WMO/CAS/WWW

¹E-mails: schen@rsmas.miami.edu

²Email: Naomi.Surgi@noaa.gov

Tropical cyclone (TC) track forecasts have continued to improve over the last several years. However, the intensity forecasts are still lagging behind. This is clearly evident in the most recent numerical model predictions of Hurricane Lili (2002). None of the current numerical model guidance captured the rapid intensification or the rapid decay before the landfall of Hurricane Lili in the Gulf of Mexico. Both the research and operational communities are focusing on improving the model performance for TC structure and intensity changes. Here we will focus on progress made in both operational and research models over the last four years since the last IWTC-IV. Key issues related to model resolutions and physical representations are discussed. This working group addresses both the operational and research modeling perspectives for advancing TC intensity and outer wind structure forecasts.

1.5.1 Introduction

Numerical weather prediction (NWP) models have increasingly played a vital role in advancing the operational forecasting skill of tropical cyclones (TCs). Over the past three decades, the average official 72-h TC track forecast error has been reduced by over 50%, which can be largely attributed to major advances in operational NWP systems. In the last fifteen years, this skill has accelerated with the introduction of global models in the mid-1980s and with further upgrades and refinements of global modeling systems during the 1990s. Another important contribution from the research community has been the development of high-resolution regional dynamic models with advanced physics.

This advancement of dynamical modeling systems has not had the same positive impact for improving intensity forecasts, which still lag behind the statistical intensity forecast models. Although current statistical models make use of the skill of global models in forecasting important input features that influence intensity changes such as vertical shear, they only provide rudimentary information on storm size and intensity and lack a comprehensive detailed description of changes in storm structure that are crucial to provide more quantitative information to meet operational demands for TC watches and warnings.
Over the last 4-5 years, both idealized simple dynamic models and high-resolution, full physics numerical models have been used to improve our understanding of physical processes that control the structure and evolution of TCs. To resolve the TC inner core structure, which is crucial in intensity forecasting, some research models use horizontal grid resolutions of 1-2 km. At this resolution, we are not only able to simulate the inner-core dynamics, but also explicitly resolve the convective processes that have previously been represented by cumulus parameterizations at lower resolutions. Recent studies have also indicated that the quantitative precipitation forecasting (QPF) in TCs has been improved, especially for the concentrated high rainrate near the inner core. These progresses are important for development of the next generation of operational forecasting models. The plan has been made to adept the high-resolution Weather Research and Forecasting (WRF) Model for prediction. Some of the development work has been underway over the last year.

The ultimate goal of the NWP models should be consistent skillful forecasts of TC track and intensity change over all TC basins. Although statistical models will continue to be used at operational forecast centers for guidance on TC intensity forecasts, future work should be focused toward developing more advanced high-resolution NWP systems. These systems must include advanced initialization of the TC vortex, high-resolution physics, and coupling to the ocean and land surface that can address the details of the TC vortex structure and the interaction of the TC with the environment.


1.5.2 Operational Modeling Efforts

The operational TC forecast community has identified improving intensity forecasts and TC- related rainfall forecasts and associated inland flooding as high priority forecast issues. Although global and relatively higher resolution regional models have advanced track forecast skill, the above forecast challenges will require very high resolution NWP coupled systems with advanced high-resolution physics. However, continued improvements of the global model systems are required to provide a realistic TC environment that includes both the atmosphere and the ocean.

a) Model physics

The sensitivity of TC structure to model physics is now well known from the high-resolution TC studies by the research community over the past 4-5 years (see next section). Even with a coarse resolution global model, this sensitivity was documented in a study by Surgi et al. (1998) that compared upgraded physics in the NCEP global model verses more rudimentary treatment of the boundary layer and deep convection. These physics play a crucial role in maintaining the TC structure through the vertical distribution of heating and the important feedback between the boundary layer and the deep convection to support the vertical coupling in the TC vortex. It was also found that enhanced model vertical resolution is also important to provide a detailed description of the heating profile that is crucial to properly represent the vortex structure.
.
In anticipation of upgrading the Geophysical Fluid Dynamics Laboratory (GFDL) model physics for benchmarking the GFDL prior to its replacement by the next generation NOAA WRF, the GFDL model is currently being tested with various new physics for deep convection and boundary layer. Fig. 1.5.1 shows the 36-h forecast for Hurricane Isidore from the operational GFDL model with a moist convective adjustment scheme and Mellor-Yamada 2.5 boundary layer. Fig. 1.5.2 shows the 36-h forecast of the GFDL model with physics of the NCEP Global Forecast System (GFS).

Note that the GFS physics provide a much more realistic overall TC structure including the spreading out of the upper-level heating along the TC anvil. Also, the impact of the improved TC structure feeds back to improve the TC track forecast, whereas the operational model accelerates Isidore northwards in the Gulf of Mexico with landfall along the Gulf Coast states in 72 h. The upgraded GFDL model slows the forecast in the central Gulf, which is more consistent with reality. The feedback of improved TC structure in improving track is also consistent with the results of Surgi et al. (1998).




Fig. 1.5.1 Cross-section of Hurricane Isidore with the old GFDL physics (i.e., Kurihara cumulus parameterization, Mellor-Yamada 2.5 closure on PBL). The shading is the total heating, the contours are the wind speed.



Fig. 1.5.2 Cross-section of Hurricane Isidore with the upgraded GFDL physics from the NCEP global model (i.e., SAS+MRF PBL).

Work on developing an appropriate microphysics package is also progressing at NCEP and NRL (Naval Research Laboratory) in Monterey, California. The NRL has adapted a microphysics package (Rutledge and Hobbs, 1983) that includes two ice components (cloud ice and snow). This microphysics option produces less and delayed precipitation that generally results in less intense TCs. More recently, NRL has an upgraded version of the microphysics parameterization that includes grauple as the third ice component, which improves the conversion calculation with consideration of mass conservation. This version is currently being evaluated for impact on intensity forecasts.

Many uncertainties remain in developing microphysics, e.g., how to partition the liquid verse ice and how many species of ice need to be represented? Previous studies have shown that different ice species have different terminal velocities that can greatly affect the downdrafts, consequently, the TC’s intensity. Much testing needs to be carried out at different model resolutions to provide more insight into this problem.

Timeliness requirement and computing resource constraints will prevent some operational centers from running TC models at grid resolutions less than 6 km. How to represent unresolved moist convection at these scales is a very important and timely topic.

b) Air-sea and air-land coupling

The effect of tropical cyclone-ocean interactions on the intensity of observed hurricanes was investigated by Bender and Ginis (2000) by coupling the GFDL hurricane model (Kurihara et al. 1998) with a high-resolution version of the Princeton Ocean Model. Bender and Ginis (2000) tested the ocean coupling for two cases of Hurricane Opal (1995) and one case of Hurricane Gilbert (1988) in the Gulf of Mexico and two cases of Hurricanes Felix (1995) and Fran (1996) in the western North Atlantic. It was shown that the coupling produced a significant cooling of the sea surface resulted in a substantial decrease in the evaporation and boundary layer moist static energy over the cold wake. A substantial decrease in SST, when the storms were moving slowly, had a large impact on the storm intensity. In each of the seven forecasts, inclusion of the ocean coupling generally led to improvements in the prediction of storm intensity, as measured by the minimum sea-level pressure (SLP), relative to the operational (uncoupled) GFDL model.

During the 1998-2000 hurricane seasons, the coupled GFDL/University of Rhode Island hurricane-ocean model described in Bender and Ginis (2000) was run in near-real time for more than 350 cases in the Atlantic and Gulf of Mexico. Improved intensity forecasts were again achieved, with the mean absolute error in the forecast of minimum SLP reduced by about 26% compared to the operational (uncoupled) GFDL model. The largest improvements (30.5 %) were achieved for intense hurricanes (less than 980 hPa) (Ginis et al. 1999, 2000).



Fig. 1.5.3 Time series of minimum SLP for the uncoupled GFDL forecast (solid line) and the coupled GFDL/URI model (dot-dashed line) compared to observed values (circles with values every 6 h) for the forecasts of Hurricane Bonnie (from Ginis et al. 2000).

The predicted 72-h time series of the minimum SLP for Hurricanes Bonnie are shown in Fig. 1.5.3. This example represents one of the best cases and is obtained during important forecast periods. On August 23, 1998, Hurricane Bonnie was moving toward the U.S. east coast. The GFDL model track forecasts were generally fairly good as the model correctly predicted that the storm's forward speed would be reduced during that period. However, the intensity forecast indicated rapid intensification, which resulted in large intensity overprediction. This overprediction occurred because the operational model had responded to very warm water underneath the storm in the global SST analysis. In the coupled model, the slow hurricane movement generated a large SST decrease below the storm. As a result, the storm intensity forecasts were significantly improved. At initial time 1200 UTC August 23 shown in Fig. 1.5.3, the coupled model correctly predicted that the hurricane would weaken by 1200 UTC August 24.

The coupling of the GFDL model to a land surface model (LSM) is an important effort to improve TC related QPF forecasts for landfalling TCs to address the inland flooding problem. Early land surface models arbitrarily cut off evaporation at landfall and increased the surface roughness. The GFDL model currently uses a bulk representation of the land surface in which potential evaporation is reduced via a cooled land surface.

A comprehensive land surface model was collaboratively developed by NCEP, the University of Oregon, the U.S. Air Force, and the NOAA/National Weather Service Office of Hydrology. Various configurations of this model are currently under testing at NCEP to account for changes in the surface fluxes as storms make landfall over various underlying surfaces. The land model includes a multi-layer effect of the direct effect/interaction of vegetation, plant interaction with evaporation, prediction of soil moisture and temperature, and canopy water content. These processes are important for storms making landfall over regions that include standing water. Fig. 1.5.4 is a schematic of the land physical processes affecting TC structural changes upon landfall that are not captured by traditional boundary layer treatments. Figs. 1.5.5a and b from Tuleya et al. (????) show substantial changes from the surface fluxes, e.g., decrease in the evaporative and sensible heat fluxes in the underlying surface at landfall. In that study, sensitivity on the 48-h intensity forecast for the landfall of Hurricane Helene was shown by comparing the TC translating over a nearly dry surface verse a saturated surface.





















Fig. 1.5.4 Schematic of physical processes for hurricanes at landfall.

































Fig. 1.5.5 GFDL model-simulated surface latent (a) and sensible (b) heat fluxes.
c) Model initialization

One of the biggest scientific challenges for the next-generation, high-resolution hurricane prediction systems is addressing the hurricane vortex initialization problem. In anticipation of running the next-generation high resolution coupled air/sea/land model system that will replace the GFDL model, e.g. the community-based Weather and Research Forecasting model for hurricanes (HWRF), NCEP is developing a local mesoscale data assimilation capability for initializing the hurricane core circulation. This capability is being developed to make use of real time radial velocities from the NOAA Gulfstream-IV doppler winds to describe the 3-D wind structure of the hurricane core circulation from the outflow layer to the surface. This assimilation system is being developed in concert with the NCEP’s current 3-DVar to make use of future advanced mesoscale variational data assimilation techniques being developed at NCEP. We believe this capability is critical to address the intensity problem by resolving the outflow layer of the hurricane core plus having detailed vertical resolution to the surface to properly simulate the boundary layer fluxes and resolve the vortex asymmetries.

d) Environmental effects

An accurate simulation of the large-scale flow is critical to capture the interaction of the TC with the environment to improve TC structure and intensity change forecasts. In this regard, global model improvements have provided a much improved description of the tropical environment with the advancement of the global modeling systems over the past 4-5 years. These improvements have resulted from including more observations in the environment, advanced data assimilation techniques, and continuing numerical model development. The data assimilation upgrades over the past several years have contributed to better defining the TC environment via the direct assimilation of satellite radiance data, e.g. microwave data, and a physical initialization that makes use of the assimilation of the SSM/I and TRMM precipitation data. NCEP has also begun assimilating the QuikScat data, which to date has showed some improvement near the surface in the tropics. However, optimal use of the scatterometer data will require substantial development of data quality control and better algorithms for the boundary layer schemes. Some global models have also included the use of prognostic cloud water/ice with related radiative effects. All of these advancements have contributed to provide a better description of the large-scale tropical diabatic heat sources and thus have resulted in an improved tropical circulation.

e) Model configuration

A recent improvement to the Coupled Ocean Atmosphere Prediction System (COAMPS^TM) for the high-resolution TC forecast is to apply the movable inner mesh capability to follow selected tropical cyclones. For an operational forecast, the application requires a method of dynamically tracking TCs in the model forecast. The TC tracking method used in the GFDL hurricane model is adopted to follow the storm center on the COAMPS^TM inner meshes every few time steps. If a TC moves away from the mesh center more than one grid distance of its mother mesh, the mesh is moved to keep the TC always near the center of the inner mesh so that its circulation can be better resolved by the high-resolution mesh during the entire forecast period. Fig. 1.5.6 shows an example of such application in which the detailed interaction between the TC Bilis and the Taiwan terrain is better resolved by the 9-km inner mesh following the cyclone.

Recently a series of high-resolution model simulations of tropical cyclones has been conducted using fully nonlinear, nonhydrostatic, primitive equation mesoscale models, such as the PSU/NCAR MM5. These simulations have demonstrated the ability of the high-resolution models to re-produce the fine structures in the inner core and outer rainbands region of tropical cyclones (e.g., Liu et al., 1997; 1999; Karyampudi et al., 1998; Braun and Tao 2000, Braun 2002; Bao et al., 2000, Rogers et al., 2002, Chen et al., 2002, Tenerelli and Chen 2002, Zhang et al. 2002). Tenerelli and Chen (2000) developed a vortex-following nested-grid using the nonhydrostatic MM5 that allows long integration (5-7 days) with very high grid resolution (~1-2 km) in the inner core region of hurricanes. They use four-level nests with 45, 15, 5, and 1.67 km grid spacing, respectively. The three inner domains move automatically with the storm. The model has been used successfully to simulate Hurricane Bonnie 1998 (Rogers et al. 2002), Hurricane Georges 1998 (Orndorff et al. 2002), and Hurricane Floyd (Chen et al. 2002, Tenerelli and Chen 2001, 2002). Fig. 1.5.7 shows an example of model simulated precipitation fields at different grid resolutions. In comparison with the TRMM precipitation radar (PR) observed rainfall, the model captures the intense rainfall in the inner core very well with 1.67 km resolution, whereas the lower resolutions unable to resolve the inner core and rainband structures. For the first time a high-resolution model is able to capture the development of the concentric eyewalls and a complete eyewall replacement cycle that was observed in Hurricane Floyd. A 6-day long animation of MM5 simulation of Floyd precipitation field can be found at http://orca.rsmas.miami.edu/floyd. Tenerelli and Chen (2002) show that the secondary rain maximum and associated a ring of potential vorticity cores lead the development of the secondary wind maximum. Braun (2002) has also conducted a very high-resolution (1.3 km) MM5 simulation of Hurricane Bob (1991) and showed clearly that the model captures the detailed dynamic and thermodynamic structure in the inner core region. These results demonstrate the capability of the high-resolution model in simulating tropical cyclone structure and intensity changes.

TCs interact with the large-scale atmospheric environment. Interactions with vertical wind shear are an important part of current statistically-based intensity forecast models and have been the subject of considerable research (e.g., DeMaria 1996; Galina and Velden 2002). Frank and Ritchie (2000) and Rogers et al. (2002) show the impacts of vertical shear on both inner-core structure and rainfall patterns of TCs are significant. Current operational models give reasonably good predictions of storm environmental vertical wind shear. In addition, real-time products are now available that give maps of vertical wind shear over many ocean basins using satellite observations such as scatterometer and cloud-drift winds (e.g., http://cimss.ssec.wisc.edu/tropic/tropic.html). The impact of these products on experimental forecasts using a dynamical tropical cyclone model has been assessed and found to be useful (Soden et al. 2001). Clearly, operational assimilation of such products in a dynamical TC forecast model will occur in the future.

Session 1.5.
< P>