Methods to predict severe weather

Gerd Tetzlaff
Institut für Meteorologie, Universität Leipzig, Stephanstr. 3, D-04103 Leipzig, Germany.
Tel.: +49-341-97-32850; Fax : +49-341-211-0937;
E-mail: tetzlaff@rz.uni-leipzig.de

Severe weather is assumed to be capable to do damage to man and nature. The occurrence of such weather is restricted to statistically rare events. What frequency range is comprised usually depends on the amount of damage occurring in the course of such an event and is not well quantified. The prediction time span has a fairly wide range. A prediction everything is called that allows to exceed the accuracy of the predictions produced by trivial methods, usually achieved either by persistency or by climate predictions. Mostly, the available data bases allow to produce both a persistency and a climate prediction. The available methods limit prediction to time intervals between about one minute to about one year. Furthermore, it has to be differentiated between the methods to produce standard types of predictions and the particular prediction of severe weather. It is assumed that severe weather means the occurrence of high amounts of precipitation or high wind speeds, or also lack of rain or other long-term shortages. All prediction methods use two basic approaches, the deterministic numerical weather prediction and the statistical approach, and in some cases the combination of both of them. It is known that the quality of the severe weather prediction exhibits lower skills than the standard predictions.

The short-term prediction or now-casting is placed in the time range from one minute to about three hours. Within this range the numerical weather prediction is not of the highest quality because of the necessary initialisation process, which costs some assimilation time. Therefore, statistically based methods are used for prediction purposes. However, these methods require a sufficient data base, both for the formulation of the prediction and the evaluation of the quality of the predictions. In the field of nowcasting there is rather little particular experience available allowing to tackle severe weather prediction, because these are statistically rather rare events. Most of the statistical methods are designed to succeed for cases that are close enough to the median values. In principle, this is inherent in the statistical methods as such. Some approaches allowing the treatment of extreme events still cannot overcome the general shortage in the data coverage in this range of events. Therefore, it has to be assumed that the methods available for statistical short-term prediction often cannot satisfy the needs of short-term severe weather prediction.

The procedure to predict severe weather in the time range from about 6 hours to about 10 days is based well on the methodology of the numerical weather prediction. This time range marks the edges of the time span that allows to exceed prediction quality compared to the predictions based on trivial methods. Therefore, the numerical weather prediction as performed by the weather bureaus gives the basis for the severe weather predictions. For better performance these methods rely on the additional application of statistical methods such as model output statistics or Kalman-filter techniques. These statistical methods indeed allow a more reliable prediction and contribute to overcome some of the shortcomings of the pure numerical products. This applies in particular as severe weather still shows patterns of subscale processes. The major problems of the numerical processes are in the vertical and the deduced horizontal exchange mechanisms, becoming apparent in the hydrological cycle. Improvements in the predictions of severe weather have thus to address the description of these processes, however, besides all the other contributors to the inaccuracies in the numerical products, such as the whole data situation, both for the initialisation and the assimilation.

Recently quite impressive progress was achieved in the seasonal forecasts over periods of several months. However, there is still no physical link in these numerical products to severe weather. Therefore, it is a challenge to be met in the future.