High-Resolution Modeling of Tropical Cyclones Using Moving Grids



C.-S. Liou and T.R. Holt
Marine Meteorology Division

Introduction: Tropical cyclones or hurricanes are serious threats to both Navy operations and the general public because of the potential for severe and hazardous weather conditions. Although the accuracy in tropical cyclone track forecasting has been significantly improved in recent years, uncomfortably large uncertainty still exists in predicting tropical cyclone structure and intensity. Besides the lack of a sufficient number of observations for properly describing the initial tropical cyclone circulation, complexity in cyclone structure represents another difficulty in predicting tropical cyclones with numerical models. As shown in the AVHRR satellite image for super hurricane Floyd (Fig. 1), vigorous deep convection appears adjacent to the calm hurricane center, the eye. Wind speed increases rapidly from near zero in the eye to higher than 50 m/s within 50 km. Beyond the deep convective eye wall, small-scale convective cells and asymmetric, spiral convective bands dominate the hurricane circulation. This sharp gradient wind distribution and the small-scale nature of deep convection require very high model grid resolution to properly simulate the tropical cyclone structure and intensity. However, a hurricane typically moves with a speed of 25 to 30 km/h. Therefore, to maintain a high-resolution grid centered on the tropical cyclone and to make the most efficient use of computer resources, numerical techniques must be developed for moving high-resolution grids to follow a selected tropical cyclone. We have developed and implemented such numerical techniques in the Navy's operational mesoscale model, the Massive Parallel Processing (MPI) version of the Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS™).*

Fig 1





FIGURE 1
AVHRR satellite image of super hurricane Floyd at 19:00 GMT 12 September 1999 (courtesty of CIMSS, University of Wisconsin).

Moving Grids for Tropical Cyclone Prediction: We need to dynamically and automatically locate the position of a tropical cyclone in a high-resolution grid domain in order to move the grids following the selected cyclone. To ensure a smooth cyclone track that follows and represents an area center of cyclone circulation, we use a "mass center" to define a tropical cyclone position. The mass center is defined as the gravity center of pressure deficits with respect to a reference pressure. When a COAMPS¨ high-resolution inner grid is assigned to follow a tropical cyclone, the inner grid is initially centered at the tropical cyclone position. The cyclone position is then dynamically tracked at every time step of the forecast. When the tropical cyclone moves away from the inner grid center more than one grid distance of its parent grid, the inner grid is moved to its new location where the grid is once again centered over the tropical cyclone. The inner grid remains fixed when the tropical cyclone movement would shift the grid into the lateral boundary zone of its parent grid.

As a high-resolution inner grid follows a selected tropical cyclone, the grid will move into an area where only coarser-resolution information is available from its parent grid. We use bi-linear interpolation to project the coarse-resolution information onto the high-resolution grids in that area. A dynamic consistence adjustment is applied afterward to reduce any imbalance generated by the interpolation. For terrain height and land-sea-ice index fields, which are time independent, we prepare the two fields initially with high resolution covering the whole domain of the outer-most coarse grid and ensure the terrain height and index to be the same at collocated grid points of all grids. In this way, the high-resolution fields are easily extracted for any high-resolution moving grid, regardless of where it moves.

Results: We choose tropical cyclone Bilis to demonstrate the ability of moving high-resolution grids in modeling the detailed structure of a tropical cyclone. Figure 2 shows the 850 hPa winds (about 1.5 km above ground) of a 27-km resolution parent grid at the initial time, with white boxes indicating locations of the moving high-resolution inner grids at different forecast times. Figure 3 shows the 850 hPa winds of the 9-km resolution moving grid for the 42-h forecast. The high resolution simulates well the complicated wind distribution when the tropical cyclone interacts with the complex terrain of Taiwan. By ensuring the same terrain height at all collocated grid points, the high-resolution grid smoothly moves over the complex terrain areas without generating any obvious numerical noise. Comparisons of COAMPS™ forecasts from moving grids and larger-area fixed grids also indicate that the grid movement introduces very minimal errors in numerical calculation (not shown). More than 100 tests have been conducted to ensure the accuracy of the numerical techniques implemented in the MPI computational environment.

*COAMPS™ is a trademark of the Naval Research Laboratory.

Fig 2

FIGURE 2
850 hPa wind speed (color, m/s) and streamlines of 27-km grid at the forecast initial time. White boxes are locations of the 9-km moving grid at different forecast times.

Fig 3

FIGURE 3
850 hPa wind speed (color, m/s) and streamlines of 9-km moving grid for a 42-h forecast.

Summary: To improve tropical cyclone structure and intensity forecasts by better resolving its complicated circulation, we have developed and implemented algorithms to move high-resolution COAMPS™ grids following a selected tropical cyclone. The success of this development provides us with a capability to run very high-resolution inner grids with affordable computer resources to further develop other modeling components, such as model physics, for improving tropical cyclone structure and intensity forecasts.

[Sponsored by ONR and SPAWAR]