Strive for Better Typhoon Forecast
- Thursday, 11th August 2016
In recent years, whenever there is a tropical cyclone over the western North Pacific and the South China Sea, the public is not only interested in the forecast tracks issued by the Observatory but also those from the other meteorological centres. Their performance will also be compared after the event, usually by comparing the deviations of forecast positions of tropical cyclones from the actual positions. However, we have to take note of sample collection when performing such kind of verification. This is because for the same tropical cyclone, different meteorological centres may start to issue the forecast tracks at different time and update the forecast at different intervals. As a result, the number of forecasts produced by different meteorological centres (i.e. number of samples) can vary largely. It is therefore very important to pay more effort in collecting and choosing the samples if one wants to compare the tropical cyclone forecasts issued by different meteorological centres.
A more objective approach is to use the same set of data for verification. For example, we can select forecasts issued at the same time and in the same area for comparison. The table below shows the average errors of tropical cyclone forecast tracks issued by different meteorological centres in 2015 using the same set of samples (forecast tracks based on observation time at 00, 06, 12 and 18 UTC for tropical cyclones within the area bounded by 7oN and 36oN, 100oE and 140oE). The track error refers to the distance between the forecast position and the actual position obtained from the Observatory's post-analysis. On average, the 24-hour track errors of different meteorological centres range from 60 to 70-odd kilometres (roughly the size of Hong Kong), and the track errors increase gradually to more than 300 kilometres (roughly the distance between Hong Kong and Shantou) for 120-hour forecast. For those cases where the difference in track errors of meteorological centres was only a few or 10-odd kilometres, it is indeed hard to tell which one performs better as the analysis of actual positions itself also involves an uncertainty of about 30 kilometres on average.
Verification of tropical cyclone tracks issued by different meteorological centres in 2015 using the same set of samples
The Observatory always strives for betterment in forecast techniques. Taking the tropical cyclone warning for shipping issued by the Observatory as an example, improvement in the forecast can be seen in the long run. When 72-hour track forecast was first introduced in 2003, the average error was about 300 kilometres. In recent years, it diminished gradually to around 200 kilometres. One of the reasons of such improvement is the advancement of computer forecast models. Among them, the European Centre showed a better performance on average in recent years. However, such good track record does not guarantee the best performance in the future. Weather forecasters therefore still have to assimilate various kinds of information and learn from past experience, so that a professional judgement can be made.
C.W. Choy & L.S. Lee
 UTC (Coordinated Universal Time) = Hong Kong Time - 8 hours
 Tropical cyclone warnings for shipping are issued by the Hong Kong Observatory for tropical cyclones located within the area bounded by 10oN and 30oN, 105oE and 125oE. Samples of verification include warnings based on observation time at 00, 06, 12 and 18 UTC.