U.S. Dept. of Commerce / NOAA / OAR / GFDL *Disclaimer

 

7. HURRICANE DYNAMICS

GOALS
  7.1 HURRICANE PREDICTION SYSTEM

ACTIVITIES F00

     7.1.1 Performance in the 1999 Hurricane Season           The 1999 Atlantic hurricane season, comprising 12 storms, turned out to be another busy year for the GFDL model. More than 290 forecasts were made at NCEP for the National Hurricane Center (NHC) at four initialization times per day. In the Eastern and Central Pacific, the GFDL model was run at NCEP for 162 cases for the NHC. As in past years, the U.S. Navy continues to run their separate GFDL forecast system to support requirements in the Northwest Pacific, Indian Ocean, and various Southern Hemispheric Basins. In addition, during the 1999 Atlantic season the Navy ran the "GFDN" version of the GFDL model after the NCEP fire to provide support to the NHC when NCEP forecasts were not available.

     7.1.2 Analysis of the Forecast Results

          Skillful forecasts of intensity have been a challenging goal in hurricane forecasting. For the 1999 season, however, the GFDL system did exhibit skill at 48hr and beyond. When adjusted for initial bias, the GFDL model showed skill relative to SHFR (climatology and persistence) for the entire forecast period with mean errors 11, 17, and 17 knots at 24, 48, and 72hrs, respectively. This is evidence that forecast system changes made in prior years, including the addition of initial asymmetries (FY99), resulted in improvements for the 1999 season. There are indications that improvements in resolution and physical parameterizations, such as ocean coupling, boundary layer and cumulus convection will lead to further improvements in these intensity forecasts (6.3, 6.4).

          On the other hand, the GFDL track forecast error in the 1999 season was again strongly affected by the difficulties in NCEP's near-storm tropical analysis. Implementation of the higher resolution T170L42 analysis/forecast system was delayed until the 2000 season. Nevertheless, the GFDL forecast errors of 68,124 and 187 nautical miles at 24, 48, and 36hr exhibited considerable skill and were competitive with the Navy and UKMET global model guidance. After the 1999 season, NCEP corrected and improved their near storm analysis and the GFDL model was rerun for the cases of Dennis, Floyd, and Gert with dramatic improvements: for 63 cases, the GFDL 72h forecasts from the reanalysis yielded skill 50% better than climatology and persistence compared to only 35% for the GFDL operational run. It is anticipated that the level of skill will be high when the analysis system is implemented for the 2000 season.

     7.1.3 The 2000 Hurricane Season

          The GFDL hurricane forecast system was successfully converted and implemented on NCEP's distributed memory computer system. The entire forecast system was converted, including pre- and post- processing stages, vortex bogussing, and forecast dissemination. Also implemented was the extension of forecast period from three to five days, which has been run successfully for tropical systems in the Atlantic and Eastern Pacific basin. The GFDL-URI (University of Rhode Island) coupled model has also been installed in parallel mode at NCEP. The POM ocean model was implemented to run in parallel with the atmospheric model, so little overhead was required in the forecast step of the coupled system.

PLANS FY01

          Performance of the GFDL system in the remaining 2000 season will be monitored. The coupled model forecasts will be compared to the operational model forecasts. Alternative physics packages will be investigated and integrated with the ocean-coupled model to improve intensity prediction. A more systematic study of grid resolution and grid configurations will be attempted in an effort to identify an efficient forecast system for the 2001 hurricane season.

  7.2 HURRICANE PREDICTION CAPABILITY

ACTIVITIES FY00

     7.2.1 Extended Prediction           Extended prediction of hurricane tracks up to 120 hours using the GFDL system has been successfully implemented and has become operational at NCEP for the 2000 hurricane season. A collaborative study continues with the tropical forecast centers of the Navy, UKMET, and ECMWF in evaluating skill to 120h for the 1995 through 1998 seasons. All models, including the GFDL model, exhibit skill into days 4 and 5 when compared to a climatology and persistence reference. It was found that simple ensembles of the various forecast models yield even more skill. Interestingly, the GFDL model was found to be a key member of this ensemble; that is, the GFDL model together with any other global model gave the best track forecast.

     7.2.2 Impact of Satellite-Observed Winds on GFDL Forecasts

          A series of experimental forecasts was performed to evaluate the impact of enhanced satellite-derived winds on numerical hurricane track predictions (mr). The winds were derived from GOES-8 multi spectral radiance observations by tracking cloud and water vapor patterns from successive satellite images. A three-dimensional optimum interpolation method was developed to assimilate the satellite winds directly into the GFDL hurricane prediction system. A series of parallel forecasts was then performed, both with and without the assimilation of GOES winds. Except for the assimilation of the satellite winds, the model integration procedures are identical in all other respects. Over 100 cases were examined from 11 different storms covering 3 seasons (1996-1998), making it possible to account for and examine the case-to-case variability in the forecast results when performing our assessment. On average, assimilation of the GOES winds leads to statistically-significant improvements for all forecast periods, with the relative reductions in track error ranging from ~5% at 12 hours to ~12% at 36 hours. The majority of forecast cases were improved using the satellite winds, with roughly 3 improved forecasts for every 2 degraded ones. The errors in intensity forecasts for the 24-72 hour periods were also improved by 4-8%.

          Inclusion of the satellite winds also dramatically reduced the westward bias which has been a persistent feature of the GFDL model forecasts, implying that much of this bias may be related to errors in the initial conditions rather than a deficiency in the model itself. A composite analysis of the deep-layer flow fields suggests that the reduction in track error and correction of the westward bias may be associated with the ability of the GOES winds to more accurately depict the strength of vorticity gyres in the environmental flow. These gyres result primarily from the interaction between the storm's vortex with the environmental gradient of vorticity. While these "-gyres" are common features in numerical models, documenting their presence in nature has been hampered by the lack of sufficient observations and the complex flow in the storm environment. As shown in Fig. 7.1, the assimilation of the satellite winds results in an

enhanced gyre dipole with greater anticyclonic circulation to the northeast of the storm and greater cyclonic circulation to the southwest of the storm. This suggests that, while the initial wind field does contain a gyre-like pattern, the strength of these gyres is under-analyzed relative to that inferred from the GOES retrievals. The combination of both statistical and physical analyses thus offers compelling evidence that the assimilation of satellite winds can significantly improve the accuracy of hurricane track forecasts.

     7.2.3 Sensitivity of GFDL Track Forecasts to Initial Conditions

          Since 1996, the UKMET global model has also been providing operational forecast guidance for storms in the Atlantic basin. On the average, this model was the best performer for track prediction during the 1999 season. In order to evaluate the performance of the GFDL forecast system based on this global analysis, forecasts of Hurricanes Dennis, Floyd, and Gert were rerun using the UKMET analysis as input into the initial condition (designated "GFDU"). In all of these cases, the initialization and forecast model procedures were otherwise identical to the operational GFDL. To evaluate the impact on the track forecast from the analysis alone, the lateral boundary forcing used for the GFDU forecasts was the same as that used for the operational forecasts (GFDL) obtained from the AVN model. In total, 79 cases were rerun using the GFDL model with the UKMET analysis. The GFDL forecasts initialized from the UKMET analysis showed considerable improvement compared to the operational model (Fig. 7.2). The 24, 48,

and 72h average track errors decreased from 63, 111, and 170 nm to 53, 85, and 132 nm, with a frequency of superior performance of 62, 71, and 60% at these time levels. Similarly, the skill relative to CLIPER increased from 38% and 39% at 48 and 72h, to 52% and 53% for GFDU. The average track error for GFDU was slightly better than the UKMET model at the longer time periods. For example, at 48h the skill relative to CLIPER was 54%, 47%, and 39% for the GFDU, UKMET, and GFDL models, respectively.

     7.2.4 Evaluation of Model Forecast Errors

          A comprehensive study of the spatial variation of model track and intensity forecast errors across the Atlantic Basin was undertaken. Such a spatial analysis can highlight geographic regions of stronger and weaker model performance, thus providing a more detailed look at a model's forecast biases than can be attained through average error statistics alone. Results indicate that for track, the barotropic and advection models have their regions of strongest performance in a band stretching from the deep tropics west-northwest through the Bahamas. The dynamic models, including the GFDL, Navy NOGAPS, and UKMET models, have their regions of strongest performance extending from the Bahamas and Leeward Islands northward to off the New England Coast. Almost all of the models in the study had difficulty with storms in the western Gulf of Mexico, with most models exhibiting a strong westward bias in that region. Most models also had trouble with storms in the northeastern Atlantic Ocean. For intensity, all of the models in the study exhibited an under-intensification bias for forecasts originating in the deep tropics. The GFDL model, which has an under-intensification bias across nearly the entire basin at 12 hours lead time, exhibited a wide region of relatively bias-free intensity forecasts across much of the western North Atlantic at 72 hours lead time.

PLANS FY01

          Analysis of extended range 5-day forecasts will continue. The investigation of forecast skill among forecast models of the various tropical operational centers will be completed. Developmental work on the initialization methodology will proceed with evaluation of past cases. In addition, work to upgrade the GFDL system by increasing the grid resolution and improving the physical parameterizations of the model will be given higher priority.

  7.3 BEHAVIOR OF TROPICAL CYCLONES

     7.3.1 Hurricane Intensity in a High-CO2 Climate

ACTIVITIES FY00

          The potential impact of a greenhouse gas-induced global warming on the intensity of hurricanes has been investigated in a series of studies (1527, hc, hz). The large scale atmospheric boundary conditions for a high CO2 climate have been either derived from a global climate model (1527, hz) or explored via a systematic examination of a wide parameter space of tropical lapse rate and tropical SST changes (hc). For the global climate model-derived conditions, a modest increase (~5-10%) in the intensity of very strong hurricanes has been simulated in the high CO2 climate, as compared to the strongest hurricanes in the control climate.

          One assumption of these and other studies to date is the neglect of the effect of hurricane/ocean coupling (i.e., the local SST cooling induced by the hurricane) on the intensity changes. To evaluate how a CO2-induced enhancement of hurricane intensity could be altered by the hurricane/ocean coupling, a series of idealized hurricane experiments was performed using the GFDL Hurricane Prediction System coupled to a 1/5 degree resolution regional version of the Princeton Ocean Model (POM). For each of a series of 72-hr experiments, a specified initial storm disturbance was placed in an idealized basic state consisting of a uniform easterly flow of 5 m/sec. The initial SST, atmospheric lapse rate, and atmospheric moisture conditions for the experiments were derived from long simulations (control and high CO2) of a global climate model. The ocean thermal stratifications were based on observed climatologies and included the CO2-induced stratification changes simulated by the global climate model. The results indicate that a CO2-induced intensification still occurs even when the hurricane/ocean coupling effects are included; ocean coupling appears to have only a small impact on the magnitude of this intensification. Analysis of these experiments was completed during the past year (mv).

     7.3.2 Tropical Cyclone-Ocean Interaction

          Unlike the steady decline in operational forecast track errors over the past two decades, intensity prediction has shown little improvement in the forecast period of one to three days. One of the important physical components lacking in both statistically and dynamically based forecast methods is that of the role of the ocean interaction beneath the hurricane. The role of ocean interaction is becoming more accepted as an important component in forecasting tropical cyclone intensity. Therefore, the GFDL group continues to collaborate with the University of Rhode Island Oceanography group in developing a new GFDL coupled hurricane-ocean model for the Atlantic to improve hurricane intensity forecasts. For the 1999 hurricane season, the GFDL/URI forecast system produced coupled model forecast results typically available 2-4 hours after the operational NCEP GFDL product. For the 1999 season, 139 forecasts were performed including the following storms: Arlene, Brett, Cindy, Dennis, Emily, Floyd, Gert, Harvey, Irene, Jose, and Lenny. The coupled model forecasts of track and intensity were made available to the National Hurricane Center via a dedicated web site. In addition, a successful conversion was made from a research model to a fully automated real-time prediction system.

          As in the 1998 season, the coupled model achieved significant improvements in the hurricane intensity forecasts compared to the operational atmosphere-only GFDL model. The mean absolute error of the sea level pressure forecasts was reduced by about 31%. Although the improvements in the minimum sea level pressure forecasts are significant, little improvement was achieved in forecasting surface winds by the coupled model. This is because the GFDL model has a tendency to underestimate the surface winds for strong hurricanes with minimum sea level pressure below 970hPa. Therefore, to gain accuracy in forecasting surface winds, the coupled model used the observed pressure-wind relationship in conjunction with the model predicted minimum sea level pressure to determine surface winds.

          Work was completed in porting the coupled model to a scalable parallel design on the new NCEP IBM-SP with the goal of implementing the model for operational forecasting for the 2000 hurricane season in the Tropical Atlantic. Implementation of an improved ocean and atmospheric initialization procedure is also under way.

     7.3.3 Tropical Cyclone Landfall

          A suite of experiments was designed and run in collaboration with URI to investigate the role of surface water on landfalling hurricanes. This study (mt) resulted in an interesting comparison of land surface conditions, extending earlier work on the impact of surface conditions on hurricane landfall. Another study was initiated with NESDIS to evaluate the rainfall from GFDL landfall forecasts for 1995-1999. Given the flooding of Floyd and Mitch, rainfall is another important parameter to investigate, and the current studies may be the first attempt to evaluate the patterns of model produced rainfall for real cases of landfalling hurricanes. Hurricane Opal (1995) presented a challenging sequence of rapid intensification followed by filling just before landfall. A collaborative study with Colorado State University scientists indicated that trough interaction played a minimal role in Opal's intensification, and that vertical shear caused its subsequent weakening, at least in the GFDL model simulation (nb).

PLANS FY01

          Other methodologies for examining the hurricane intensity/climate change issue will also be explored, such as the use of higher resolution GCMs. The sensitivity of prior results to physical parameterizations will be explored. Work will continue to improve the tropical cyclone-ocean coupled model and initialization. A plan to extensively analyze the storm behavior at landfall will be made using improved model physics.

  7.4 MODEL IMPROVEMENT

ACTIVITIES FY00

          A major activity during the past year was the successful conversion of the GFDL modeling system to a distributed memory system. This ensures the efficiency of the forecast system for operational implementation at NCEP for the 2000 tropical season, as well as for future research applications at GFDL. The conversion effort was quite difficult due to the complex nature of the GFDL moving nested grid framework. The new design incorporates two-dimensional decomposition. For operational use, the present model uses 42 processors to achieve the approximate speed attained by the Cray T90 12 processor run. Also converted were the pre- and post-processing steps of the forecast system, including the vortex bogussing step. The code was redesigned to use an arbitrary number of sub-domains. It was found that this distributed system is highly dependent on the efficient communication between the distributed processors.

          Work on the improvement in the spatial resolution of the hurricane model continues in order to cope with the increase in the global model resolution (6.1.3), as well as to represent the hurricane structure more accurately. The modeling system has been run with a doubling of resolution of 1/2°, 1/6°, and 1/12° in the horizontal and with 42 levels for a selected number of case studies of Floyd (1999), Opal (1995) and Mitch (1998). More physically realistic features have been noted. High resolution GFDL forecasts of Floyd's rainfall were quite encouraging, indicating large amounts at landfall in the Carolinas with a swath of heavy rainfall through the Mid-atlantic and up to New England (Fig. 7.3). More cases need to be evaluated to check for

increased forecast skill. Interestingly, the improvement in intensity error was disappointing, indicating that model physics packages may be deficient.

          A comprehensive evaluation of the physical parameterizations of the GFDL model has been under way, including horizontal and vertical turbulent mixing, dissipative heating, surface exchanges, and convective parameterization. A suite of experiments using Mellor-Yamada level 2.5 mixing was tested for a 74-case suite of Dennis, Floyd, and Gert. Substantial increase in intensity skill was achieved, especially for the first two forecast days with improvements of 17 to 12, 16 to 13 and 15 to 13 knots at 12, 24, and 36h, respectively. Perhaps more impressively, the negative bias (under-forecast) for winds was reduced from ~10 knots to ~4 knots throughout the 72h forecast period. Preliminary results indicate improvements in storm structure when prediction of turbulent eddy kinetic energy is included. The model's forecasts of both track and intensity are also sensitive to different convective parameterizations.

PLANS FY01

          Developmental work to improve the hurricane model initialization, ocean interaction, model physics, and resolution will continue. This is critical for improved intensity prediction skill. A significant number of case studies will be needed to evaluate the impact of these efforts on forecast skill of track and intensity. Additional work will be continued to assimilate more data into the GFDL forecast/analysis system. Physical mechanisms of evaporation of rain and sea spray together with cumulus and dissipative heating will be evaluated.

*Portions of this document contain material that has not yet been formally published and may not be quoted or referenced without explicit permission of the author(s).