Home | About Us | Contacts | Help | Feedback | SEARCH Global | Australia | NSW | Vic. | Qld | WA | SA | Tas. | ACT | NT | Ant. | Weather & Warnings | Hydrology | Climate | Numerical Prediction | About Services | Learn About Meteorology | Registered User Services |

Global Guide
Main Menu

Global Guide to Tropical Cyclone Forecasting:
CHAPTER 3: TROPICAL CYCLONE MOTION

3.4 MAKING THE POSITION FORECAST

To conform with the warning cycles used in most offices (Chapter 7), we divide the forecast period into two: Extended range forecasts of greater than 36 h and short range forecasts of up to 36 h. We illustrate the discussion with examples from the 1991 season in the western North Pacific, obtained from JTWC (1992).

Reports at IWTC-I and IWTC-II indicate that current practice varies considerably in the length of forecast periods employed, which range from 24-72 hours. We strongly recommend that all forecast offices make 72 hour forecasts, at least for internal use. The forecasts at this time period provide valuable support to longer-term planning and help with determining the shorter-term forecasts. Further, an archive of 72 h forecasts provides valuable information for judging the improvement of forecast skill.

Figure 3.15: Mean forecast errors relative to CLIPER for selected forecast aids used by NHC during 1976-1979 (Neumann and Pelissier, 1981b).

Caution does need to be used with 72 h forecasts. Once made, the forecast team is committed to a specific scenario, which may be difficult to maintain when a tropical cyclone is approaching a bifurcation. Some conservatism and reluctance to change this scenario must ne maintained to prevent the "windshield wiper" effect of rapidly changing backwards and forwards between potential forecast scenarios. Such windshield wiper forecast changes are a cause of major confusion and loss of confidence amongst forecast users.

Each forecast period requires special strategies and uses different forecast techniques, the relative performance of which is indicated in Fig. 3.15. Care must be taken to account for the degree of uncertainty in the forecast at each time period. Historical mean forecast errors provide a first order indicator of this uncertainty, but forecasters also should be aware of the type and degree of scatter in these forecasts. As shown in Fig. 3.16, the error distributions at all time periods are significantly skewed. A few large errors have a marked effect on the mean, which is consistently larger than the median.

These large errors typically come from erratic cyclones, or those that exhibit sudden changes of direction or speed (Chen, 1988). This has been shown objectively by Neumann (1981), who ranked one-thousand 48-h forecast errors over the North Atlantic according to forecast error. He found that the 10 cyclones with the lowest error followed smooth tracks, but that the 10 with the highest error were quite erratic. The average error in the erratic group was about twice that of the those with smooth tracks. Neumann (personal communication, 1993) finds that erratic movement occurs in all basins, with the Australian region being the worst. The difficulty with such track forecasts has lead to the use of terms such as "unusual" to describe them. We note, however, hat smooth, steady tracks are in the minority and that such "unusual motion" is more usual than implied by its title.

Figure 3.16: Frequency distributions of analysis and forecast errors for JTWC forecasts in the western North Pacific during 1991 (JTWC, 1992).

3.4.1 Extended Range Forecasts (36-72 hours)

This period is used mainly as an alert and preparation guide, and to provide an initial envelope for determining the shorter period forecasts. Dynamical and statistical techniques dominate (Fig. 3.15) and the use of probabilistic forecasts can be a very useful aid to objective planning.

Since the cyclones are normally well out over the ocean, most position analysis is by satellite. Fortunately, however, there is less relative sensitivity to initial position error than at shorter periods. Care needs to be taken to smooth out small-scale oscillations that have little impact on the longer term track. However, sensitivity to initial analysis conditions can lead to either damping or strong amplification of initial analysis errors in numerical models.

The major forecast errors arise from misinterpretation or poor forecasting of the evolving situation and are most common near recurvature or other expected, or actual, sharp direction or speed changes. Expected changes in direction are responsible for a similar proportion of major forecast errors to those of unforecast track changes (Fig. 3.17). For Typhoon Yunya (Fig. 3.17a), JTWC missed the recurvature, leading to two major errors in set of otherwise quite good forecasts. The very next typhoon, Zeke (Fig. 3.17b), was expected to recurve over Hainan Island, leading to a series of incorrect forecasts.

Figure 3.17: Illustration of major forecast errors: a) missed recurvature of Typhoon Yunya, b) expected recurvature of Typhoon Zeke. The figures contain all 72 hour forecasts from JTWC overlain on the cyclone track (JTWC, 1992).

There is no easy solution to this large forecast error problem, but considerable care in analysis and selection of appropriate forecast aids can help. This requires that forecasters have an intimate knowledge of the details of the available techniques and of their strengths and weaknesses in different circumstances. This can be helped by regular, objective operational evaluation of the available techniques, for example, Neumann and Pelissier (1981) and the Annual Tropical Cyclone Reports of the JTWC. Also, it is possible to combine several forecast techniques in a statistically optimal manner to produce improved forecast guidance (Leslie and Fraedrich, 1989). Mundell (personal communication, 1993) also has developed a technique for optimal blending of forecast aids at JTWC.

An example of avoiding bias, or errors in a particular techniques is provided by the poor climatological forecasts from Typhoon Ellie in Fig. 3.18. Following a potential recurvature point (arrowed) the typhoon moved consistently westward to the south of the subtropical ridge. The climatology at this high latitude is dominated by northward moving and recurving storms, as is clearly seen in Fig. 3.18. When such anomalous tracks can be identified, the climatological techniques can be disregarded in favour of dynamical techniques that are better able to resolve the situation.

Figure 3.18 also indicates another standard bias in climatological techniques. Notice how the forecasts as the cyclone approaches China tend towards stronger and sharper recurvature and keep the cyclone off the coast. This arises because westward moving cyclones decay over land and the longer range forecasts are dominated by cyclones that moved along, or away from the coast.

Figure 3.18: Series of 72 hour climatological forecasts for Typhoon Ellie (JTWC, 1992).

3.4.2 Short Range Forecasts (less than 36 hours)

The 12-36 hour period is used mainly for establishing warning zones and initiating civil defence procedures, such as evacuation of coastal zones. Persistence, climatology and statistical techniques become more important relative to the dynamical techniques. The causes of major forecast errors described in the previous section apply at this range, but with less impact. Moderate sensitivity to initial position error is experienced, and care needs to be taken applying the position analysis recommendations in Section 3.2. Often tropical cyclones are nearing, or within the range of shore-based radar, with attendant analysis improvements and pitfalls.

Within 12 hours, coastal communities normally will be experiencing heavy rain, and strong to gale force winds. The forecasts are geared towards fine-tuning coastal zone evacuation and damage amelioration decisions. Persistence dominates at this short forecast period and considerable care needs to be taken with position location and correctly analysing short-period oscillations of the eye.

Recommendation: Considerable attention must be taken with maintaining the best possible track of the tropical cyclone. In particular, great care should be taken to avoid misinterpretations with of recent positions and trends in the cyclone track.

For example, short-period oscillations can cause considerable uncertainty, or error in the forecast track if they are confused with longer term trends. This is illustrated in Fig. 3.19, for a cyclone moving at 5 ms^-1 with a small-scale trochoidal oscillation of amplitude 14 km and period 12 hours, which produces a cyclic speed variation of 2 ms^-1.

Figure 3.19: Illustration of the effects of incorrect interpretation of a small-scale cyclic oscillation of amplitude 14 km with period 12 hours for a cyclone moving at 5 ms^-1 and 25 hours (450 km) from the coast: a) cyclone moving directly towards the coast, b) cyclone moving at an angle of 30^o to the coast.

For a cyclone 25 hours (450 km) from and moving directly towards the coast (Fig. 3.19a), the resultant forecast errors would be ±180 km along the coast with no significant difference in landfall time, or between 4 h early and 6 h late at the forecast landfall point. The along track error is acceptable, provided sufficient evacuation time has been allowed. The cross-track error places an additional 200 km uncertainty in landfall forecast location and introduces a potential for poor warnings, but is still within normally acceptable ranges. In this case the along- and across-track errors are decoupled, neither has any direct effect on the other. For the same cyclone approaching the coast at an angle of 30^o (Fig. 3.19b), the errors become coupled and the uncertainty increases dramatically. Forecast landfall could be 10 h early and 210 km to the right of the actual landfall, or 67 hours late and 1400 km beyond this point. Thus, an incorrect interpretation of the oscillation could have the cyclone arriving ashore 67 hours ahead of and 1400 km away from the forecast landfall point!

Contents Chapter 3.5