Accurate forecast of tropical cyclones/hurricanes (TC) is a high priority topic of research, due to their potential large economic impact as well as public safety issues. Hurricanes rank among the most destructive and costly of natural phenomena. For example, the warm waters particularly along the Bahamas and the Greater Antilles provide a significant energy source for hurricanes en route to Florida. As a result, Florida receives more hurricanes than elsewhere in the United States (Elsner and Kara, 1999). In 1992, hurricane Andrew caused more than $30 billion in direct economic losses in south Florida. The recent hurricanes of 2004 (Charley, Frances, Ivan and Jeanne) that spun toward and crossed Florida, caused over $20 billion in losses. Even more alarming, Pielke and Landsea (1998) reported that total loss estimates for a category-4 hurricane striking Miami, Florida would be more than $60 billion. The most recent hurricane of 2005 (Katrina) caused $100 billion economic cost. This trend in damages highlights the importance of and need for better TC track and particularly intensity forecasting.
Despite large reductions in track forecast errors over the past three decades (cf. Fig. 1), hurricane forecast errors have not reached estimated predictability limits (McAdie and Lawrence, 2000). In contrast to the improvements in track forecasts, there has been little improvement in forecasts of storm intensity and surprisingly little effort to forecast storm structure and overall size (Liu et al., 1997; Camp and Montgomery, 2001). Since, most of the damaging hurricanes undergo rapid intensification prior to landfall, such as hurricane Katrina, (globally about 10-15% of all hurricanes experience a period of rapid intensification), considerable improvement is still needed in the understanding and prediction of TC intensity change and inner core structure.
The TC forecasting has received significant attention both on the scale of weather prediction (~3-5 days forecasting) and on the climate scale. On the climate scale, TC behavior is influenced by climate factors, such as changes in large scale
J.B. Elsner and T.H. Jagger (eds.), Hurricanes and Climate Change, 339
doi: 10.1007/978-0-387-09410-6, © Springer Science + Business Media, LLC 2009
Fig. 1 Shown are the analysis and 24, 48, and 72 hour National Hurricane Center forecast track errors from 1975 to 1998 (after McAdie and Lawrence, 2000). The linear regression trend lines show that in this 23 year period the 48 and 72 hour forecast track error improved by 50%, while the 24 hour forecast track error improved by 37%. In spite of these improvements, 72 hour forecast track error is still about 400 km, the 48 hour forecast track error is about 275 km, while the 24 hour track error is about 175 km
Fig. 1 Shown are the analysis and 24, 48, and 72 hour National Hurricane Center forecast track errors from 1975 to 1998 (after McAdie and Lawrence, 2000). The linear regression trend lines show that in this 23 year period the 48 and 72 hour forecast track error improved by 50%, while the 24 hour forecast track error improved by 37%. In spite of these improvements, 72 hour forecast track error is still about 400 km, the 48 hour forecast track error is about 275 km, while the 24 hour track error is about 175 km circulation anomalies associated with the El Niniio-Southern Oscillation (ENSO) or the North Atlantic Oscillation (NAO) or by shifts in tropical storm origins or regional changes in land use (e.g., Elsner et al., 1998). On the scale of weather prediction, TC behavior is characterized by strong multi-scale interactions: the hurricane vortex is hundreds of km in horizontal size (synoptic-scale), the eye is tens of km (mesoscale), and the embedded convective clouds are in the order of km (cloud-scale), with a vertical scale of up to 20 km (e.g., Gopal et al., 2002). Moreover, other atmospheric phenomena, such as dust aerosols, may affect the development of TC's (Dunion and Velden, 2004). Such issues are still under investigation and are somewhat controversial.
The local environmental flow in which the hurricane vortex is embedded is the main factor that determines its short-term track. At longer timescales, an accurate forecast of the flow at considerable distance from the cyclone is needed. The internal structure of the TC and interaction between the structure and the hurricane's environment are the primary influences on hurricane intensity. These spacetime scales and scale interactions should be represented as accurately as possible, when TC behavior is studied. Otherwise, significant components of the problem may be neglected.
This wide range of temporal and spatial scales characterizing atmospheric processes demonstrates the need for multiscale capabilities in hurricane modeling. However, the current atmospheric simulation systems are scale specific and cannot resolve the full spectrum required for the accurate forecast of local scale phenomena (e.g., convection). Even with recent advances in computational power, the current architecture and physics of today's generation of atmospheric models cannot simulate the full spectrum of scale interactions of the atmosphere. Nevertheless, numerical weather prediction (NWP) models should be capable of predicting three dimensional, time dependent atmospheric mean flow and turbulence fields over complex terrain in an unsteady synoptic environment. In addition, these models should have a sufficient grid resolution to account for local scale phenomena and mean vertical planetary boundary layer structure. The numerical accuracy of the algorithms and consequently the eventual predictions of these models are closely related to the completeness of model physics, dynamics, accurate initial and boundary conditions, and the computational domain and mesh on which the model calculations are performed.
As to the completeness of model physics and dynamics, over the last two decades, with the advancements in supercomputers and improvements in the formulation of atmospheric physics and dynamics, NWP models have been continuously improved. However, analysis errors (i.e., errors in initial conditions) are still being considered as one of the main source of the TC forecast errors for the NWP models (Lorenz, 1990). As for the atmospheric analyses, their quality depends on two main factors. First, the quality of the analysis is limited by the technique through which the analysis is derived. Both the NWP model used to generate first guess fields for the analysis, and the statistical estimation method that combines information from the first guess and observations are built using approximations because of limits in our knowledge and computational capabilities. The second dominant factor regarding the quality of the analysis is data coverage. Observations with better geographical coverage, higher quality and/or more comprehensive data should lead to improved atmospheric initial conditions and in turn improved NWP forecast, provided that these observations can be assimilated rapidly enough to be useful.
The current in-situ atmospheric measurement system is inadequate to determine the current state of the atmosphere and hence our ability to forecast is limited by this constraint. These observations at regular time intervals and at fixed geographical locations have been able to define the large-scale features of the atmospheric system and have served the interests of climatologists, synoptic forecasters, and early NWP forecast systems. However, there is always need for additional data to resolve critical features in the atmosphere (such as the details of convective structure of hurricanes). The current and next generation of satellite-based atmospheric observations is likely to provide a significant increase in our observational coverage. The spatial and temporal resolution of the retrieved atmospheric properties will expand and the net result on the data flow is expected to increase. This paradigm shift may require a shift in our utilization of data. For example, instead of using all data in an egalitarian and brute-force method, we may need to consider adding some form of intelligence to our data processing system to identify, extract, and utilize just the necessary critical information for a given forecast cycle (Boybeyi et al., 2007).
Undoubtedly, another obvious method to improve the accuracy of the prediction of NWP models is to enhance the spatial grid resolution. However, introducing fine spatial resolution throughout the domain is not always practical since the size of the modeling domain, the numerous interactions between the various atmospheric processes, and the complexity of the numerical algorithms places restrictions on the grid resolution that can be achieved using current computers. These limitations prohibit the use of a uniform high spatial resolution that is appropriate to resolve the smallest scales of interest (in case of TC forecasting, the smallest scale of interest is "convection"). One of the alternatives is then to develop methodologies capable of providing local refinement in certain key regions.
Hence, TC track and intensity forecast can be improved by;
• Detailed measurements on scales from the storm's large scale environment to its small inner-core and
• Appropriate numerical grid resolution to resolve the smallest scales of interest (e.g., eye, eye wall, and spiral bands) and multi-scale atmospheric flow interactions.
This chapter discusses the above requirements in details for a better TC forecasting. Section 2 provides discussion on remotely sensed observations, while Section 3 provides discussion on high fidelity numerical modeling coupled with high resolution observations. Finally, Section 4 provides some concluding remarks.
Was this article helpful?