2.4 Methods and Applications of Thunderstorm Nowcasting
2.4.1 Scale Dependencies
The nowcasting of thunderstorms is complicated by the fact that multiple-scale forcing mechanisms often are operating at the same time. There is a mutual interaction between extra-tropical synoptic-scale processes and severe convection (Doswell and Bosart, 2001). Whereas synoptic-scale processes provide a setting in which severe convection develops (Johns and Doswell III, 1992; Barnes and Newton, 1983) the idea becomes accepted that convective processes themselves influence synoptic-scale processes via mesoscale processes. Although there is no universally accepted method to separate scales of mo- tion, all approaches have one basic observation in common: the lifetime of a meteoro- logical feature, like for example the rain field, is dependent on the scale of the feature, with large features evolving more slowly than small features. With this knowledge the dynamically motivated approach to divide scales (Emanuel, 1986; Doswell III, 1987) seems more appropriate than the scale division based on powers of 10 in space and time (Doswell III, 1987).
Like the question about meteorological scales, no universally valid definition for weather forecast regimes exists. In contrast to meteorological scales, weather forecasts are commonly defined based on the time scale of their prognoses only. Since temporal and spatial scales are related, the spatial aspects of the forecast targets are automati- cally incorporated in the forecast regimes. According to Wilson et al. (2004) and also in use at the DWD the term nowcasting is used for prognoses with a time horizon up to 6 h, whereas prognoses on longer time scales are called forecasts. Very short-term nowcasting refers to prognoses with forecast times up to 2 h. Adapted to the outlines of this work very short-term nowcasting will be used for thunderstorm prognoses up to 30 min.
Certainly the most important question in the field of forecast research is the general question about the predictability of the atmosphere. An extensive study series (Ger- mann and Zawadzki, 2002) to (Germann et al., 2006) addresses to this question from storm to synoptic scales. Some crucial observations strongly influence current forecast research. Based on continent-scale radar composite images they used the lifetime of radar reflectivity patterns in Eulerian and Lagrangian coordinates as a measure of pre- dictability. An important finding has been that the range of predictability increases with increasing scale (Germann and Zawadzki, 2002). They also found that advection explains a significant part of the variation of the precipitation rate at a given location favoring a Lagrangian-persistence approach over an Eulerian approach to extrapolate field patterns (Germann and Zawadzki, 2002). In their Lagrangian persistence approach they identified two sources of forecast uncertainty (Germann et al., 2006). One source is the permanent growth of precipitation and the other is the change in the storm mo- tion field. They found that the contribution of changes in the motion field to forecast
Figure 2.12: Scheme of precipitation predictability versus forecast time (taken from Germann et al. (2006)).
uncertainty is small in comparison with the fluctuation of precipitation.
In their study (Germann et al., 2006) they present a forecast scheme which concep- tually illustrates the skill of different forecasting techniques as a function of forecast time. The scheme is given in figure 2.12. The envelope, thick dashed, curve is achieved if the best forecast technique is taken for every forecast time. But due to the problems nature the estimate will never be above the exact predictability. Nowcasting techniques are usually focused on analyses and extrapolation of the trend of a single variable on smaller scales, e.g. radar based rainfall distribution. Models resolve the larger, slower evolving scales, and local details are filled in by parameterisation or statistics. Never- theless an unavoidable downward trend in information content exists with increasing forecast time due the fact that the atmosphere is a complex, non-linear system. Follow- ing figure 2.12 shows the well observed characteristic that for extrapolation methods the initial information capture is close to perfect. But because the method lacks information about atmospheric physics the information is lost rapidly with increasing nowcast time. In contrast thereto the limited resolution and the imperfect assimilation algorithms let models poorly represent the observed state, whereas the information loss with proceed- ing forecast time is not so large for the earlier part of the forecast. And finally no forecast method is able to assess higher skill than the long-term average.
It can be seen in figure 2.12 that there exists a theoretical intersection, when the infor- mation content of a model exceeds the information content of a nowcasting method. The scheme implies that two points have to be addressed in order to improve the nowcasting of weather features to the theoretical possible upper limit. First the nowcast techniques covering the different nowcast scales, like trend extrapolation and models, have to be brought to perfection and second the cross point has to be determined as exactly as possible. This is a complex and delicate issue since both points interact. It depends on the possibilities of the employed nowcasting methods and on the measures used to quantify the information content of the nowcasts. But finally and most important the
method depends on the principal nowcast objectives and the target objects the nowcast tool is designed for.