland 1983), and statistical models such as NHC90 (Neumann and McAdie 1991) and the Climatology and Persistence Statistical Model (CLIPER; Neumann 1972). A list of acronyms is given in Table 1.

In the present study, along with these NHC components, we have also included a suite of global models from operational centers such as the U.S. Navy Operational Global Atmospheric Prediction System (NOGAPS; Hogan and Rosmond 1991; Goerss and Jeffries 1994), United Kingdom Met Office Global Model (Heming 1997), the National Centers for Environmental Prediction Aviation global model (NCEP AVN; Surgi et al. 1998; Lord 1991), and the Geophysical Fluid Dynamics Laboratory Multiple Mesh Mesoscale (GFDL MMM) model (Kurihara et al. 1995). Furthermore, specialized intensity models that have been developed and run operationally, namely, the Statistical Hurricane Intensity Forecast (SHIFOR; Jarvinen and Neumann 1979) and the Statistical Hurricane Intensity Prediction Scheme (SHIPS; DeMaria and Kaplan 1997, 1994; Reynolds and Smith 1993), as well as the mesoscale dynamical GFDL forecast model (Horsfall et al. 1997), are also included in this study.

The present study is based on most of these multimodel forecasts where we invoke what is called a “multimodel superensemble,” following Krishnamurti et al. (1999, 2000a,b, 2001), to construct an objective consensus forecast. In section 2 we describe the FSU modeling component, followed by the methodology of present work in section 3. The datasets used are described in section 4.

2. The FSU modeling component

In addition to the above models, a suite of seven FSU models are also being currently run (for 1–6-day forecasts) in real time to provide additional forecast datasets for the construction of a multimodel superensemble. (The methodology is described in sections 5 and 6 of this paper.) These models include the following:

  1. FSU GSM runs at a resolution of triangular truncation 126 waves, without rain-rate initialization. This has an approximate horizontal transform grid separation of 95 km and is known as the control experiment. An outline of the components of the current version of the FSU GSM is provided in the appendix. This is a dynamical–physical model that is very comprehensive in its basic parameterization of physical processes, land surface, and air–sea transfer processes. It utilizes a spectral transform method and the semi-implicit time differencing scheme.

  2. FSU GSM runs at a resolution of triangular truncation 126 waves including the rain-rate initialization (called physical initialization; Krishnamurti et al. 1991a; Krishnamurti et al. 1996).

  3. An ensemble of five forecasts is carried out with a regional spectral model developed by Cocke (1998). This model utilizes Fourier functions as its basis functions. This model works with departure variables with respect to the global model solutions, thus utilizing doubly periodic boundary conditions. This model has the same vertical discretization as the global model; they both make use of the so-called Charney–Phillips vertical discretization. Currently these two models share the same physical algorithms. An ensemble of the above five-member regional spectral model forecasts is prepared using the EOF-based initial perturbations following Zhang and Krishnamurti (1999). The five members utilize five initial locations for the storms' positions 50 km from the vortex central position. These are located at the official “best-track” initial position, ±50 km to the left or right of that position, and ±50 km to the north or south of the official position. At these locations, a synthetic (Rankine-like) asymmetric three-dimensional vortex is initially placed following Thu and Krishnamurti (1992). The synthetic vortex includes a number of parameters, such as the radial size of the vortex, maximum sustained surface winds, lowest minimum surface pressure, and depth of the initial vortex. The vertical structure of the initial Rankine-like vortex is taken from normalized vertical structure (via lookup table) based on past storms. These are carefully chosen for respective storms from the daily parameters provided by the NHC. The regional spectral model is integrated out to six days using these respective initial positions. Thus, in all, there were seven forecasts that were prepared at FSU on a daily basis, which includes the GSM runs made with the control and the physically initialized experiments.

3. Methodology

a. Computational techniques

The FSU GSM was originally coded for vector computers. A version of the model containing roughly 150 000 lines of code was ported onto superscalar parallel platforms and was set to execute in 64-bit precision. Most of the testing and porting was done using the T106 spectral resolution. The memory requirements for the T126 with 14 vertical layers were close to 20 Megaword (MW). There are various processes involved for the execution of a single time step in the GSM, including the calculation of the physical parameters and the continuous transformation of the various meteorological variables between the gridpoint (G) and the spectral (S) domains. These transformations are achieved via the use of the direct and inverse fast Fourier transforms (FFTs) and Legendre transforms (LTs). In porting the model to a distributed memory parallel system, the serial code structure was maintained, with the exception of the parts requiring use of the Fourier and Legendre transforms. These parts were modified to accommodate necessary communication among the processors participating in the parallel execution to assure that the results obtained from the serial and parallel versions were identical. Furthermore, the code was optimized for fast execution on as many as 12 processors, via proper translation and verification of all the model routines.

b. Superensemble methodology for hurricane forecasts

The forecast production time line is partitioned into two parts: a training phase and a forecast assimilation phase. The training phase includes roughly 75 or more individual sets of past tropical storm or stronger forecasts from these multimodels; however, this set is limited by the requirement that no major changes to the individual models have occurred between the training and the forecast phase. Given the storm position and intensity [maximum sustained surface (10 m) wind, typically 1-min sustained average] from these past hurricane forecasts plus the “observed” best-track and intensity datasets, the training phase includes the evaluation of the linear regression–based statistical coefficients for days 1–6 of forecasts, where each time of the forecast (e.g., hour 12, 24, 36, etc.) is handled separately. Also each variable (latitude, longitude, and intensity) per forecast time is handled separately. (Note that this process uses scalar variable values, not gridded NWP fields; the NWP fields for the global and regional numerical models are used to determine the position and intensity values for a given tropical system.)

During the forecast period, the superensemble forecasts are constructed using the aforementioned statistics and the current multimodel forecasts. The details of the multiple regression methodology are presented in Krishnamurti et al. (1999, 2000a). These forecasts were carried out for each of the tropical system forecasts for the 1999 Atlantic hurricane season. This procedure first required a detailed tabulation of forecast data from past storms, that is, time, latitude–longitude (position), and intensity datasets for past storms from all models. In this paper we shall review the results of the initial test forecasts from 1998 and the real-time Atlantic tropical system forecasts for 1999.

c. Superensemble time line for hurricane forecasts

While hurricane research can produce storm forecasts after the fact with the best available storm-track and intensity fields, this effort has been to produce superensemble forecasts in real time, and this has presented several challenges. The issues of using the available forecast products for producing improved forecasts, with up-to-date storm information, were the same for research as it is in operational usage.

Our initial studies using 1998 forecast data utilized the 1200 UTC forecasts from National Oceanic and Atmospheric Administration/Tropical Prediction Center (NOAA/TPC), NOGAPS, Met Office, as well as the FSU model components. Some of the initial datasets for the FSU model forecasts, integrated out to 6 days into the future, were only available as late as 23 h after the initial 1200 UTC time period. Therefore, their usage in a real-time operational environment was limited since newer model forecasts at hour 24 were already becoming available. In order to address this situation and produce useable forecasts in real time, a method was devised to extract a 5-day real-time forecast from the latest available 6-day forecast datasets generated by the FSU superensemble system, as seen in Fig. 1.

The superensemble forecast technique can be implemented to handle base variables or incremental changes from the previous forecasts or observed values. When incremental forecasts are produced, the observed (latest) storm information can be utilized at any point along the forecast time line during the initial 12–24 h. It was found that in order to have the best prediction capability, the incremental technique was better suited since it allowed the latest 1200 UTC positions and intensities to be incorporated as the initial forecast information for the ensuing 5-day forecasts. To have a fair comparison of all available forecasts at 1200 UTC, some of the forecasts had to be interpolated to the 1200 UTC forecast values; these were in fact used in the assessment of the 5-day real-time superensemble forecasts. The list of products available at 1200 UTC is presented in Fig. 1 along with the superensemble time line. Future tests of the incremental technique, including the year 2000 experimental Atlantic forecasts, utilize as much information from earlier forecasts (including more recent 0000 and even 0600 UTC forecasts) as possible.

4. Datasets

A variety of datasets is being used in the FSU modeling stream and for the multimodel component of this study. These include the following.

a. Base analysis

We rely on the operational daily high-resolution (0.5° latitude–longitude and 15 vertical levels) 1200 UTC European Centre for Medium-Range Weather Forecasts (ECMWF) analysis (Hoyer 1987). This includes datasets for all the basic variables (u, υ, T, q, z). Furthermore, these datasets include the current 7-day averaged, 1° × 1° field of SST obtained from NCEP (Reynolds and Smith 1993).

b. Rainfall estimates

The FSU models invoke rain-rate initialization following Krishnamurti et al. (1991a). This requires rainfall estimates between hours −24 and 0. Currently, there are five satellites that provide microwave radiance datasets for deriving rainfall estimates. These include four U.S. Air Force Defense Meteorological Satellite Program (DMSP) satellites [called F11 (now defunct), F13, F14, and F15] and the National Aeronautics and Space Administration Tropical Rainfall Measuring Mission (NASA TRMM) satellite. The resolution of this dataset is roughly 40 km. The procedure for the derivation of rain rates from these satellites is described in Krishnamurti et al. (2000b, 2001). Here, we make use of several current microwave data–based rain-rate algorithms. Given these five satellites, it is possible to obtain a reasonable coverage of rainfall-rate estimates every 6 h between 50°S and 50°N.

c. OLR datasets

These datasets are also needed within the physical initialization for the FSU models, where an (outgoing longwave radiation) OLR-based rain-rate algorithm is used to derive a first-guess rainfall, following Krishnamurti and Bounoua (1996). Gridded OLR fields are received with a delay of 1 day from NESDIS. The resolution of this dataset was 1.5° latitude–longitude through 1999 and is currently received at a resolution of 1.0° latitude–longitude.

d. Multimodel products

Hurricane track and intensity forecast datasets, described earlier, were also received from a number of U.S. and global modeling centers. Primarily, these include forecasts at 12-hourly intervals from NCEP, GFDL, NOGAPS, Met Office, FSU (three model inputs), the statistical and in-house models of the NHC/Hurricane Research Division (HRD), and the official forecasts from the NHC. These model components were referenced in section 2 and are further described later in this paper.

5. Superensemble track and intensity forecasts for 1998

For a summary of the 1998 Atlantic hurricane season, please refer to Pasch et al. (2001). A number of forecast models provided hurricane forecast information, such as NOAA/NCEP, GFDL, Met Office, NRL/NOGAPS, the official forecasts of TPC/NHC, and the FSU suite of models. Table 2 provides a short summary of these models. In addition to these forecasts, there exist in-house prediction models at the National Hurricane Center that routinely provide statistical hurricane intensity forecasts (SHIPS, SHIFOR). Unfortunately, it was not possible to acquire uniform datasets for a string of years without major model changes impacting the forecast datasets. Four of the aforementioned models—NCEP Medium-Range Forecast (MRF) FSU, GFDL, and NOGAPS—underwent some major resolution and internal changes after the 1997 season. Thus, it was not possible to derive applicable statistics based on the multimodel performance with the 1997 or prior datasets and use those to improve the forecasts during the 1998 Atlantic hurricane season. It was still possible to produce and assess forecasts for each tropical system of the 1998 season using a cross-validation approach. This entailed deriving the multimodel statistics from all of the storms of 1998, sequentially, excluding the specific storm that was being forecasted. This cross-validation technique is a robust approach for assessing the validity of the proposed method for forecasts of the 1998 systems. There were 14 named storms in 1998, each of which lasted one or more days. Thus, it was possible to develop a multimodel forecast database for 78 common forecast cases from the 1998 season. The databases are the multimodel forecasts, the observed and the official forecast estimate of the track (position) and intensity (maximum sustained surface wind speed) every 12 h, starting from an initial time and ending at hour 72 (with the FSU models continuing to 144 hours), as dictated by the termination of such forecasts. Separate statistics were calculated for each of these forecast periods. The coefficients (weights) for the models vary for each forecast period (i.e., hours 12, 24, … 144). It should be noted that not all forecast runs were carried out to 3 days or beyond (e.g., landfalling or dissipating system). Thus, the number of training datasets varies for each of these forecast time periods. We found that separate weights for each forecast period gave better results rather than providing a single homogeneous set of multimodel statistics for all of the forecast periods. A complete dataset for all the above models, observations, and official forecast was essential for completing the required statistics.

First, we shall examine the overall statistics of the track and intensity forecasts for the year 1998. Here, skills for day-1, day-2, and day-3 forecasts are shown. The results for the superensemble track and intensity forecasts (cross validation) are presented in Figs. 2a,b. During 1998, the most skillful track forecasts came from the NHC forecasters. The superensemble forecasts for the season were superior to those of all other models and official forecasts for each of the three days. The superensemble track forecasts, in the training phase, have position errors of the order of 100, 150, and 200 km for days 1, 2, and 3 of forecasts, respectively. The corresponding intensity rms errors for the superensemble “forecasts” are 5, 7.6, and 9.7 m s−1 for days 1, 2, and 3 of the forecasts. Intensity forecasts from the superensemble were only slightly better than those for the best models. The rms errors of intensity for the control (i.e., the training) and the forecast periods from the superensemble were better than those of all other models. Also shown in Fig. 2a is the skill of the ensemble mean, which was slightly worse than that of the superensemble. The intensity errors, in Fig. 2b, are expressed in meters per second [these are the maximum sustained surface (10 m) winds, reported by the respective models]. The improvements of intensity forecasts, by the superensemble, over the best model, are roughly 10%–25%.

Goerss (2000) showed that the error of the ensemble mean compared to forecast errors of the individual models (such as GFDL, NOGAPS, and the U.K. Met Office models) is less during a season of typhoon forecasts. The ensemble mean has a tendency to reduce errors of the individual models and produce slightly better forecasts. In this study, we display position and intensity errors of both the ensemble mean and the superensemble for the hurricanes of 1998 and 1999. The superensemble tends to improve the forecast more in comparison to the ensemble mean. The reason for this is that the superensemble is more selective in the bias removal of the member models. The rms track errors are improved over the best model by 10, 40, and 80 km for days 1, 2, and 3 from the superensemble forecasts. Although that appears to be a small improvement, it is still significant, considering the importance of landfall forecasts on issues of loss of life and property arising from hurricanes. An example of a 72-h superensemble track forecast for Hurricane Georges with two different start dates (22 and 23 September 1998) is shown in Figs. 3a,b.

Specific examples of intensity forecasts from the superensemble are shown for Hurricanes Bonnie (Figs. 4a,b) and Georges (Figs. 5a,b). Two separate initial conditions were chosen for each of these hurricanes in these diagrams. A number of models provided 3-day forecasts; these include FSU (three input models), NOGAPS, and GFDL (not all are illustrated in Figs. 3a,b). We have added two other intensity-forecast estimates provided by the National Hurricane Center—SHIPS and SHIFOR—which are statistical models described by DeMaria and Kaplan (1997). SHIFOR is a statistical hurricane forecast model developed in house at the Hurricane Research Division of NOAA. This is based on climatology and persistence, applies only to storms over the ocean, and is analogous to the CLIPER model for storm-track prediction. SHIPS is another simple climatological scheme of the Hurricane Research Division that makes use of parameters such as maximum possible intensity, current intensity, vertical shear of the tropospheric horizontal wind, persistence of intensity change over the last 12 h, eddy flux convergence of momentum at 200 hPa, zonal wind, and temperature within 1000 km of the storm center. These schemes are described in further detail in DeMaria and Kaplan (1997).

The longer-range forecasts for days 4, 5, and 6 came from the FSU global and regional models, which provided the forecast inputs through this period. Overall, it is apparent that some skill in the intensity forecasts is evident from the superensemble. The intensity forecasts from SHIPS and SHIFOR were reasonable for the first 3 days of forecasts; however, it is apparent from the overall statistics, presented earlier, that the superensemble outperforms all other models in the intensity forecast over the season.

6. Summary of 1999 Atlantic hurricane season, including the FSU real-time superensemble track and intensity forecasts

The 1999 Atlantic hurricane season had above-average activity, with 12 named tropical cyclones, including the most category-4 hurricanes since 1886, numbering five. For the full summary of this season, please refer to Lawrence et al. (2001). Fortunately for the 1999 superensemble forecasts, the 1998-based hurricane training set, both with respect to model characteristics and tropical system structure, appears to have been very representative for the tropical cyclones of 1999. Below we will present the skill for both the track and intensity forecasts.

The skill of track forecasts for all of the 1999 Atlantic hurricanes from the participating multimodels and the superensemble is summarized in Fig. 6a and the intensity errors are presented in Fig. 6b. In order to assess the skills, all of the interpolated forecasts available at 1200 UTC were compared. Here we have performed a linear interpolation in time where it was necessary. The following multimodels were compared: the interpolated National Hurricane Center's official forecast (NHCI); the interpolated NCEP AVN (AVNI), the operational NCEP Hydrometeorological Prediction Center forecast (OHPC), the interpolated GFDL MMM Model (GFDI) being run at NCEP, the interpolated equivalent baratropic VICBAR model (VCBI), the interpolated U.K. Met Office Model (UKMI), the interpolated NOGAPS model (NGPI), the ensemble mean (ENSM) and the FSU superensemble (FSU SENS). The ordinate in this diagram illustrates the (great circle) rms position error in kilometers and abscissa denotes forecast time intervals (every 12 h).

Here we note that the calculated position errors of the superensemble are less than those of the multimodels and of the ensemble mean during the 3-day forecast. The day-3 reduction in position errors compared to the NOGAPS and the AVN is quite substantial (i.e., on the order of 100–150 km). The superensemble reduces the 3-day errors compared to the best model (the GFDL) by roughly 82 km. Overall, these reductions of errors by the superensemble seem quite encouraging.

For a further breakdown of storm-specific errors, Figs. 7a,b illustrate the track error for each storm with more than one forecast case for hour-48 and-72 forecasts, respectively. The best model forecast, the ensemble mean, and superensemble track errors are compared. Note that the superensemble regularly outperforms the best individual model, especially in the longer range (72-h). The objectivity of the superensemble is therefore appealing, since for each storm event, including between hours 24, 36, 48, and 72, the best individual performing model changes. The superensemble nearly always outperforms the ensemble mean for a given set of track forecasts. Therefore, the superensemble forecast can provide more useful operational guidance than any specific model or sets of averaged fields.

As for the intensity forecasts, the relative improvement at day 3 was less than with the track forecasts, and for hours 12 and 24, GFDI outperformed the FSU superensemble forecast. The small variation in error between the better models suggests either limited skill, especially in the longer range, or low correlation with respect to observations, which has a negative impact on the superensemble weighting coefficients. Also, the relative skill improvement with the intensity forecasts is much less than with the track forecasts. Thus, presently any improvement using the superensemble technique for intensity forecasts is limited without further enhancement to this technique.

7. Track forecasts for key Atlantic hurricanes of 1999

Although there were eight Atlantic hurricanes during 1999, we shall only illustrate the tracks and intensity forecasts for some of these hurricanes, that is, Floyd, Irene, and Lenny. However, it should be noted that the overall skill score summaries include all of the storms of 1999. Table 3 provides a list of the storms of 1999, noting that the stronger hurricanes had 60 m s−1 sustained surface winds. The track forecasts for Hurricane Floyd made for start dates 1200 UTC 13 September 1999 and 1200 UTC 14 September 1999 are shown in Figs. 8a,b. In these illustrations, we show the interpolated 1200 UTC official forecasts from the National Hurricane Center, along with the forecasts of AVN, Met Office, GFDL, and the FSU superensemble forecasts. Also shown is the official best (observed) track determined by the National Hurricane Center in postanalysis. Apparent here is that the superensemble forecast tracks were, in general, superior to the other models for the more intense storms and also exhibited a greater degree of consistency. One of the problems noted during the real-time operation was the frequent prediction for landfall at Florida, Georgia, South Carolina, and finally North Carolina by various individual models. This led to major evacuation of the population from coastal areas from Florida to the Carolinas. The superensemble forecast tracks were more consistent with respect to individual model forecast landfall positions as well.

The predicted and observed best tracks for Hurricanes Irene (15 October 1999) and Lenny (15 and 16 November 1999) are illustrated in Figs. 9 and 10a,b, respectively. Here again we note the overall superior performance of the FSU superensemble compared to the individual member models and the official best-track forecast of the National Hurricane Center. The forecasts made on 14 October for Irene were not very impressive by any of the models or by the superensemble. Thereafter, the forecasts for Irene from the superensemble were superior compared to other forecasts. The forecast of Lenny, by the superensemble, toward St. Martin, past St. Croix in the U.S. Virgin Islands, versus the predicted Puerto Rico track by all the multimodels demonstrated a major strength of the superensemble. That was especially noteworthy for the 72-h forecasts made on the 16 November 1999 (Fig. 10b). In these forecasts, most multimodels predicted a track directly over the center of Puerto Rico. Hurricane Lenny provided a major challenge for all the models. The predicted track in most of the better models called for a northeastward track. The real-time superensemble forecasts (based on the multimodels) consistently called for a more eastward motion toward St. Martin. At the time, we questioned the accuracy of our real-time forecasts as they were being prepared at FSU. No previous storms in our training database had traversed eastward at such a low latitude in November. It became evident (after these excellent forecasts from the superensemble) that the latitude was not entirely a governing factor in determining the multimodel biases. There were several other factors such as the intermodel performance in past forecasts that contributed to the overall correction (described by the statistics).

8. Intensity forecasts for key Atlantic hurricanes of 1999

Our experience with the skill of intensity forecasts from the multimodel superensemble for 1999 was quite similar to that for 1998. Intensity forecasts have been a difficult issue for tropical numerical weather prediction. In the overall statistics of intensity forecasts, where all the storms of 1999 were included, we did notice a slight improvement from the superensemble over all member models as well as the official best observed estimates of intensity for the first three forecast days. In general, the superensemble forecasts for Hurricanes Floyd, Irene, and Lenny of 1999 were superior to those of the member models. The superensemble forecast of intensity for Hurricane Floyd shown in Fig. 11 is generally within a category to a category and a half (based on the Saffir–Simpson scales) of the observed estimates of the intensity. These observed intensities display some large fluctuations in intensity during a 4–6-day forecast period. These are in fact obtained from direct research aircraft surveillance (such as the NOAA P3 flights and the dropwindsonde data). The models generally fail to identify such intensity fluctuations. The forecasts of storm intensity for Floyd show the nature of the intensity forecast problem, especially beyond 48 hours. On 11 September, the maximum intensity was underestimated during most of the 5-day forecast. The forecasts for 12 September underestimated the intensity during the first 24 h. Maximum winds were reasonably handled through forecast at 84 h and thereafter the intensity was overestimated. The forecasts for 13 September started out with a reasonable category-5 storm; thereafter, a slow decrease of intensity somewhat consistent with the observed best estimates was noted. The model winds were, however, roughly one to two categories higher. The forecasts made on 14 and 15 September clearly demonstrated a weakening in the intensity in the ensuing 72 h. The summary of the 1999 Atlantic hurricane intensity errors is illustrated in Fig. 6b. What is apparent here is that the intensity forecasts from the superensemble are only slightly better than the member models, in general. The intensity forecasts for Hurricanes Irene and Lenny are presented in Figs. 12 and 13a,b, respectively. Here we note some improvement of skill beyond those of the multimodels.

Overall, the intensity issue needs to be addressed more directly using the superensemble of mesoscale higher-resolution models including coupled atmosphere–ocean models. That work requires planning and careful implementation of multimodels and the training phase, and it is expected to make an impact in the intensity skills, especially beyond days 2 and 3.

9. Concluding remarks

The official forecast of hurricanes at the NHC is often called an expert “consensus” forecast. Experienced forecasters at NHC prepare such a consensus from among the suite of model forecasts available to them, along with the past and current conditions of the tropical cyclone at hand. Although the superensemble is an objective consensus that attempts to collate the collective wisdom of the various models, it may still lack the intuitive knowledge of the experienced line forecasters. It is only when pieces of such intuitive knowledge is passed on in an objective manner that the superensemble can benefit from it. The two approaches complement each other and, presently, both are available for operational forecasts at the hurricane centers.

It was possible to obtain improved results on hurricane forecasts for the storms of the 1998 season where we had used a cross-validation approach (i.e., using all the storms of the season to provide the superensemble training weights except the one being predicted). For the 1999 season we had started with the statistics derived for the year 1998 and updated those as the newer storms had undergone their forecast and verification phase. These two approaches for the years 1998 and 1999 appear to work extremely well in the context of the superensemble forecasts. The 1200 UTC rms track errors for the tropical storm and hurricane cases during the year 1999 from the superensemble forecasts are 155.1, 248.1, and 332.3 km for days 1, 2, and 3, respectively. These averaged errors represent improvement over the best individual model by about 10, 40, and 80 km, respectively.

As we proceeded to the year 2000, however, the method pursued for the 1999 season did not work in providing the major improvements noted earlier. There appeared to be one major reason for this behavior of the superensemble. Midseason major model changes were noted in several of the models, particularly the UK Met office, NCEP AVN, and the GFDL impacted by the changes in the AVN (as it provides the boundary conditions to GFDL MMM). This made it somewhat difficult to maintain high skills in real-time since midcourse (year 2000 based) correction of statistics could be tested only experimentally.

In Fig. 14 we present the track error statistics for the year 2000. When the model coefficients obtained by using the year 1999 storm cases in the training phase of the superensemble were used to predict the tracks of hurricanes during 2000, the errors of the superensemble were larger (FSU SENS, dark bars next to the far right of each time group) compared to other member models and the ensemble mean. During this same season, we experimentally retrained the coefficients using the earlier 2000 storm cases and found that the track errors from this version of the superensemble (SE EXP, hatched bar on the far right of each group) were reduced. Even still, the ensemble mean outperformed the superensemble in nearly all time periods for both cases (FSU SENS and SE EXP). This establishes the positive impact of the training phase of the superensemble on the forecast phase and the fact that the participating member models should have a reasonable consistency in terms of the model physics and initialization schemes. When significant changes are made to the forecast models, a sufficiently large, updated training set is required to capture the updated model characteristics. Otherwise, the superensemble may not outperform an ensemble mean of the better models.

Figures 6a,b and 7a,b show that some improvement beyond the other component forecasts is being realized from use of the multimodel superensemble. The track forecast improvements in these summary diagrams do not quite reflect the details of the track forecasts for some of the difficult storms. During the course of the real-time events in 1999, as Hurricanes Floyd, Irene, and Lenny traversed through the western Atlantic and the Caribbean, the superensemble tracks (such as those shown in Figs. 8a,b, 9, and 10a,b) revealed small improvements by days 1, 2, and 3, which led to considerably improved positions for the landfall for these storms. The intensity forecasts from the superensemble are only slightly better than the best current operational models.

Much progress has been made in recent years toward the improvement of mesoscale research models for hurricane forecasting. These are beginning to use nonhydrostatic detailed microphysical models (Wang 2001). Given results from a number of such models, it may be possible to obtain further improvements in the forecasts of tracks and especially the intensity using superensemble techniques. Recent results of high-resolution coupled atmosphere–ocean modeling of hurricanes attempted by Bender and Ginis (2000) appear very promising for the intensity issue. That model was able to predict the heat content of the oceanic mixed layers and the sustained presence of warm SST anomalies that affect the hurricane's life cycle. Here again, we feel that much further progress should be possible if a number of such models were put together for the construction of a multimodel superensemble.

Acknowledgments

The research reported here was funded by NASA Grants NAG5-9662 and NAG8-1537, NSF Grants ATM-9910526 and ATM-0108741, and FSURF COE. The authors wish to express their special thanks to Dr. Tony Hollingsworth for providing ECMWF analysis in real time. We also thank Dr. Nihat Cubukcu, Mr. Brian Mackey, and Mr. Adam O'Shay for their reviews and aid with this effort.

REFERENCES

  • Aberson, S. D., and M. DeMaria, 1994: Verification of a nested barotropic hurricane track forecast model (VICBAR). Mon. Wea. Rev., 122 , 28042815.

    • Search Google Scholar
    • Export Citation
  • Bender, M. A., and I. Ginis, 2000: Real-case simulations of hurricane–ocean interaction using a high-resolution coupled model: Effects on hurricane intensity. Mon. Wea. Rev., 128 , 917945.

    • Search Google Scholar
    • Export Citation
  • Bender, M. A., R. J. Ross, R. E. Tuleya, and Y. Kurihara, 1993: Improvements in tropical cyclone track and intensity forecasts using the GFDL initialization system. Mon. Wea. Rev., 121 , 20462061.

    • Search Google Scholar
    • Export Citation
  • Bengtsson, L., M. Botzet, and M. Esch, 1995: Hurricane-type vortices in a general circulation model. Tellus, 47 , 175196.

  • Businger, J. A., J. C. Wyngard, Y. Izumi, and E. F. Bradley, 1971: Flux profile relationship in the atmospheric surface layer. J. Atmos. Sci., 28 , 181189.

    • Search Google Scholar
    • Export Citation
  • Chen, Q-S., L-E. Bai, and D. H. Bromwich, 1997: A harmonic-Fourier spectral limited-area model with an external wind lateral boundary condition. Mon. Wea. Rev., 125 , 143167.

    • Search Google Scholar
    • Export Citation
  • Cocke, S., 1998: Case study of Erin using the FSU Nested Regional Spectral Model. Mon. Wea. Rev., 126 , 13371346.

  • DeMaria, M., and J. Kaplan, 1994: Sea surface temperature and the maximum intensity of Atlantic tropical cyclones. J. Climate, 7 , 13241334.

    • Search Google Scholar
    • Export Citation
  • DeMaria, M., and J. Kaplan, 1997: An operational evaluation of a statistical hurricane intensity prediction scheme (SHIPS). Preprints, 22d Conf. on Hurricanes and Tropical Meteorology, Ft. Collins, CO, Amer. Meteor. Soc., 280–281.

    • Search Google Scholar
    • Export Citation
  • Goerss, J. S., 2000: Tropical cyclone track forecasts using an ensemble of dynamical models. Mon. Wea. Rev., 128 , 11871193.

  • Goerss, J. S., and R. Jeffries, 1994: Assimilation of synthetic tropical cyclone observations into the Navy Operational Global Atmospheric Prediction System. Wea. Forecasting, 9 , 557576.

    • Search Google Scholar
    • Export Citation
  • Harshvardan, and T. G. Corsetti, 1984: Longwave parameterization for the UCLA/GLAS GCM. NASA Tech. Memo. 86072, 52 pp.

  • Heming, J. T., 1997: UK Meteorological Office forecast performance during the unusual Atlantic hurricane season of 1995. Preprints, 22d Conf. on Hurricanes and Tropical Meteorology, Ft. Collins, CO, Amer. Meteor. Soc., 511–512.

    • Search Google Scholar
    • Export Citation
  • Hogan, T., and T. Rosmond, 1991: The description of the Navy Operational Global Atmospheric Prediction System's spectral forecast model. Mon. Wea. Rev., 119 , 17861815.

    • Search Google Scholar
    • Export Citation
  • Holland, G. J., 1983: Tropical cyclone motion: Environmental interaction plus a beta effect. J. Atmos. Sci., 40 , 328342.

  • Horsfall, F. M., M. DeMaria, and J. M. Gross, 1997: Optimal use of large-scale boundary and initial fields for limited-area hurricane forecast models. Preprints, 22d Conf. on Hurricanes and Tropical Meteorology, Ft. Collins, CO, Amer. Meteor. Soc., 571–572.

    • Search Google Scholar
    • Export Citation
  • Hoyer, J. M., 1987: The ECMWF spectral limited-area model. Proc. ECMWF Workshop on Techniques for Horizontal Discretization in Numerical Weather Prediction Models, Shinfield Park, Reading, United Kingdom, ECMWF, 343–359.

    • Search Google Scholar
    • Export Citation
  • Jarvinen, B. R., and C. J. Neumann, 1979: Statistical forecasts of tropical cyclone intensity. NOAA Tech. Memo. NWS NHC-10, 22 pp.

  • Juang, H. H-M., and M. Kanamitsu, 1994: The NMC Nested Regional Spectral Model. Mon. Wea. Rev., 122 , 326.

  • Kanamitsu, M., 1975: On numerical prediction over a global tropical belt. Rep. 75-1, Department of Meteorology, Florida State University, 282 pp.

    • Search Google Scholar
    • Export Citation
  • Kanamitsu, M., K. Tada, K. Kudo, N. Sato, and S. Isa, 1983: Description of the JMA operational spectral model. J. Meteor. Soc. Japan, 61 , 812828.

    • Search Google Scholar
    • Export Citation
  • Kitade, T., 1983: Nonlinear normal mode initialization with physics. Mon. Wea. Rev., 111 , 21942213.

  • Krishnamurti, T. N., and L. Bounoua, 1996: An Introduction to Numerical Weather Prediction Techniques,. CRC Press, 293 pp.

  • Krishnamurti, T. N., S. Low-Nam, and R. Pasch, 1983: Cumulus parameterization and rainfall rates II. Mon. Wea. Rev., 111 , 815828.

  • Krishnamurti, T. N., D. Oosterhof, and N. Dignon, 1989: Hurricane prediction with a high-resolution global model. Mon. Wea. Rev., 117 , 631669.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., J. Xue, H. S. Bedi, K. Ingles, and D. Oosterhof, 1991a: Physical initialization for numerical weather prediction over the tropics. Tellus, 43AB , 5381.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., K. S. Yap, and D. K. Oosterhof, 1991b: Sensitivity of tropical storm forecast to radiative destabilization. Mon. Wea. Rev., 119 , 21762205.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., H. S. Bedi, G. D. Rohaly, D. K. Oosterhof, R. C. Torres, E. Williford, and N. Surgi, 1996: Physical initialization. J. Atmos. Oceans, 35 , 369398.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., W. Han, and D. K. Oosterhof, 1998: Sensitivity of hurricane intensity forecasts to physical initialization. Meteor. Atmos. Phys., 65 , 171181.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., C. M. Kishtawal, T. LaRow, D. Bachiochi, Z. Zhang, C. E. Williford, S. Gadgil, and S. Surendran, 1999: Improved skills for weather and seasonal climate forecasts from multimodel superensemble. Science, 285 , 15481550.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., C. M. Kishtawal, D. W. Shin, and C. E. Williford, 2000a: Multimodel superensemble forecasts for weather and seasonal climate. J. Climate, 13 , 41964216.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., C. M. Kishtawal, T. LaRow, D. Bachiochi, Z. Zhang, C. E. Williford, S. Gadgil, and S. Surendran, 2000b: Improving tropical precipitation forecasts from a multianalysis superensemble. J. Climate, 13 , 42174227.

    • Search Google Scholar
    • Export Citation
  • Krishnamurti, T. N., and Coauthors. 2001: Real-time multianalysis–multimodel superensemble forecasts of precipitation using TRMM and SSM/I products. Mon. Wea. Rev., 129 , 28612883.

    • Search Google Scholar
    • Export Citation
  • Kuo, H. L., 1965: On formation and intensification of tropical cyclones through latent heat release by cumulus convection. J. Atmos. Sci., 22 , 4063.

    • Search Google Scholar
    • Export Citation
  • Kuo, H. L., 1974: Further studies of the parameterization of the influence of cumulus convection on large-scale flow. J. Atmos. Sci., 31 , 12321240.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., and M. A. Bender, 1982: Structure and analysis of the eye of a numerically simulated tropical cyclone. J. Meteor. Soc. Japan, 60 , 381395.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., G. J. Tripoli, and M. A. Bender, 1979: Design of a movable nested-mesh primitive equation model. Mon. Wea. Rev., 107 , 239249.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., M. A. Bender, and R. T. Ross, 1993: An initialization scheme of hurricane models by vortex specification. Mon. Wea. Rev., 121 , 20302045.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., M. A. Bender, R. E. Tuleya, and R. J. Ross, 1995: Improvements in the GFDL hurricane prediction system. Mon. Wea. Rev., 123 , 27912801.

    • Search Google Scholar
    • Export Citation
  • Kurihara, Y., R. E. Tuleya, and M. A. Bender, 1998: The GFDL hurricane prediction system and its performance in the 1995 hurricane season. Mon. Wea. Rev., 126 , 13061322.

    • Search Google Scholar
    • Export Citation
  • Lacis, A. A., and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth's atmosphere. J. Atmos. Sci., 31 , 118133.

    • Search Google Scholar
    • Export Citation
  • Lawrence, M. B., L. A. Avila, J. L. Beven, J. L. Franklin, J. L. Guiney, and R. J. Pasch, 2001: Atlantic hurricane season of 1999. Mon. Wea. Rev., 129 , 30573084.

    • Search Google Scholar
    • Export Citation
  • Lord, S. J., 1991: A bogussing system for vortex circulations in the National Meteorological Center global forecast model. Preprints, 19th Conf. on Hurricanes and Tropical Meteorology, Miami, FL, Amer. Meteor. Soc., 328–330.

    • Search Google Scholar
    • Export Citation
  • Lord, S. J., H. E. Willoughby, and J. M. Piotrowicz, 1984: Role of a parameterized ice-phase microphysics in an axisymmetric, nonhydrostatic tropical cyclone model. J. Atmos. Sci., 41 , 28362848.

    • Search Google Scholar
    • Export Citation
  • Louis, J. F., 1979: A parametric model of vertical eddy fluxes in the atmosphere. Bound.-Layer Meteor., 17 , 187202.

  • Mackey, B. P., and T. N. Krishnamurti, 2001: Ensemble forecast of a typhoon flood event. Wea. Forecasting, 16 , 399415.

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

  • Neumann, C. J., 1972: An alternate to the HURRAN tropical cyclone forecast system. NOAA Tech. Memo. NWS SR-62, 22 pp.

  • Neumann, C. J., and C. J. McAdie, 1991: A revised National Hurricane Center NHC83 model (NHC90). NOAA Tech. Memo. NWS NHC-44, 35 pp.

  • Pasch, R. J., L. A. Avila, and J. L. Guiney, 2001: Atlantic hurricane season of 1998. Mon. Wea. Rev., 129 , 30853099.

  • Reynolds, R. W., and T. M. Smith, 1993: An improved real-time global sea surface temperature analysis. J. Climate, 6 , 114119.

  • Rotunno, R., and K. A. Emanuel, 1987: An air–sea interaction theory for tropical cyclones. Part II: Evolutionary study using a nonhydrostatic axisymmetric numerical model. J. Atmos. Sci., 44 , 542561.

    • Search Google Scholar
    • Export Citation
  • Surgi, N., H-L. Pan, and S. J. Lord, 1998: Improvement of the NCEP global model over the Tropics: An evaluation of model performance during the 1995 hurricane season. Mon. Wea. Rev., 126 , 12871305.

    • Search Google Scholar
    • Export Citation
  • Thu, T. V., and T. N. Krishnamurti, 1992: Vortex initialization for typhoon track prediction. J. Meteor. Atmos. Phys., 47 , 117126.

  • Tiedke, M., 1984: The sensitivity of the time-mean large-scale flow to cumulus convection in the ECMWF model. Proc. ECMWF Workshop on Convection in Large-Scale Numerical Models, Reading, Berkshire, United Kingdom, ECMWF, 297–316.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., S. Tibaldi, and A. J. Simmons, 1983: Reduction of systematic forecast errors in the ECMWF model through the introduction of envelope orography. Quart. J. Roy. Meteor. Soc., 109 , 683718.

    • Search Google Scholar
    • Export Citation
  • Wang, Y., 2001: An explicit simulation of tropical cyclones with a triply nested movable mesh primitive equation model: TCM3. Part I: Model description and control experiment. Mon. Wea. Rev., 129 , 13701394.

    • Search Google Scholar
    • Export Citation
  • Williford, C. E., R. J. Correa-Torres, and T. N. Krishnamurti, 1998: Tropical cyclone forecasts made with the FSU Global Spectral Model. Mon. Wea. Rev., 126 , 13321336.

    • Search Google Scholar
    • Export Citation
  • Zhang, D-L., and E. Altshuler, 1999: The effects of dissipative heating on hurricane intensity. Mon. Wea. Rev., 127 , 30323038.

  • Zhang, Z., and T. N. Krishnamurti, 1999: A perturbation method for hurricane ensemble predictions. Mon. Wea. Rev., 127 , 447469.

APPENDIX

An Outline of the FSU Global Spectral Model

The global model used in this study is identical in all respects to that used in Krishnamurti et al. (1991a). The following is an outline of the global model:

  1. independent variables—x, y, σ, t;

  2. dependent variables—vorticity, divergence, surface pressure, vertical velocity, temperature, and humidity;

  3. horizontal resolution—triangular 126 waves;

  4. vertical resolution—14 layers between roughly 50 and 1000 hPa;

  5. semi-implicit time differencing scheme;

  6. envelope orography (Wallace et al. 1983);

  7. centered differences in the vertical for all variables except humidity, which is handled by an upstream-differencing scheme;

  8. fourth-order horizontal diffusion (Kanamitsu et al. 1983);

  9. Kuo-type cumulus parameterization (Kuo 1965, 1974; Krishnamurti et al. 1983);

  10. shallow convection (Tiedtke 1984);

  11. dry convective adjustment;

  12. large-scale condensation (Kanamitsu 1975);

  13. surface fluxes via similarity theory (Businger et al. 1971);

  14. vertical distribution of fluxes utilizing diffusive formulation where the exchange coefficients are functions of the Richardson number (Louis 1979);

  15. longwave and shortwave radiative fluxes based on a band model (Harshvardhan and Corsetti 1984; Lacis and Hansen 1974);

  16. diurnal cycle with respect to the radiative processes;

  17. parameterization of low, middle, and high clouds based on threshold relative humidity for radiative transfer calculations;

  18. surface energy balance coupled to the similarity theory (Krishnamurti et al. 1991b);

  19. nonlinear normal model initialization—five vertical modes (Kitade 1983); and

  20. physical initialization (Krishnamurti et al. 1991a).

Fig. 1.
Fig. 1.

FSU superensemble production time line

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 2.
Fig. 2.

(a) The 1998 Atlantic tropical system cross-validation-based track errors, hours 12–72, including FSU superensemble and ensemble mean forecasts; (b) the 1998 Atlantic tropical system cross-validation-based intensity errors, hours 12–72, including FSU superensemble and ensemble mean forecasts

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 3.
Fig. 3.

(a) Hurricane Georges 72-h track forecast, starting 1200 UTC 22 Sep 1998; (b) Hurricane Georges 72-h track forecast, starting 1200 UTC 23 Sep 1998

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 4.
Fig. 4.

(a) Hurricane Bonnie 72-h intensity forecast, starting 1200 UTC 20 Aug 1998; (b) Hurricane Bonnie 72-h intensity forecast, starting 1200 UTC 25 Aug 1998

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 5.
Fig. 5.

(a) Hurricane Georges 72-h intensity forecast, starting 1200 UTC 21 Sep 1998; (b) Hurricane Georges 72-h intensity forecast, starting 1200 UTC 22 Sep 1998

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 6.
Fig. 6.

(a) The 1999 Atlantic tropical system track errors, hours 12–72, including FSU superensemble and ensemble mean forecasts; (b) the 1999 Atlantic tropical system intensity errors, hours 12–72, including FSU superensemble and ensemble mean forecasts

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 7.
Fig. 7.

(a) The 1999 Atlantic individual storm track errors, hour 48, including FSU superensemble and ensemble mean forecasts; (b) the 1999 Atlantic individual storm track errors, hour 72, including FSU superensemble and ensemble mean forecasts

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 8.
Fig. 8.

(a) Hurricane Floyd 72-h track forecast, starting 1200 UTC 13 Sep 1999; (b) Hurricane Floyd 72-h track forecast, starting 1200 UTC 14 Sep 1999

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 9.
Fig. 9.

Hurricane Irene 72-h track forecast, starting 1200 UTC 15 Oct 1999

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 10.
Fig. 10.

(a) Hurricane Lenny 72-h track forecast, starting 1200 UTC 15 Nov 1999; (b) Hurricane Lenny 72-h track forecast, starting 1200 UTC 16 Nov 1999

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 11.
Fig. 11.

Hurricane Floyd 72-h intensity forecast, starting 1200 UTC 13 Sep 1999

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 12.
Fig. 12.

Hurricane Irene 48-h intensity forecast, starting 1200 UTC 15 Oct 1999

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 13.
Fig. 13.

(a) Hurricane Lenny 72-h intensity forecast, starting 1200 UTC 15 Nov 1999; (b) Hurricane Lenny 36-h intensity forecast, Starting 1200 UTC 17 Nov 1999

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Fig. 14.
Fig. 14.

The 2000 Atlantic tropical system track errors, hours 12–72, including FSU superensemble and ensemble mean forecasts

Citation: Monthly Weather Review 131, 8; 10.1175//2571.1

Table 1.

List of acronyms

Table 1.
Table 2.

Numerical models used for superensemble prediction in this study

Table 2.
Table 3.

Storm statistics for 1999 Atlantic named tropical systems: tropical storm (TS) and hurricane (H)

Table 3.
Save