PRESSING FORECAST AND WARNING CHALLENGES

3.1 Introduction. The ultimate goal of the operational forecast centers is the protection of life and property from tropical cyclones. The centers are keenly aware of operational and research deficiencies that prevent them from fully meeting this goal. Accordingly, a list of research objectives related to the most pressing forecast challenges, as perceived by TPC, CPHC, and JTWC, was compiled and is shown in Table 3.1. Because the missions of the three centers vary, their priority assessments also vary.

Since a similar assessment was first prepared in 1992, several changes have occurred that have influenced the needs of the warning centers and the priorities of those needs. Advances in technology; downsizing of the military; budgetary considerations; the cost of wind damage from Hurricane Andrew, Hurricane Iniki, and Typhoon Omar; the acquisition of a high altitude jet aircraft, and changes in the NWS organizational structure have all had some influence on determining the pressing forecast and warning needs.

The research objectives in Table 3.1 are divided into ten categories. Each of these categories will be discussed individually in sections 3.2 through 3.11.

The most important, most visible, and most specific user products issued by the centers are the tropical cyclone forecasts/advisories (formerly designated as marine or military advisories). Regardless of the forecaster's knowledge of tropical cyclone initial position and intensity, or his/her confidence in the projections through 72 hours (48 hours in the Southern Hemisphere), these advisories require exact specification of initial and/or forecast tropical cyclone position, intensity (including maximum sustained winds, gusts, radii of various wind speeds), eye diameter, radius of 12-foot seas, and coastal effects if the storm is to make landfall. Actions of virtually the entire forecast and warning system are based on these forecasts/advisories. Therefore, it is not surprising that the specific items required in the forecast/advisory issuance are typically rated as having a high priority research need by each center.

Although the items listed in Table 3.1 are listed as separate entities, they are not independent. For example, Carr and Elsberry (1994) discuss the relationships between motion and intensity from both a user impact and dynamical perspective. With regard to user impact, the authors state:

"Traditionally, characterizations of the accuracy of tropical cyclone forecasts have focussed primarily on track forecast error. However, error in forecasting the evolution of a significant wind area is a complex combination of the track, intensity, and size forecast errors. For example, even if the intensity and size of a midget typhoon were to be precisely forecast, a relatively small cross-track forecast error might lead to an extreme overestimate of surface winds for a location near the forecast track. Conversely, an accurate intensity and track forecast accompanied by a size forecast that fails to account for substantial growth of the extent of gale-force winds will lead to a large underestimate of wind speed at a location forecast to be on the periphery of the tropical cyclone. The key point is that the track, intensity, and size forecasts are inextricably linked from the perspective of impact on the user."

The authors further demonstrate that the motion, intensity, and size are also related in a theoretical sense.

Table 3.1. Consolidated tropical cyclone research objective priority rating for warning centers. A single/double asterisk indicates the priority has been upgraded/downgraded since the last assessment in 1992. A triple asterisk indicates a new item.

3.1.1 Research and Development Categorization. Where a noted deficiency arises from a lack of understanding, a program of basic research is needed. Exploratory research is intended to convert basic research advancements at universities or research laboratories to prediction methods that will address forecast problems. Applied research efforts typically are applications to specific problems at the forecast center and are expected to be implemented if validated in a fully operational mode. Advanced development normally addresses improvements to existing hardware systems.

3.2 Position and Motion. Before further discussion of this topic, the exact meaning of the term "initial position and motion" needs clarification. The precise center of circulation (usually taken as the geometric center of the eye) of a tropical cyclone often has small-scale (less than 40 km) trochoidal oscillations ("wobble") about a reasonably conservative mean track that is more representative of the synoptic-scale steering forces. These small-scale track meanders, illustrations of which are given in WMO (1979) and Holland (1991), can be quite troublesome and misleading to the forecaster.

In addition to uncertainties introduced by trochoidal motion, the forecaster must also deal with the inherent subjectivity of satellite image interpretation and possible navigation problems. Aircraft "fixes" are also not immune to estimation errors (Gray et al. 1991). Whether a sudden turn or acceleration in a tropical cyclone track is real, an artifact of the measurement system, or a temporary internal wobble is very difficult to determine in an operational environment. The important point is that improvements in the ability to "fix" a tropical cyclone may not always translate into better, more representative tropical cyclone initial motion vectors. Indeed, more accurate "fixes," without the ability to interpret the position, may compound the problem of determining the more representative synoptic-scale motion.

The importance of using best-track or synoptic-scale initial position and motion on the short-range forecast is clearly indicated in Table 3.2. Here, the Atlantic CLIPER (Neumann 1972) model was activated in both an operational and a best-track mode. In the operational mode, the model is supplied with current and past position, motion (and intensity) based on the often conflicting information available at warning time. In the best-track mode, the model is reactivated on "perfect" data, compiled after all information on the storm is received and assessed. Although the improvements in the forecast tropical cyclone position are only modest for extended projections, they are substantial at short range. Obviously, this amount of improvement can never be fully realized in operations since the best track implies a knowledge of the storm after the fact. Nevertheless, the table does point out that even a small improvement in initial position and motion has a dramatic effect on the improvement of the short-range position forecast. Further experiments with the CLIPER model show that the initial motion vector is much more important than the initial position in determining CLIPER errors and errors in models that are based on CLIPER output. Other examples of the detrimental effect of incorrect initial tropical cyclone positioning on model predictions are given by Neumann (1975).

3.2.1 Positioning

3.2.1.1 Initial Positioning Error. Initial positioning error (IPE) is defined as the distance between an estimated warning time position and the actual position as later determined from the best track. It is also referred to as the zero-hour forecast. IPE often contains what is sometimes called a "pseudo-component" because some of the small-scale motions (wobble) of the tropical cyclone eye, as discussed above, are subjectively removed when constructing a best track. Thus, the best track is a smoothed representation of the actual track of the tropical cyclone center.

Table 3.2. Comparison of 1972-1987 Atlantic basin forecast errors nm and (km) using the TPC CLIPER model in both an operational (OPNL) and best-track (BTRK) mode (from Gray et al. 1991).

3.2.1.2 Magnitude of Initial Positioning Error. Over the 10-year period 1984-1993, IPE over the Atlantic, eastern North Pacific, and western North Pacific basins have averaged near 16, 15, and 21 nm, respectively. The lower errors over the eastern Pacific basin are consistent with the climatologically persistent tracks for that basin, where simple extrapolation works fairly well in estimating a warning time position, and also reflect the earlier practice of using a single satellite platform for determining position, which provides an "artificial" consistency. Differences between the Atlantic and western Pacific are harder to explain but may be somewhat fictitious because of differences in the methods of computing the best track and treatment of the "pseudo-error" mentioned above. Nevertheless, slightly lower IPE over the Atlantic is consistent with frequent aircraft reconnaissance fixes over that basin, which give a better estimate of the center when satellite imagery is not well defined, as is often the case in the weaker tropical cyclones. Only about 25 percent of the total tropical cyclone positions are based on satellite imagery of eyes in the visible spectrum (OFCM 1988).

Positioning errors have been decreasing over the years--a trend which can be attributable mostly to the availability of satellite imagery in the middle to late 1960's. Atlantic errors have decreased from near 28 nm in 1967 (OFCM 1982) to 16 nm today. However, the rate of decrease in IPE has been leveling off and has only decreased a few nm since 1980. Further improvements in IPE can be attained by improving the position estimates for weaker systems. Additional efforts are needed to improve satellite fix technologies, especially for the weaker systems, and NOAA/NESDIS and the University of Wisconsin, through their various satellite studies, are working this problem. AFGWC also participates through its Special Sensor Microwave/Imager (SSM/I)-related studies. The SSM/I 85 GHz channel, in particular, may provide the resolution needed for better positioning; however, current gaps in the swath coverage limit its applications. HRD/AOML, through its research aircraft missions, has developed various algorithms or methods for identifying the tropical cyclone center using Doppler radar; recent dual-Doppler radar missions have provided extraordinary detail. These algorithms are needed for input into other algorithms that further process these data, such as an algorithm to filter out transitory motions of the eye, which might lead to further reductions in IPE.

3.2.2 Initial Motion. As pointed out in section 3.2, short-period track forecasting is critically dependent on knowing the representative motion of the larger scale storm envelope, as opposed to small-scale transient motions of the geometric storm center. Development of an improved technique that would account for the noise inherent in the raw fixes would be an important contribution to the initial motion problem. Past research in this area has included the Kalman filter approach discussed by Titus and Jarrell (1985), the direct tracking of the entire storm envelope rather than the eye (Sheets 1985), and the use of the CLIPER model (Curry et al. 1987) to aid in obtaining a conservative warning-time position.

3.2.3 Track Forecasts. The track of the tropical cyclone is probably the most important element of the tropical cyclone forecast/advisory. The track is also an important component in a correct intensity and size forecast because, as pointed out by Carr and Elsberry (1994), these forecast parameters are interrelated. Objective track prediction models have been a prime force in the reduction of tropical cyclone forecast errors. The models are typically not used directly, but they do influence the forecaster's decision-making process. Research in this area has been conducted by a number of agencies over a number of years. Since recent studies indicate that forecast errors have been decreasing and have a potential for further decline, additional research is needed mainly in statistical-dynamical and purely dynamical modeling. However, there is also a need for improvements in some of the more basic statistical models. In addition, barotropic models have shown promise in the tropics. Because of their simplicity, these models have utility in the forecast centers for performing sensitivity tests (e.g., to evaluate quickly whether or not the uncertainty of an initial position error would lead to parallel (little impact) or highly diverging (great impact) forecasts).

Track forecasts are currently issued through 72 hours, but a need is developing for still longer range projections. Because of the large errors associated with longer range predictions, the centers have always been apprehensive about users taking specific actions based on forecasts beyond 48 hours. Currently, JTWC (and TPC) are internally evaluating their ability to issue outlooks for 5 days (120 hours).

The NWS Hurricane Probability Program (Sheets 1984) was inaugurated in 1983 to address the forecast uncertainty issue. However, the forecast tropical cyclone track is still taken too literally and additional exploratory and applied research are needed to develop methods for improving user interpretation of tropical cyclone forecast products. With current advances in computer presentation capabilities, this research could include a reassessment of the hurricane strike probability concepts developed for the western North Pacific (Jarrell 1982), or the issuance of a forecast cone similar to current Japanese forecasts. However, probability forecasting requires a considerable amount of user indoctrination in the proper interpretation of often small strike probabilities for a given site.

3.2.3.1 Improvement in Track Forecasts. As discussed in Chapter 2, tropical cyclone motion forecasts have been gradually improving, but there is a need and a potential for additional improvements. Running a statistical-dynamical model and a numerical model on "perfect" initial data (see Chapter 5) suggests that the current forecast errors could be reduced by a factor of two. However, basic research is still needed to estimate the limits of predictability insofar as tropical cyclones are concerned. There is also a need to determine the rate of forecast improvement in some of the other basins, perhaps using the concept of forecast difficulty (Neumann 1981).

Basic research is still needed on the relationship between tropical cyclone motion and the environment. This would include the deep-layer mean concept (see section 3.2.3.4). Experiments such as TCM-90 (Elsberry and Abbey 1991) are enabling tropical cyclone motion to be synthesized into its various components.

3.2.3.2 Binary Tropical Cyclone Interactions. On the average, three times per season, two western North Pacific tropical cyclones become separated by distances less than 1500 km and potentially interact in a binary fashion (WMO 1993). Although this can happen in other basins, binary tropical cyclones over the western North Pacific basin typically occur in the monsoon trough where large-scale steering forces are weak and the binary interaction dominates (Dong and Neumann 1983). As pointed out by Carr et al. (1995), true binary tropical cyclone interaction occurs at rather small separation distances; the occurrence of these is rather rare. These tropical cyclone interactions complicate track forecasting.

Based on the work of Brand (1970) and several other investigators, Steven Hallin at JTWC developed a PC-based program called Fuji that identifies the point where two tropical cyclones begin to interact, calculates the mutual rotation during the interaction, and indicates when the interaction ceases, if appropriate. During the mutual interaction phase, the midpoint can be advected using any model to calculate that component of the translation. The rotation of each storm is independently added to the translation of the midpoint to derive the tracks of each cyclone. The program has led to about a 50 percent reduction in the track forecast errors of interacting storms. However, the midpoint may not be the optimum point from which to determine the translation, and research is needed to identify the optimum location. The very active 1995 Atlantic tropical cyclone season may indicate the need to adapt the program for the Atlantic. The program was transitioned to TPC in 1993. Research involving binary tropical cyclones should also consider the interactions of tropical cyclones with other synoptic/subsynoptic circulations.

3.2.3.3 Use of Satellite-Derived Products. Although satellites are essentially observational tools, forecasters generally think that they can glean 24-36 hours of predictive value from animated satellite images, primarily from an assessment of trends and persistence. Bao (1981), Fett and Brand (1975), Lajoie and Nicholls (1974), and others have developed empirical relationships between tropical cyclone motion changes and cloud-system orientation, or the presence or absence of clouds. Dvorak and Mogil (1994) and Mogil and Dvorak (1995) demonstrate the use of water vapor imagery in tropical cyclone track prediction. Because these methods do not consider all of the factors responsible for tropical cyclone motion, the amount of operational skill they demonstrate has been limited. Nevertheless, each has something to offer, and more research is needed (for example, Jeffries et al. 1995; Puri and Miller 1990; Krishnamurti 1995) on assimilating satellite measurements into initial analyses.

3.2.3.4 Deep-Layer Averaging. The Atlantic NHC90 model and the JTWC JT92 models use a fixed 1000 to 100 mb deep-layer-mean (DLM) while the BAM models (Marks 1992) use intensity-dependent layer averaging for tropical cyclone "steering" computations. Research on Australian data sets (Velden and Leslie 1991) suggests that the depth of deep-layer averaging for steering purposes should be a function of tropical cyclone intensity. While this is conceptually logical and is confirmed by other studies for the Atlantic, such as Dong and Neumann (1983), it is not clear, except in extreme cases of a sheared environment, that intensity-dependent layer averaging can be profitably used by the forecaster.

The BAM model is routinely run using shallow-, medium- and deep-layer averages, and the three usually different forecast tracks provide the forecaster with information on the impact of the different atmospheric steering levels. At least one of these BAMs runs typically performs very well; however, based on verification statistics over the Atlantic (Gross 1991), the specific intensity versus DLM depth relationship is not clear. Research groups have studied deep-layer averaging in different basins with regard to both wind and geopotential height fields. While there is general agreement insofar as the intensity dependent concept is concerned, there is some disagreement as to the best layers to use under given tropical cyclone situations. Part of the problem may be related to different analysis methodologies in the different basins. Thus, research in this area may be basin or even analysis dependent.

Most of the research studies on deep-layer averaging have been accomplished in a research environment where best-track data are used to establish relationships. The question arises as to the use of different layer averages in an operational environment where analyses may be preliminary and the current and forecast intensity of the storm are not known with the same precision as with the best-track data and the use of a final analysis. With these uncertainties, it may be better to use a fixed rather than a variable layer average. This can only be determined by additional applied research.

3.2.3.5 Systematic Approach to Tropical Cyclone Prediction. A new approach to tropical cyclone track prediction is being developed by the Naval Postgraduate School. This approach is based on the fact that families of tropical cyclone tracks tend to be associated with recognizable and repetitive parent synoptic regimes. Identification of these regimes and statistical knowledge of tropical cyclone characteristics and motion associated with the regime could help the forecaster to improve over the forecasts from numerical models and other objective aids. Carr and Elsberry (1994) and Carr et al. (1995) have proposed a systematic approach to tropical cyclone track forecasting that helps forecasters to (1) interpret the tropical cyclone motion ramifications of evolving global model fields, and (2) anticipate errors in the tropical cyclone forecast tracks provided by the global model and by other objective aids that use the numerical model output. Recognition of conditions where the numerical guidance is apt to falter would go a long way toward reducing average forecast error. Operational testing is currently underway at JTWC (Boothe et al. 1995).

3.2.3.6 Statistical Models. The so-called statistical-synoptic models are gradually being retired from the prediction model inventories of the various centers, although one such model (PSS) is still being used for the eastern Pacific basin (see section 2.3.1.1). In view of the improved performance and timeliness of statistical-dynamical (see section 3.2.3.7) and purely dynamical methods, statistical-synoptic models, which do not utilize prognostic fields, are becoming obsolete.

CLIPER-class models (Neumann 1972) and the Half Persistence and Climatology (HPAC) models do not make direct use of any observed environmental data. As long as this limitation is understood by forecasters, they can be used to define the threshold skill/no-skill level for the given forecast situation. Also, such models perform quite well in basins or portions of basins where climatological patterns predominate, such as the eastern North Pacific. In fact, forecasters often assess the forecast situation as climatological or aclimatological in order to weigh the aids that they use in forecast preparations.

Currently, the eastern North Pacific CLIPER model is used for the entire basin from the North American west coast westward to the dateline. Applied research is needed to determine if a separate CLIPER model is needed for the central North Pacific region (140 to 180W) where tropical cyclone motion is less persistent than in the area east of 140W. Higher track forecast errors west of 140°W (see section 2.3.2.2) suggest this need.

Temporal and spatial distributions of tropical cyclone tracks show that they tend to be repetitive and to be associated with identifiable and repetitive synoptic patterns (Neumann 1979). Analog models attempt to identify an existing track with a track family through a series of computer algorithms. However, they are only mildly successful, and the process of identifying parent distributions of tropical cyclone tracks remains somewhat subjective. Recent work at the Naval Postgraduate School (Boothe et al. 1995; Carr and Elsberry 1994) suggests that the ability to identify synoptic patterns can be quite profitable in the forecast process, and continued research is needed.

3.2.3.7 Statistical-Dynamical Models. Statistical-dynamical models have demonstrated the ability to make productive use of global model output insofar as motion prediction is concerned (DeMaria et al. 1990). However, the increased resolution and tropical cyclone bogussing (the insertion of synthetic observations to represent the tropical cyclone) in current global models are masking the steering flow from which the statistical-dynamical models derive predictive skill. While these changes are clearly beneficial to the numerical models, the performance of the TPC NHC90 and the JTWC JT92 models (statistical-dynamical models) is deteriorating as numerical model resolution increases. This is a classical statistical pitfall where the attributes of the data upon which the models were developed have dramatically changed. The solution is to remove the tropical cyclone vortex before the numerical model output is used (Neumann 1993; Kurihara et al. 1995) and to develop new model prediction equations based on a data set of recent increased resolution analyses.

3.2.3.8 Numerical Models (NOAA). The current suite of numerical models for the Atlantic and eastern Pacific basins is given in section 2.3.1. The GFDL regional model was first run in a semi-operational mode during the 1994 season, and recent TPC verification statistics indicate that the model is now one of the best-performing track prediction models over the Atlantic and the eastern Pacific. While the model also predicts intensity, the results have been disappointing, and the model developers have a clear strategy for improvement (Kurihara et al. 1995). Further, the GFDL model needs to be evaluated for its ability to handle multiple cyclone situations, especially where the cyclones may be interacting with each other or with other atmospheric circulations. The model did not perform well with multiple storms in the Atlantic during 1995. Finally, the GFDL model needs to be evaluated for other basins in which it will be used. A limited evaluation of the GFDL model has been performed on 16 western North Pacific tropical cyclones. The results of the 124 model runs are shown in Table 3.3. A homogeneous sample comparison also was made with NOGAPS forecasts. While the results are encouraging, the evaluation represents a limited sample and may not reflect the most difficult forecast situations during the life of the tropical cyclones.

Table 3.3. Position errors (nm) for a completely homogeneous set of cases involving the GFDL model, NOGAPS, and the official JTWC forecast. The percent improvement of GFDL to NOGAPS is also shown.

Tropical cyclone motion forecasts are also available from the global aviation model; however, recent performance statistics from this model were rather poor (see Table 2.2). Numerical models, in general, will benefit from better knowledge of the initial state of the atmosphere. Recent HRD/AOML research experiments with Omega Dropwindsonde (ODW) enhanced analyses (Burpee et al. 1996) clearly demonstrate this potential. Planned Global Positioning System (GPS) dropwindsonde surveillance missions with the recently acquired NOAA jet aircraft (see section 3.4.1.3) will also provide enhanced analyses. The development of the GFDL model, the ODW synoptic flow experiments, and the planned operational surveillance missions are good examples of successful and ongoing technology transfer from the research to the operational environment.

3.2.3.9 Numerical Models (DOD). NOGAPS--an 18-level, T159 triangular truncated global spectral model--is the DOD's primary numerical prediction model (see section 2.3.3.1). While it is difficult to determine the exact performance characteristics of the model because several changes have been made in the model resolution and physics over the last few years, it is possible to identify some general strengths and weaknesses. While the NOGAPS has a relatively good bogus scheme, the bogus may be improved in terms of vertical extent (depth) and size. The model also appears to have a specific bias with regard to weak and small tropical cyclones.

The most serious deficiency in the NOGAPS model is its inability to consistently track the vortex, especially at the 48-72 hour forecast periods. This is absolutely critical if the DOD expects to produce extended outlook forecasts to 120 hours. It is also critical in achieving the stated accuracy requirements of 50, 100, and 150 nm for 24, 48, and 72 hour track forecasts, respectively. Another deficiency is the tendency for the model to merge small but discrete low pressure centers or cyclones into a single center. Finally, more skill is needed in predicting the onset of monsoon surges and the movement and development of Tropical Upper Tropospheric Trough (TUTT) cells. While these problems may be a result of sparse data and limited model resolution, it may be possible to model the TUTT and monsoon influences, and insert appropriate synthetic observations. While additional work is needed to address these problems, NOGAPS, for the most part, performs well in the tropics.

3.3 Surface Wind Description. The primary need is the ability for improved specification of surface wind parameters used in public advisories and in forecasts/advisories. This would include the initial maximum wind and gusts, the radii of 100, 64, 50, and 34 knot winds, and forecasts of these surface wind parameters at 12, 24, 48, and 72 hours. Secondarily, the term also implies a knowledge of the vertical distribution of wind, the relationship between surface winds and sea-level pressures, the radius of maximum wind, and the eye diameter, the initial value of the latter being one of the required items in the forecast/advisory.

Skill in operational surface wind forecasts is limited, and the difficulty is compounded by the inaccurate specification of the initial conditions. Even in Hurricane Andrew, which devastated South Florida, maximum winds and their variation with time and space, sustained winds for different averaging times, the presence or absence of tornadoes, and the relationship between surface winds and those measured at flight level remain controversial.

Mainly through use of satellite microwave sensors, planetary boundary layer (PBL) models, and modifications to the Dvorak technique, a number of research groups are attempting to improve the capability to estimate surface winds in and around a tropical cyclone. The general feeling among both operational and research groups is that the sensors from both polar-orbiting and geostationary platforms have not been fully utilized and that a multisensor approach is appropriate. Because of rainfall contamination, limited footprints, model assumptions, and general subjectivity, the ability to describe the wind field with the detail needed for future numerical models is limited, and new techniques for obtaining surface (and other) winds in tropical cyclones are needed. Deployment of fixed meteorological buoys, establishment of additional in situ methods, and Doppler radars provide additional capability to obtain surface winds. When storms pass over these in situ measurement platforms, the data are very helpful in calibrating remote-sensing systems and in better understanding structural changes.

HRD/AOML and the University of Massachusetts have demonstrated the potential of obtaining surface winds from aircraft using a combination of data from a C-band scatterometer and a stepped-frequency microwave radiometer. On-board computer workstations can provide real-time surface wind speed and direction estimates for relay to forecast centers. Work also continues (OFCM 1995) on using over-the-horizon radar (OHR) to obtain distant wind fields. The OHR is a phased-array radar that transmits and receives electromagnetic energy over very long distances by utilizing reflection from the ionosphere. Ocean waves provide backscatter signals that give a good indication of the local wind direction, but only a fair to poor indication of the speed. Although the technique appears promising, there are many practical problems with the methodology that have led to a lack of progress in the operational use of the technique. Details are provided by Foley (1995).

Four events have brought new emphasis to the problem of describing tropical cyclone surface winds: (1) the loss of aircraft reconnaissance in the western North Pacific basin; (2) the critical role that storm-surge prediction now plays in emergency planning; (3) the extent of gale-force winds, which has become a key element in the timing of emergency actions; and (4) NOAA's acquisition of a jet aircraft. The significance of the latter is twofold: (1) long-range plans call for these high-flying aircraft to have the ability to estimate surface winds; and (2) current evidence points to the upper troposphere and perhaps the lower stratosphere as the locus of important interactions relative to tropical cyclone intensity changes (NCAR 1995).

3.3.1 Intensity Estimate. The Dvorak technique (Dvorak 1984) is widely used for the estimation of tropical cyclone intensity, but this extremely useful technique needs to be upgraded to take into account various deficiencies noted by users in the various basins. Perhaps separate versions of the technique are needed for each basin. A promising multispectral satellite approach to improving tropical cyclone intensity analysis and prediction is currently being undertaken at NRL West at Monterey (Hawkins et al. 1995).

A number of initial intensity estimates are available to the forecaster; however, many of these are asynoptic, are not at the surface, use different wind-averaging times, are not at the optimum location, etc. This makes it difficult for a forecaster to assimilate these "bits" of information. Ideally, these data need to be collated and normalized into a surface wind field analysis. Powell and Houston (1995) of the HRD/AOML have demonstrated that such an analysis is possible, and work in this area is ongoing. In addition to assisting the forecaster, such analyses are useful in damage assessment (Powell et al. 1995) and in storm-surge work (Houston et al. 1995). Additional wind measurements would greatly help in the preparation of these analyses.

Associated with the requirement of specifying surface winds is the requirement of specifying gusts. Although the sustained wind vs. gust relationship varies from storm to storm, forecasters have only average relationships to use when forecasting gusts. On the other hand, the observed peak gust in the strong quadrant appears to be the most representative value for determining maximum sustained wind over land and water. Black (1993) presents a summary of current knowledge relative to this topic. The relationships among maximum winds aloft, peak gusts, and surface winds need to be examined closely, and the Doppler radar should be a beneficial tool in this evaluation.

3.3.2 Relationship Between Winds and Pressures. The relationship between winds and pressures is important since both are used interchangeably as measures of intensity. Many basin-dependent relationships, such as by Atkinson and Holliday (1977) for the western North Pacific, Love and Murphy (1984) for northern Australia, and Kraft (1961) for the Atlantic, have been developed. While these relationships are useful in most instances, there are some serious deficiencies: (1) the winds in midget typhoons are underestimated by the Atkinson-Holliday relationship (Guard 1995a); (2) use of the Kraft (1961) relationship for the Atlantic (which was derived from Gulf of Mexico data) yields winds that are too high for a given pressure in the more northerly latitudes (Neumann 1995); and (3) some wind-pressure relationships have been extrapolated beyond the range of the developmental data (a statistical pitfall).

Given the importance and credence given to wind-pressure relationships (WPR) to operationally determine maximum surface winds, it seems that it is time to evaluate carefully current WPRs. There are basic and compelling reasons to do this:

• Considerable scatter occurs between observed minimum pressure and observed maximum wind.

• The wind measuring technology, especially with reconnaissance aircraft, has improved considerably over the past decade.

• Specific questions remain:
  • Are relationships between flight-level winds and surface winds valid?
  • Are relationships between peak gusts and sustained winds valid?
  • Does deep convection bring strong winds from the free atmosphere into the boundary layer?
  • At what level are the maximum winds located?
  • What variables need to be included in statistical wind-pressure relationships?
  • Would hurricane intensity be more accurately described as a range of winds?

• One effect of using WPRs has been to discard wind observations that deviate significantly from those expected from the corresponding pressure values. (This is not a bad practice whenever winds are lower than expected but is more serious when winds are stronger. There are many more reasons to explain weaker-than-expected winds, and even if wrong, errors are toward over-warning versus under-warning.) The WPRs have been used to "average" the data to make it fit a common curve. This procedure has influenced post-storm analyses and resulted in an intensity data sample that may be questionable. If the best-track data are then used as "truth" for validating new research and techniques, the WPR becomes incestuous.

A stumbling block in establishing new relationships is that the archived databases are often poor. In the Atlantic, archived intensities for early storms contain simultaneous winds and pressures that are highly suspect. In the Pacific, examination of wind/pressure data sets maintained by various agencies have widely different values for given storms at and after landfall. This is particularly troublesome both when attempting to develop and validate intensity forecasting techniques and when using these data sets in damage assessment and storm-surge models. Part of the problem is the different wind/pressure relationships utilized by different countries and the use of different wind-averaging times. Additional work on these data sets is needed.

Little documentation is available on the various archived tropical cyclone data sets that specify historical tropical cyclone positions, winds, and pressures in the various basins. Engineering interests use these data to establish maximum possible storm intensity, wind-pressure relationships, design criteria, damage assessment models, etc., and assume the data are accurate; however, it is known or suspected that:

• Documented winds are too high in some years for both the Atlantic and western North Pacific basins (Black 1993; Landsea 1993).

• There are inconsistent simultaneous winds and pressures in the Atlantic data set (Neumann 1994).

• The landfall locations are incorrect on many of the early Atlantic tropical cyclones (Neumann 1995).

• Early eastern North Pacific storms were poorly documented after they moved inland.

These and numerous other problems are generally known to people working extensively with these data and should be better documented. The TPC does provide Technical Memoranda describing their tropical cyclone data sets, but the publications are in need of revision. The limitations of these data sets need to be assessed and known errors corrected.

3.3.3 Intensity Forecasts. The prediction of tropical cyclone intensity by simple statistical models based on climatology and persistence has little skill, and these models specifically lack the ability to identify the unusual and dangerous rapidly developing storms. The most destructive hurricanes reach extreme intensity not through slow intensification but through the process of "rapid deepening" that may take a hurricane from Category 1 or 2 on the Saffir-Simpson Hurricane Scale to Category 4 or 5 in a day or two. Hurricanes Gilbert, Hugo, Andrew, and Opal are examples of the phenomenon. Rapid deepening is a forecast challenge because it appears to be essential to the development of most major hurricanes and because it can happen so quickly that events outrun the forecasts. For example, Hurricane Opal intensified from Category 2 to Category 5 as it approached the Gulf Coast in October 1995. An unprecedented disaster was avoided only when the hurricane weakened just offshore.

It is clear that upper tropospheric interactions between hurricanes and cyclonic features in their immediate environments often trigger rapid deepening, but the detailed dynamics are not well understood. Forecasters have no skill in forecasting the onset of rapid deepening. Research experiments, using aircraft, such as the NOAA's Gulfstream IV-SP, flying in the upper troposphere, are essential to better understanding and forecasting of rapid deepening.

A second forecasting challenge is "concentric eyewalls." In the concentric eyewall case, a second ring of convection forms around a preexisting hurricane eye. The outer eyewall has a local wind maximum and tends to propagate inward in response to the latent heat released by convection just as single eyewalls do. Eventually, the outer eyewall constricts around the inner eyewall and strangles it, often leading to weakening if the hurricane has initial winds greater than 100 knots (50 m/s). Hurricanes Andrew and Opal both experienced "such eyewall replacements." Andrew resumed rapid deepening and passed onshore with essentially the same intensity that it had before the replacement. Opal, on the other hand, remained in a weakened state. While the process of multiple eyewall development is well understood, no reliable predictors exist to forecast timing and the amount of weakening or reintensification.

Statistically based intensity change models that include environmental predictors (DeMaria and Kaplan 1994; Fitzpatrick 1995) and empirical models of inland decay of landfalling storms (Kaplan and DeMaria 1995) show some promise. The mesoscale modeling work of Pielke and collaborators at Colorado State University, using the Regional Atmospheric Modeling System (RAMS), for intensity prediction is also encouraging (Eastman 1995). Research is needed, however, to better identify, through both observational studies and numerical model simulations (e.g., Krishnamurti et al. 1989; Kurihara et al. 1995), the internal and external (environmental) parameters associated with intensity changes. A minimum set of measurements required for intensity prediction needs to be determined.

3.3.4 Radii of Surface Winds. While the forecast/advisory message specifically requires the initial and the forecast radii of certain threshold winds, specifying the radii of certain critical winds is but one aspect of the much larger problem of specifying and understanding the entire three-dimensional tropical cyclone structure and how this structure relates to the environment in which the tropical cyclone is embedded, including the underlying land or sea boundary. Although simply defining the existing structure of a tropical cyclone is a formidable problem, forecasting of structural changes is even more formidable and will obviously require detailed numerical modeling efforts.

To construct initial analyses of wind radii on a routine basis, the observational capability and understanding of wind structure in tropical cyclones must be greatly improved. Wind structure is extremely complex, and a research project to gain a better understanding of surface wind distribution is the first step in this effort. This will require a better understanding of the asymmetric structure and temporal changes of this structure through four-dimensional analyses even though ultimate interest is primarily at the surface. Air-sea interactions (Ginis et al. 1995) may also play a major role. Additionally, other problems in specifying the wind field, such as the transition to semi-tropical or extratropical systems, monsoon interactions, and the effects of varying terrain on islands and continental coastal/inland areas on landfalling tropical cyclones, will also require investigation.

Continued research is required to develop airborne and satellite sensing techniques to observe the surface wind distribution. Airborne stepped-frequency microwave radiometers (SFMR) for surface wind speed and rainfall rate measurement and a C-band scatterometer (C-SCAT) for surface wind speed and direction can be used in combination to determine surface wind vectors. This instrumentation is currently being tested on a NOAA WP-3D aircraft. Future plans call for SFMRs to be installed on the operational WC-130 aircraft, and the SFMR/C-SCAT combination will be installed and tested on NOAA Gulfstream IV-SP.

DMSP SSM/I measurements can detect outer wind structure, but improved techniques are needed to remove contamination from rainfall in regions closer to the tropical cyclone center. Currently, these measurements are useful mainly to determine the outer extent of tropical storm-force winds. JTWC and NCEP are also using data from the scatterometer on the European Remote Sensing (ERS-1) satellite, which employs an active microwave sensor to track ocean surface waves. The measured backscattered microwave radiation is then used to define the ocean wind field. While the swaths of data are narrow, the data appears to be very accurate for wind speeds less than 50 kt, and the resolution is also very good. With the launch of ERS-2 (to replace ERS-1) and of a joint NASA-Japan polar-orbiting satellite carrying the NASA scatterometer (N-SCAT), the amount of coverage should greatly increase.

Finally, the use of airborne and coastal Doppler radar in determining the initial three-dimensional wind field has great potential. Airborne Doppler radar wind data, such as that obtained from the two NOAA/AOC WP-3D aircraft in eastern Pacific Hurricanes Jimena of 1991 and Olivia of 1994 and Atlantic Hurricanes Iris and Luis of 1995, are good examples. Dual-Doppler capability from the coastal WSR-88D radar network needs to be developed as well. This would also assist in determining the precise location of the wind center of tropical cyclones.

3.3.5 Hurricane/Typhoon Vertical Wind Structure. Doppler radar data are confirming that the level of maximum winds in tropical cyclones may be much lower than thought a few years ago and may be near the top of the mixed layer, which averages just above 500 m. This is considerably lower than the level of routine aircraft reconnaissance flights. Black (1993) provides a summary of current knowledge of low-level vertical wind distribution in tropical cyclones. The radars also indicate that deep convection can transport the low-level winds vertically very rapidly and efficiently. The temporal changes (spin-up and spin-down times) of tropical cyclone winds, especially for tropical cyclones above 125 knots, are poorly understood. For example, very intense tropical cyclones may spin-down much more rapidly than weaker ones. A research initiative may be needed to resolve the temporal changes in near-surface winds for intensifying and weakening tropical cyclones.

3.4 Synoptic Environment.

3.4.1 Tropical Analysis. Because of data deficiencies, small standard deviations of parameters being measured, non-geostrophic relationships, and lack of proven conceptual models, tropical analysis has always presented special problems. While some tropical cyclone forecast centers continue to produce their own tropical analyses, increased emphasis on centralized analysis and prognoses on a global scale and the concept of a "paperless office" are causing local analyses to be phased out. A good example is the TPC automated low, high, and deep layer analysis package; however, the inability of present computerized methods to handle small-scale features, such as tropical easterly waves, continues to foster locally produced surface analyses. Caution should be taken in the complete elimination of hand-drawn analyses. The value is not in merely producing a final analysis product. A much deeper value is inherent in the process of assessing each piece of data spatially and temporally, integrating that assessment into an analysis, and then reanalyzing it as more data are available. Doswell (1986) strongly cautioned against removing the forecaster from the analysis. The interactive analysis work of Jeffries et al. (1995) needs to be assessed as an automated alternative.

3.4.1.1 Exploit Existing Data. While NOAA's Gulfstream IV-SP will provide valuable in situ observational data in the tropical cyclone environment, there is a need to make better use of existing nonconventional data sources, some of which are still under development. This would include surface winds and precipitation from aircraft using the combined SFMR and C-SCAT methodology and from satellites using SSM/I and scatterometer methodology. Also, algorithms need to be developed to obtain maximum benefit from Doppler radar. Included would be dual-Doppler wind fields from aircraft and "pseudo" dual-Doppler winds from the WSR-88D coastal radar network.

Since the late 1980's, HRD/AOML has provided TPC with real-time analyses of the hurricane core structure based on aircraft observations. These analyses, in storm relative coordinates, enable forecasters to better assess the structural changes taking place. HRD/AOML has also developed other analysis schemes, such as the nested scheme used by the VICBAR barotropic model and the analysis of the surface wind field. There are other potential sources of derived surface winds in and around tropical cyclones. Many of these are off-time, off 10-meter level, or off 1-minute averaging times and tend to be isolated both temporally and spatially. Applied research focused on the capability to assimilate these and other conventional data into a routine surface wind analysis, say within 500 km of tropical cyclone centers, is needed.

3.4.1.2 Improve Specification. Many of the synoptically important features of the tropics have horizontal scales that are not adequately resolved by the regular observational network. Taken collectively, there is a wealth of environmental data on a variety of scales available from weather satellites. With the successful launch of the GOES-8 and GOES-9 satellites and their improved sensor suites, considerable research is possible to develop a satellite multisensor approach to extract improved estimates of the various environmental parameters in and around tropical cyclones. Velden (1995) suggests that the data being provided by GOES-8 (and GOES-9) are superior to observations from previous geostationary satellites and have the potential to impact positively both the subjective and objective analysis of tropical cyclones and their environment. The water vapor images from GOES-8 are of unprecedented quality and preliminary indications (Velden 1995) are that image sequences are providing substantially improved wind vectors in cloud-free regions. A long-standing problem in need of continued work is the correct assignment of heights to derived wind vectors. Also needed are improved analysis methods for winds derived from the water vapor imager, which are representative of deep atmospheric layers.

Sensors from polar-orbiting satellites either, including the passive microwave SSM/I sensors on the DMSP polar orbiters, the Microwave Sounding Unit (MSU) on the NOAA polar orbiters, the microwave sounders on the DMSP spacecraft (SSM/Tl & T2), and the scatterometer on the European Remote Sensing (ERS-1) polar orbiter, have not been fully utilized. Additional exploratory and applied research efforts are needed to assimilate these data into current analyses.

3.4.1.3 Aircraft Observations. As indicated earlier, current evidence points to the upper troposphere and perhaps the lower stratosphere as the locus of important interactions relative to tropical cyclone intensity changes (NCAR 1995). Satellite imagery has not been able to provide the needed data. NOAA's Gulfstream IV-SP jet aircraft will provide some of these high level data. As the operational capabilities of the Gulfstream IV-SP evolve, it will have three roles:

• Surveillance. In its intial role, the Gulfstream IV-SP aircraft will collect high level horizontal observations and GPS dropwindsonde observations, which will be transmitted to NCEP for enhancement of initial analyses used by global and hurricane models. Exhaustive tests (Burpee et al. 1996) have demonstrated that these enhanced analyses can provide significant improvements in hurricane track prediction models. This potential for improving hurricane forecasts was a major justification for acquisition of the jet aircraft. The aircraft will initially have only a Dopplerized C-band nose radar. Ultimately, it will be equipped with a second instrument that combines a vertically scanning Doppler radar for observing profiles of the three-dimensional wind field below flight level with a scatterometer for remotely sensed surface winds. This instrument is essential for the jet to reach its potential as a reconnaissance platform.

• Reconnaissance. Reconnaissance missions, as specified in the National Hurricane Operations Plan (OFCM 1996), will continue to be performed by AFRES WC-130 aircraft and supplemented, as required, by NOAA's WP-3D aircraft. It is necessary to stress that the new jet aircraft will not replace the need for the existing AFRES and NOAA aircraft fleet. For the immediate future, these are entirely different programs. The only overlap will be that WP-3D research aircraft will have an additional role in validating new instrumentation for the jet aircraft. This new instrumentation should eventually permit the jet aircraft to conduct reconnaissance and research missions in addition to, or in conjunction with, the surveillance missions.

• Research. The 1995 hurricane field program plan (Marks and Friedman 1995) of the HRD/AOML lists eight major experiments that require the use of one or both WP-3D aircraft and occasional use of AFRES WC-130 aircraft. The basic objective of NOAA's hurricane research field program is the collection of descriptive data that are required to support analytical and theoretical studies. Although these studies are designed to improve the understanding of the structure and behavior of hurricanes, the ultimate purpose is to develop improved methods for hurricane prediction. There is ample evidence that this ultimate purpose is being realized. A classic example is provided by the hurricane synoptic-flow experiments (Burpee et al. 1996) where a reduction in forecast error has been demonstrated.

In anticipation of the proposed multiple roles of jet aircraft (surveillance, reconnaissance, and research), it is very important that current capabilities of WP-3D and WC-130 aircraft in providing multilevel data not be degraded. The general tendency since aircraft reconnaissance began in the mid-1940's has been to fly at increasingly higher levels away from the boundary layer where data are needed and in situ measurements are preferred. Because the tropical cyclone structure varies from storm to storm, there have been some difficulties in obtaining reliable surface wind fields from aircraft flying at mid-levels. These difficulties could be amplified at very high flight levels. Thus, before jet aircraft are accepted for other than high level surveillance missions, considerable research is needed to ensure that reliable boundary-layer wind and pressure observations will be available. The use of combined C-SCAT and SFMR instrumentation (see section 3.3.4), currently being tested on the WP-3D aircraft, would appear to be the most promising approach. As the missions of the jet aircraft evolve to reconnaissance and research, the optimum instrumentation for these missions must be determined, developed, and procured. New sensors (e.g., SFMR) or modification/upgrading of existing sensors for the WC-130s and the WP-3Ds are also being planned and programmed.

A miniature autonomous aircraft called the Aerosonde is being developed. The aerosonde offers an economical method for obtaining in situ meteorological data over the data-sparse tropical regions (Holland et al. 1992). The Aerosonde recently completed successful field testing as part of the international Maritime Continent Thunderstorm Experiment off northern Australia (McGeer 1996). The goal is to produce an economical recoverable aircraft ($10,000/copy) that can obtain vertical and horizontal atmospheric data (wind, pressure, temperature and humidity) and that can operate for tens of dollars per flight hour. The plan is to add a turbo-charged engine in the 1998-1999 time frame that will allow the aircraft to operate with a duration of 4 days and up to an altitude of 16 km. The aircraft will be further tested in other meteorological field experiments. Testing will include its ability to penetrate the core of tropical cyclones and gather eye/center data. This research is being funded by ONR, the Australian Bureau of Meteorology, Taiwan, and Sencon Environmental Systems Pty Ltd of Australia.

3.4.1.4 Improvements to the Dvorak Method. For many years, forecast centers have relied on intensity estimates derived from the Dvorak method (Dvorak 1975) applied to visible and infrared imagery (Dvorak 1984). As with other procedures, continued operational usage has disclosed some shortcomings in the method: multiple cloud decks obscuring the features required for more accurate classification, identifying rapidly intensifying or weakening storms, analyzing tropical cyclones generated in the monsoon trough, as well as proper analysis of midget typhoons. Additional criticism stems from the technique's subjectivity and improper application; it is not simple to apply, even by experienced analysts. However, extensive global use of the technique and continued improvement in satellite imagery and sensor technology are leading to local modifications and improvements to the technique. NESDIS, NRL, JTWC, and the University of Wisconsin have been quite active. The greatest needs are to improve the technique's assessment of tropical cyclone intensity and intensity changes and to better adapt the technique for the large variety of tropical cyclone structures observed in the western Pacific and other strongly monsoonal basins.

3.4.1.5 Light Detection and Ranging (Lidar) Winds. It has been clearly demonstrated (Burpee et al. 1996) that increased tropospheric wind observations in numerical models will improve tropical cyclone forecasts. Feasibility studies have shown that it is possible to measure such winds from a space-based Doppler lidar. Targets would be cloud particles or naturally occurring aerosols suspended in the atmosphere and moving approximately with the speed of the wind. The concept, current research, and potential improvements in climate and weather forecasting, including improvements in tropical cyclone intensity and motion forecasts, are reviewed by Baker et al. (1995). Space-based lidar wind systems would appear to be an ideal method for obtaining winds on a global scale, and a wide range of research, as well as promotional and funding efforts to assure inclusion of such systems on future spacecraft, are needed. These systems, however, are extremely expensive and the issue of cost effectiveness must be considered (beyond the scope of this plan).

3.4.2 Improve Forecasting. There is little doubt that better initial analyses will yield improved forecasts. The main challenge for numerical prediction is to develop the methodology to ensure that all relevant data are included in operational analyses. This would require advanced development efforts to improve techniques for inserting synthetic observations, satellite-derived winds and temperatures, and reconnaissance data in such a way as to ensure that the data assimilation technique will accept representative observations and reject faulty data. Also, improved model physics, including advanced physical parameterizations of the tropical boundary layer, surface fluxes, and cumulus convection, will generate a better background field for observations and provide a more realistic depiction of the large-scale tropical circulations that evolve during the forecast.

3.4.3 Tropical Cyclone Genesis. Once the speed of a vortex exceeds about 10 m/s, frictional convergence becomes strong enough to organize the convection on the scale of the vortex. At weaker intensities, the reasons why storms develop on the observed scale are unclear. Nevertheless, forecast offices are required to issue routine tropical weather outlooks that briefly describe significant areas of disturbed weather and their potential for tropical cyclone development out to 48 hours. Not only are the environmental factors responsible for the genesis of tropical cyclones not well understood, but they are probably not even being measured or at least not being measured at the proper scale or atmospheric level. Genesis cannot be easily defined because there appear to be many diverse pathways to tropical cyclone development. A summary of the unanswered questions in regard to genesis is given by Frank (1987).

By far the most common approach to genesis prediction is subjective/objective pattern recognition techniques using satellite imagery, such as those developed by Dvorak (1984) and Zehr (1995). However, surface and upper-air analysis signatures, genesis indices, decision ladders, and time clustering (conditional climatology) have also been used. Until there is a better physical understanding of genesis, continued research is needed to assist the forecaster in determining which cloud clusters will develop and which will not. In the more objective sense, studies with properly initialized numerical models of relatively small scale will be needed to develop learning tools for determining which cloud clusters have the potential for development into tropical cyclones. Aircraft field experiments in developing waves, such as those conducted by the HRD/AOML (HRD 1994), are also needed. As noted in Table 3.1, the forecast centers assign a rather low priority to genesis when compared to priorities assigned to the understanding and prediction of motion, intensity, and structural changes of existing tropical cyclones. This is understandable in that the consequences of a missed genesis forecast are minor compared to the consequences of a missed track or intensity forecast. However, the requirement for the prediction of genesis is sufficient justification for continued research.

3.4.4 Other Model Upgrades

3.4.4.1 Monsoon Models. Winds within deep monsoon flow may penetrate vertically to 200 hPa (40,000 ft) with the strongest winds in a very narrow column hundreds of nautical miles long and frequently as little as 25 nm wide (Guard 1986; Harr et al. 1993). The sparse observations over oceanic areas are not adequate to properly initialize the global models, which often do not properly reflect the intensity or the depth of the monsoon, especially under surging conditions when convection is vigorous. These surges are frequently associated with tropical cyclone genesis and are frequently directly linked to mature typhoons. Additional efforts are needed to (1) properly model monsoon surges and (2) insert the monsoon structure into the global models through the use of synthetic observations.

3.4.4.2 Tropical Upper Tropospheric Trough (TUTT) Models. TUTT systems appear to be important in initiating persistent convection that can eventually become a tropical cyclone, in occasionally contributing to rapid development of midget tropical cyclones, and in often providing an efficient outflow channel during rapid intensification in tropical cyclones. Again, sparse observations over the oceans fail to resolve the true vertical and horizontal structure of the cells, and model initialization is flawed. The active 1995 Atlantic hurricane season revealed that TUTT cells can be as important in that basin as in the Pacific, and it has been pointed out (Fitzpatrick et al. 1995) that the NCEP MRF model does not handle the TUTT cell very well in the Atlantic. Exploratory research is needed to model these cells, and then an applied research effort is needed to bogus the cells into the models using synthetic observations. Satellite water vapor data may help in this goal.

3.4.4.3 Twin Cyclones. Periodically, convection builds along the equator and persists sufficiently long to induce strong westerly winds. The westerlies perpetuate the convection, which eventually splits and moves poleward in each hemisphere. Occasionally, both of the cloud masses develop into twin tropical cyclones in the western Pacific and in the Indian Ocean. Although the event has been studied by several investigators (e. g., Lander 1990), the physical processes in the evolution are not completely understood. Exploratory research is needed to model the evolution of these storms and then an applied research initiative is needed to ensure the global models can duplicate the process, either through improved physics, improved convective initialization, or insertion of synthetic observations.

3.4.4.4 Hybrid Storm Systems. In the Atlantic basin and northeastern portions of the western North Pacific basin, tropical cyclones sometimes evolve from purely baroclinic (cold-core) systems. Hebert and Poteat (1975) provide a satellite imagery pattern recognition system for identifying and estimating the intensity of such systems in the Atlantic, but the mechanism for their formation and maintenance, and for related events such as the formation of tropical cyclones in areas of low sea surface temperatures (Merrill 1987), are not well understood. In the western Pacific, monsoon depressions that may be coded at 25-35 kt using the Dvorak technique with a shear-type pattern may have 40-70 kt winds displaced 150 nm or more from the light and variable wind center. These variations from the Dvorak scheme may be basin dependent, but an intensity scheme needs to be developed to accurately estimate surface winds associated with such systems and to transition smoothly from the monsoon depression phase to the tropical cyclone phase. Until such time as sufficient observational data permit a more realistic physical initialization of models, research efforts should focus on refinements in the currently used satellite pattern recognition techniques.

3.4.4.5 Global Teleconnections. Prompted by the work of Dr. William Gray and associates at Colorado State University (CSU), there has been renewed interest in seasonal prediction of basin-wide tropical cyclone frequency and intensity based on larger scale global circulations. A summary of the relationships between selected meteorological factors and their influences on changes in interannual tropical cyclone activity in the various worldwide basins is provided by Gray et al. 1992. His successful prediction of a very active 1995 Atlantic season is particularly noteworthy. Possible development of similar schemes for other basins needs to be investigated.

The NOAA Office of Global Programs (OGP) is sponsoring a Pacific El Niño-Southern Oscillation (ENSO) Applications Center that resides jointly at the University of Hawaii and the University of Guam. Research being conducted there and at other national and international institutions is revealing specific relationships between ENSO and changes in the distribution of tropical cyclones in the Pacific.

The prospect of global warming raises many questions as to the effect on tropical cyclone frequency and intensity. While scientists at a joint session of the World Meteorological Organization and the International Congress of Scientific Unions held in Huatulco, Mexico, in November 1993, concluded that global warming from a predicted doubling of carbon dioxide in the next 5-6 decades would probably not significantly alter the numbers or intensities of tropical cyclones (Lighthill et al. 1995), this conclusion is not unanimous (Emanuel 1995; Broccoli et al. 1995). Also, there may be some second-order effects, such as changes in monsoon circulations or jet stream location, which may be significant to tropical cyclone behavior. As climate models become more sophisticated and better able to resolve smaller scale features on smaller time scales, this question will have to be revisited. Thus, continued research is needed.

3.4.4.6 Extratropical Transition. As tropical cyclones move into more northerly latitudes, transition from tropical (warm-core) to extratropical (cold-core) characteristics occurs. Forecast centers have the responsibility of determining the transition point, at which time tropical cyclone advisories are discontinued. In later verification of forecast errors, the extratropical stage is excluded. Since forecast errors are apt to be higher after this stage is reached, this adds a certain degree of subjectivity to verification data.

Some tropical cyclone models are not designed to handle the extratropical stages of tropical cyclones. In tropical cyclone best-track data sets, the method of treating the extratropical stage is not universal. In the Atlantic, the extratropical stage is indicated, and the track is continued for some time afterwards. In the western North Pacific, the track is usually discontinued after the extratropical stage is reached. Thus, statistical models developed for the western North Pacific basin exclude the extratropical stages of tropical cyclones.

Observational studies of the distributions of wind, temperature, moisture, convection, and precipitation during extratropical transition are needed. Complementary studies on the strengths and weaknesses of numerical models during extratropical transition are also necessary.

3.5 Rainfall. Tropical cyclones contain extremely warm, moist air. They cover an area of up to 1 million square km and move relatively slowly. Hence, they are capable of producing very heavy rain that may not necessarily be correlated with the tropical cyclone intensity. For example, weak 1994 tropical storm Alberto generated tremendous floods in Georgia, Florida, and Alabama, claiming at least 30 lives and caused about $500 million damage. Merrill, in WMO (1993), points out four reasons why quantitative precipitation forecasting (QPF) of tropical cyclone rainfall is very difficult:

• Rainfall, is difficult to measure accurately.

• Errors in track prediction can be a main contributor to rainfall forecast errors.

• Interactions between tropical cyclones and other weather systems are not well understood.

• Even within clearly defined threat areas, mesoscale processes may determine the distribution of heavy rainfall.

3.5.1 Tropical Cyclone-Specific Satellite Rainfall Estimates. Satellite estimates of rainfall associated with tropical cyclones have been developed; e.g., Spayd and Schofield (1984). While these techniques, based largely on a persistence forecast, are reasonably good over the oceans, they will probably need recalibration because of the higher quality data available from GOES 8/9. The effectiveness of these techniques decreases upon landfall--particularly in mountainous terrain--when the tropical cyclone rainfall-producing mechanisms undergo change. Continued research is required to refine these techniques to reduce the large errors that are frequently present in tropical cyclone landfall and post-landfall situations.

3.5.2 Other Tropical Cyclone Rainfall Forecasting Techniques. Aside from the standard QPF methodology, reliable methods of tropical cyclone rainfall prediction are not available. Effects of terrain and translation speed on precipitation distribution are understood only in a general way. It is not always apparent why some tropical cyclones produce substantially heavier rainfall than others under apparently similar conditions.

Rainfall estimates from the newly installed WSR-88D coastal radar network are reasonably good, but there seems to be a tendency for underestimation (Spratt and Nash 1995; Korotky et al. 1995), which requires investigation. Analysis of these data should help in short-range QPF and should provide considerable insight into structural changes of tropical cyclones and associated rainfall after landfall. In addition, the use of satellite microwave technology for tropical cyclone rainfall estimation needs to be further exploited.

The Tropical Rainfall Measurement Mission (TRMM) is a NASA-sponsored experiment to measure tropical rainfall accurately and ultimately calculate its contribution to the global energy budget. A suite of sensors will be placed on a satellite to be launched in an equatorial orbit. This mission provides an opportunity to greatly expand our knowledge of tropical cyclone rainfall as well as the limitations of the remote sensors. New rainfall algorithms will emerge that must be adapted for operational tropical cyclone rainfall analysis and prediction.

3.6 Doppler Radar. The coastal WSR-88D Doppler radar network promises to be a major source of data on structural changes as tropical cyclones make landfall or pass offshore, and should provide a wealth of data on downdrafts, mesocyclones, tornadoes, microbursts, precipitation patterns and processes, trochoidal motion, etc. Since these data will initially be available to local NOAA WSFO and military offices, there is excellent potential for these offices to participate in the research process. This trend can already be noted in papers presented at the AMS 21st Conference on Hurricanes and Tropical Meteorology, where it was noted that:

• There is a strong tendency to underestimate tropical-cyclone related rainfall.

• There is good agreement between winds aloft and peak surface gusts within tropical cyclone rainbands.

• Some weaknesses exist in the current WSR-88D algorithms for tropical-cyclone work.

3.6.1 Develop Algorithms. Much of the software currently available for the WSR-88D radar network is focused on severe local storms, so there is a need for additional software development focused on tropical cyclone needs. This would include algorithms for:

• "Pseudo" Doppler winds.

• Definition of the three-dimensional wind fields, including, where possible, extending winds vertically down to the surface.

• Locating the tropical cyclone center.

• Extending the range of Doppler wind acquisition.

• Projecting areas and rates of precipitation.

Users of Doppler radar in tropical cyclone situations need to be very specific about documenting strengths and weaknesses noted in hardware and software so that these can be better tuned to the tropical cyclone problem. For example, experience at JTWC suggests that the cumulative tropical rainfall is underestimated by at least 50 percent. These algorithm needs must be specifically documented and communicated through the proper channels to groups, such as the WSR-88D Operational Support Facility (OSF) at Norman, Oklahoma, which is actually developing the software.

3.6.2 Interpretation of Radar Presentations. Forecasters viewing Doppler radar presentations are astonished by the number of transitory features that are not explained by existing conceptual models. Thus, there is a need for a wide range of studies using archived data with the goal to provide better guidelines on proper interpretation of these data. These studies should lead to a better understanding of such events as rapid deepening or filling, tornadoes, heavy rain, gustiness, the relationships between maximum winds observed above the boundary layer and those observed at the surface, etc. In addition, more work is needed to relate reflectivity data and Doppler data so that wind information can be acquired outside the Doppler range.

3.6.3 Data Archiving. Digital recorders are being installed at all DOD and NOAA WSR-88D sites. These data and associated weather event logs will be sent to National Climatic Data Center and made available to researchers (Crum 1995). For coastal locations, analyses of these data should provide considerable insight into tropical cyclone structural change as storms move inland. For island locations, inner-core details might provide clues to such problem areas as genesis and intensity changes. Procedures need to be streamlined to facilitate the transfer of data to researchers with minimum delay.

3.7 Tornadoes. Hurricane-spawned tornadoes are common in landfalling situations, and some general rules have been developed for their occurrence. Forecasters will typically issue tornado watches in the right forward quadrant region of all landfalling storms, especially in regions of onshore flow within about 500 km of the storm center. It is often difficult to distinguish tornadic activity from ordinary gustiness and straight-line winds. The coastal WSR-88D radar network and its associated data archiving should provide considerable observational information on hurricane-spawned tornadoes. This observational evidence should provide for a better basic understanding of the physical processes responsible for these mesoscale events. Although the National Severe Storms Laboratory (NSSL) performs ongoing research in this area, important contributions can be made by the local forecast offices as well.

3.8 Storm Surge. The Sea-Lake Overland Surges from Hurricanes (SLOSH) model (as reviewed by Jelesnianski in WMO 1993) has been used to develop atlases (Jarvinen et al. 1985) of likely storm-surge scenarios for 31 bays and estuaries along the Gulf of Mexico and Atlantic coastal regions, and for Puerto Rico and the Hawaiian Islands. Updating of these models and the atlases, as well as verification of storm-surge forecasts in actual landfalling situations using best-track data, will be a continuing research effort. Best-track verification sometimes suggests a need for refinement in offshore sea bathymetry or inland near-coastal topography used in the model. Also, it might indicate surge sensitivities peculiar to a given basin with complex coastal and terrain features.

The surge from most landfalling tropical cyclones affects 100-200 km of coastline for a duration of several hours. Unfortunately, tropical cyclone track forecasts have errors of a similar magnitude for the crucial 24 hours preceding landfall. Model surge computations cannot provide a valid surge estimate with such imprecision in the track forecast. Thus, improvements in track forecasts, discussed in an earlier section, will directly lead to improved storm-surge forecasts.

Storm-surge models are developed and tested using historical storm data. Storm-surge modelers have noted that the archived tracks of some of these historical storms are incorrect with regard to point of landfall and/or intensity. Another problem is that the tracks of these early storms have been overly smoothed. While this track smoothing is acceptable and even desirable for general forecasting purposes, it is detrimental for storm-surge verification purposes, and these historical data need to be corrected for storm-surge purposes.

Although the SLOSH model uses a parameterized surface wind field, there has been some recent research (NWS and HRD/AOML) into improved SLOSH performance using actual rather than parameterized fields. In fact, the availability of real-time surface wind fields (Houston and Powell 1994; Houston et al. 1995) has prompted comparison of standard parameterized SLOSH model forecasts with the SLOSH forecasts from actual surface wind fields. Because of (1) the "Atlas" concept, (2) the uncertainty in the forecast track, and (3) the necessity of using both observed and forecast wind fields in the SLOSH model, the benefits of using real-time observations and forecasts are limited from an operational perspective. Nevertheless, the possible improvements derived by using real-time observations and by running mutiple custom calculations in real time as landfall situations develop require continued investigation and research efforts.

3.8.1 The Coastal Effects of Reefs on Storm Surge. During Hurricane Iniki, the SLOSH program and MEOW did not accurately predict the height of "storm surge"--total wave height-inundation for Kauai and western Oahu (see section 2.4.2.4 for details). Conflicting ideas as to the value of coral reefs in protecting islands should be examined.

3.9 Sea State. Forecasts/advisories require the specification of the radius of 12-ft seas; however, the centers have little skill in either observing or forecasting sea state related to tropical cyclones.

3.9.1 Estimation. The estimation of the radius of 12-ft seas is quite crude. At the TPC, for example, the quadrant radii of 34-kt winds (itself an estimation) is typically used as the current radius of the 12-ft seas. JTWC is evaluating skill in predicting the horizontal surface wind distribution that is critical to sea-state estimation and other aspects critical to DOD operations. Feedback from this evaluation might require a restructuring of current operational procedures being used at JTWC for determining radii of maximum winds. However, there is no known research involving the estimation of the sea state itself. Sea-state estimation is highly subjective and for decades has been closely associated with the Beaufort wind scale. Improved observational techniques are needed, and satellite sensors that can detect wave height show some promise in providing greater and more consistent coverage. Verification is rarely available, and an applied study that would perform a post-analysis of the sea state for a number of storms is recommended.

3.9.2 Forecasting. Currently, there is no requirement in the forecast/advisory message for a specific forecast of the radius of 12-ft seas. However, a forecast of the quadrant radii of 34-knot winds is required, and an expansion/contraction of these radii would imply a similar expansion/contraction in the radius of 12-ft seas. FNMOC has implemented the Navy's Third Generation Wave Model (WAM) (Wittmann et al. 1995), which provides forecasts of wave heights based on forcing by surface winds from the global NOGAPS or regional NORAPS (Hodur 1987) atmospheric models. Verification scores for WAM are substantially better than those for the previous Global Spectral Ocean Wave Model (GSOWM). Improvements can be attributed both to the skill of the WAM model, and improvements in the skill of the atmospheric models. FNMOC also plans to "loosely couple" the global WAM with NOGAPS in order that the surface roughness predicted by the wave model will be fed back for use in NOGAPS wind-stress calculations.

Current and past research on ocean response to tropical cyclones is reviewed by Ginis (1995). Topics included are: (1) observations of the ocean response to tropical cyclones, (2) theoretical studies of internal waves generated by tropical cyclones, (3) numerical modeling of the ocean response to a tropical cyclone, (4) simulations of tropical cyclone-ocean interaction, and (5) storm surges.

3.9.3 Coastal Effects. Near-shore phenomena such as shoaling, refraction, breaking waves, wave run-up, reflection from shore obstacles, interactions with local currents, etc., are extremely localized and difficult to treat empirically. Also, little is known about the effects of increased wave height on the coastal zone when the storm is 12-48 hours away from landfall. Some of the characteristics of the behavior of coastal waves over and around coral reefs have been described by Guard and Lander (1993) in their adaptation of the Saffir-Simpson Hurricane Scale to the tropical Pacific. These phenomena require specialized local treatment of an engineering nature that need to be researched by local officials when establishing construction set-back lines and building codes. The coastal zone is also a high-interest area for the Navy.

3.10 Information and Data Management. This topic includes a wide range of related issues. In order to perform their duties, forecasters require a vast array of information ranging from raw and processed analog or digital data to local/regional/global analyses and forecasts. These products are always in a state of flux, and there is a danger of providing too much information, to the point of being counterproductive. Ideally, new products would not be sanctioned until fully tested in a research and simulated operational environment. Since tropical cyclones are relatively rare events, forecasters are often asked to evaluate products before operational utility is established. The introduction of new products implies the retiring of older products. Thus, continuous evaluation is needed to keep this process in balance.

3.10.1 Data Requirements. As pointed out earlier, the most important, most visible and most specific user products issued by the operational centers are the TPC and CPHC forecasts/advisories and the analogous JTWC tropical cyclone warnings. Although minor differences between centers exist, in general these forecast/advisory/warning releases require exact specification of initial and forecast tropical cyclone position, intensity, size, etc., through the 72-hour projection. The most critical data requirements should logically relate directly or indirectly to short and long term improvements in forecaster ability to prepare and constantly improve the quality of these messages.

It is essential to systematically validate the measurements required to provide an acceptable and ever-improving level of service and then to develop the optimum mix of remote and in situ systems that would be complementary and would satisfy these data requirements. Also, vast amounts of data are available from satellites, Doppler radar, ASDL (Aircraft-to-Satellite Data Link), etc., but not all of these data are being retrieved or assimilated into initial analyses. For example, it is often possible for forecasters to note discrepancies between synoptic analyses and satellite imagery, such as tropical waves not being reflected on low-level analyses. In data-sparse regions, tropical cyclone motion sometimes appears to be inconsistent with analysis features. Therefore, aside from the need for additional data, there is a critical need to improve data assimilation techniques, especially in data-sparse regions.

3.10.2 Damage Assessment. Damage assessment models provide the means of translating tropical cyclone wind, rainfall, and storm-surge forecasts into damage forecasts. In recent years, there has been renewed interest in this topic and models (Powell et al. 1995; Bruno 1995) have been developed. This is a very useful concept in that it provides utility companies and disaster preparedness agencies with advance knowledge as to damage potential from an existing or historical storm. One weakness of damage assessment models, if used in real time, is that they are very dependent on the exact point of landfall. In this respect, they are similar to storm-surge models that use an "Atlas" (see section 3.8) concept to bypass partially the problem of track forecast inaccuracies. Perhaps similar methodology, the use of hurricane strike probabilities, or the CLIPER model should be examined for use in determining alternate tracks.

Tropical cyclone intensities are often expressed in terms of the Saffir-Simpson Hurricane Scale (Simpson 1974, Saffir 1977, and OFCM 1996) that relates hurricane intensity to potential damage; intensity is expressed as a numerical category that pertains to a range of wind. Table 3.4 illustrates the hurricane categories, wind ranges, storm-surge ranges, and severity description. The scale, with a complete description of wind and storm-surge damage, can be found in Appendix E of the NHOP. The storm-surge heights are approximate since they are dependent on offshore bathymetry as well as storm parameters. The original Saffir-Simpson Hurricane Scale could not be directly applied to the Pacific or to most tropical areas. Guard and Lander (1993) modified the Saffir-Simpson Hurricane Scale for use in the tropical Pacific and in other tropical regions. This modified scale makes extensive use of wind damage to vegetation as well as structures. The scale has potential application in all tropical areas of the globe. Additional research is needed to develop these scales or modify existing scales for use by WMO member countries.

3.10.3 Tropical Cyclone Typing. It has long been recognized that tropical cyclone tracks tend to be repetitive and can be associated with likewise repetitive synoptic patterns (Neumann 1979). Analog track models use a series of computer algorithms to use this principle to associate an existing storm with its family. The concept is similar to fitting to a parent probability distribution and using known properties of the distribution to infer a solution to a given situation. Although computerized analog track models are no longer popular, the analog concept is still valid and work is underway (Carr and Elsberry 1994; Carr et al. 1995) to enable forecasters to identify a synoptic pattern or regime and use known attributes of this regime to assist in the forecast process. A major project on this topic is underway at the Naval Postgraduate School and JTWC. If successful, the technique could be adapted to other forecast centers.

Table 3.4. Saffir-Simpson Hurricane Scale relating a storm intensity Category to expected ranges of wind and storm surge (OFCM 1995).

3.10.4 Objective Aid Evaluation. Statistical and numerical objective aids are in use at all centers. Because aids work well in certain situations and poorly in others, forecasters need to be aware of these model strengths and weaknesses. A problem develops when the attributes of some of the models are not constant. This could be caused by (1) software changes (presumably improvements) in the model or (2) changes in another model that feeds into the model under consideration. A good example of (2) is provided by the statistical-dynamical models that perform well/poorly when the steering flow is not/is influenced by the tropical cyclone vortex in the numerical model (Neumann 1993). Another example relates to the depth of the deep-layer average used in the BAM models (Gross 1991). Accordingly, model performance statistics must be kept up to date, and new methods must be developed for forecasters to effectively use this information.

A related issue is the necessity to retire models when they have lost their operational utility, as well as a need to modify local model software when required because of changes in procedures at the centralized analysis and forecast centers, such as FNMOC or NCEP. Many examples of local model deterioration because of procedural changes at the national centers can be cited, and continued efforts are needed to detect and correct and even anticipate this condition. This also requires that the forecast centers (TPC, CPHC, and JTWC) be keenly aware of the structural characteristics of their local objective aids.

3.10.5 Global Model Evaluation. At FNMOC and NCEP, global models undergo continuous modification. Recent changes include tropical cyclone bogussing and increased model resolution, both of which should improve the overall performance of the model. The three forecast centers also evaluate the output from the global models and closely monitor tropical analyses and prognoses as they relate to their local products. These evaluations and studies should be ongoing.

3.11 Non-Meteorological Items. As pointed out elsewhere in this document, track forecast errors have been decreasing at a rather slow rate whereas coastal population has been increasing at a very rapid rate. Since this effectively increases the tropical cyclone threat, the centers have focused on disaster preparedness as one way of addressing this problem.

3.11.1 Presentation of Information. Some modifications to the tropical cyclone warning systems are made, usually at the Interdepartmental Hurricane Conference and the USCINCPAC Annual Tropical Cyclone Conference. These changes have usually been the result of the meteorologist's perceptions of the needs of customers rather than based on valid investigation of improvements that would benefit response, which should be the basis for future modifications and improvements.

3.11.2 Action Motivation Studies. Southern (1993) pointed out that the strategy behind a tropical cyclone warning is a prime element in an effective warning system. The wording of messages or television graphics should reflect the urgency and severity of a tropical cyclone threat and should motivate people to a desired course of action. There is little scientific evidence to indicate the extent to which warnings are successful in this regard. Systematic studies are needed to determine how best to motivate a nonhomogeneous population at risk to follow various courses of action. These studies should incorporate the skills of social and behavioral scientists. The sociological studies of Bern et al. (1992), in the aftermath of tropical cyclone-02B in Bangladesh, revealed many shortfalls in the warnings that played a role in the poor response of the people to the warnings. As a result of the study, many changes were made to improve the warnings to better motivate the desired actions from the people. Sociologists and others with non-meteorological expertise could play an important role in evaluating current warning methods. Research in this area needs to be supported.

3.11.3 Evacuation Studies.

3.11.3.1 Comprehensive Evacuation Plans. By examining repeated runs of the SLOSH storm-surge model (Jelesnianski 1993) for various coastal basins, local disaster officials are designing comprehensive evacuation plans to be activated during tropical cyclone threats. The number of people to evacuate, areas of evacuation, location, time needed for evacuation, when and how to evacuate, conditions requiring evacuation, etc., are all part of these plans. The ability to accomplish orderly evacuations consistent with the forecast lead time in major population centers is a major problem and continued efforts are needed to design and update these plans.

3.11.3.2 Wind Damage. Disaster preparedness plans have traditionally focused on coastal evacuation in areas where storm surge is the major threat. Recent hurricanes, notably Hugo of 1989 and Andrew of 1992, have focused attention on coastal and inland wind damage. Meteorologists need to work more closely with engineers to determine better methods to infer the maximum wind speed from observed damage. Such information would be invaluable to help fill in the gaps where intensity measurements were not available. A similar relationship needs to be made with botanists to relate the damage to vegetation due to winds. Guard and Lander (1993) have added considerable structure and vegetation damage information in adapting the Saffir-Simpson Scale to the tropical Pacific. Additional research is needed to improve methods to forecast inland wind and to identify inland geographical areas subject to high winds from tropical cyclones.

3.11.3.3 Vertical Refuge. Even after local tropical cyclone coastal evacuation plans have been activated, unexpected situations, such as sudden deepening or track changes, may occur. Under such conditions, people may be unable to evacuate to designated shelters or may even be trapped in automobiles. In response to this possibility, some "last-resort" type of refuge in high-rise buildings is needed. This would require a study for designating public and private buildings suitable for such a purpose.

3.11.4 Communications. Reliable communications are essential to a tropical cyclone warning system. The present communications system has evolved over a number of years in response to forecast center and user needs; however, both incoming and outgoing communications can be improved. Great strides have been made by HRD/AOML in getting real-time aircraft and Doppler radar data into the TPC. Current GTS communications limitations do not allow special observations to be transmitted out of the immediate region--foreign data upon which both TPC and JTWC depend. This prevents most of the maximum intensity data at landfall from being transmitted to the centers. This problem is especially serious at JTWC. In addition, once winds reach certain levels, normal communications are often curtailed until after the winds subside. Alternate methods are needed to insure that maximum intensity information from all locations can be received at TPC and JTWC. Thus, a comprehensive study is needed to evaluate how well the system is performing. Such a study would need to consider the electronic and physical aspects of the current system, the message content, language, timing, etc., on a regional as well as a cultural basis.

3.11.5 Economic Aspects. An early economic study (Neumann 1975) is still being used to estimate the costs associated with raising U.S. coastal hurricane warnings. Neumann's original estimate of preparedness costs for a typical 300 mile length of coastline was $25.1 million dollars per landfall event. Because of inflation and vastly increased population, TPC (Jarrell et al. 1992) now estimates that preparedness costs are close to $192 million per event.

Pielke (1995) has estimated that the annual losses from hurricanes in the U.S. (not including preparedness costs) average about $5 billion. This, assuming 5 hurricane landfalls every 3 years (Hebert et al. 1993), translates into about a $3 billion per hurricane event. Anderson and Burnham (1973) cite estimates by the insurance industry that 15 percent of hurricane damage ($450 million per event) is preventable but that only about 20 percent of population prepares for a hurricane. If we believe these latter numbers, only $90 million (20 percent) of the $450 million is being realized. Except for the annual damage estimate and average annual landfalls, the above numbers come from the 1970's and even then were not well documented. There is no recent research on these values but if they were developed, the $90 million (or its validated equivalent) could be interpreted as the current value of a hurricane warning and the $450 million (or its equivalent) as the potential value of a hurricane warning.

Virtually every cost-benefit analysis relating to hurricane warning service in the past several decades has emphasized the wasted cost associated with over-preparation alone. Much of the emphasis on that component came because it is easily understood and readily estimated. Nevertheless, that component could be an order of magnitude less than the potential increase in prevented damage. Such prevention is possible by restricting building in highly vulnerable areas, by enhancing hurricane resistance of coastal buildings, and by improved public response. The latter will come from better forecasts, particularly at longer ranges, allowing time for action before property becomes a secondary consideration to preserving life. With better basic forecast information comes the requirement for better, more convincing, articulation of the information and communication of the threat to the public. The following questions are key to the long-range direction (health and funding) of the warning service and its supporting research:

• What portion of the damage is preventable?

• What can we do to increase that portion?

• What fraction of the public prepares under certain circumstances?

3.11.6 Media Interface. Forecast centers must rely on radio, television, and newspapers as the primary means for disseminating their warnings. To do this effectively, newscasters, TV meteorologists and weathercasters, and reporters need to be aware of the terminology, schedules, capabilities, hazards, warning procedures, etc., of the forecast centers. Conversely, forecasters need to be aware of problems, schedules, and deadlines faced by the news media. Training sessions that address these joint needs have been organized, but additional efforts are needed to determine if there are better ways to accomplish this information interchange.

3.11.7 Architectural-Engineering Studies. Studies are needed on how best to improve engineering and architecture in aesthetically pleasing structures that can be built at reasonable cost in high wind areas. Such a study might include meteorologists, social scientists, politicians, insurance companies, builders, engineers, architects, prospective buyers, etc.

3.11.8 Disaster Preparedness in General. Practically all of the subtopics under the general heading of "non-meteorological items" relate, in some way to disaster preparedness. A recent study (Pielke 1995), funded by NSF, focuses on the adequacy of current capabilities in that area and discusses ongoing research to meet requirements. Pielke's findings are based on societal responses to the August 1992 Hurricane Andrew event in South Florida. The study lists and discusses the following ten fundamentals of hurricane preparedness:

• Future hurricanes will strike the United States.

• Assuming an average number of hurricanes strike the U.S. each year, the U.S. can expect to suffer annually hurricane-related losses of at least an average of $3 billion.

• Improved hurricane track forecasts have the potential to contribute to reduced vulnerability.

• The U.S. knows how to reduce its vulnerability to hurricanes.

• Coastal communities have great potential to reduce their vulnerability to hurricanes.

• Hurricane impacts highlight the level of societal preparedness.

• To reduce vulnerability, coordination and communication among different groups is essential.

• Society's complacency about hurricanes is an obstacle to improved preparedness.

• The impacts of a hurricane depend greatly on where it makes landfall; for example, the impacts of Hurricane Andrew in south Florida could have been much worse.

• The impacts of a hurricane depend greatly on a community's vulnerability; for example, with better preparedness, the impacts of Hurricane Andrew in south Florida could have been reduced.

3.12 Summary of Needed Research. Table 3.5 is a summary of the research items discussed in section 3.

Chapter 2
Chapter 4
Table of Contents
Publications List
Home