The Figure 2 gives an example of an ensemble forecast for the fog event during the night of 14-15 October 2005. As can be seen in the upper left panel, the cooling during the night is predicted rather good, but because the modeled fog disappears around 0800 UTC the temperature rises too fast in the morning. Note how the temperature forecasts from different members slowly diverge. The probability for a liquid water content above 0.01 g kg−1 is over 70 % and indeed fog formed that night. The timing however was not perfect according to the visibility observations shown in the lower right panel.
Systematic verification (false alarm rate and hit rate) of the model results suggested that the exnsemble prediction scheme has skill that would be valuable as an input for the aviation forecaster.
Having in mind that the base effect is of great importance for fog events (a rare phenomenon) quite satisfying results were found. In Figure 3 the verification results are displayed for different thresholds of observed visibility. It becomes evident that inclusion of advection significantly improves the forecast, here derived from the gradients and wind.
In conclusion, the numerical model ensemble is able to increase the discrimination up to a hit rate of 60 % with a false alarm rate of 30 % or for the 2100 UTC initialization to a hit rate of 80 % with a false alarm rate of 45 %. It has to be noted that these forecasts are purely machine based, and if the model results were interpreted by a human forecaster, even better skills could be achieved.
A case study outlines the good performance of the 3D fog model and also the potential of sophisticated satellite cloud products, which can be used for verification purposes. However, for conclusive skill assessment a whole season needs to be simulated and analyzed, which will be part of future research.
A case study (Figure 4) outlined the good performance of the 3D fog model and also the potential of sophisticated satellite cloud products, which can be used for verification purposes. However, for conclusive skill assessment a whole season needs to be simulated and analyzed, which will be part of future research.
Contact:
Dr. Christoph Schmutz, Leiter Stab Wetter
Figure 2: 1D ensemble prediction of the fog event from 14-15 October 2005. The first two panels show computed temperature and humidity at 2 m height for each member (thin lines), the ensemble mean (thick colored line) as well as the corresponding observations (thick gray line). In the lower left panel the ensemble mean liquid water content is contoured together with observed (thick gray line) and modeled wind speed. The last panel indicates the probability that a liquid water content of 0.01 g kg?1 is exceeded. Figures were taken from the developed semi-operational system.
|
Figure 3: ROC for fog occurrence in the 1D ensemble for different types of computed advection; ADV = pure advection, d/dt = total rate of change in profile, CONTROL = no advection. Visibilities below 500 m, 1000 m and 1500 m were used as observational thresholds of an observed fog event, respectively. Numbers above the symbols indicate the ensemble forecast probability in percent that has to be exceeded in order to be classified as a modeled fog event. Az indicates the area under the ROC.
|
Figure 4: Spatial distribution of fog as seen by satellite (black shaded areas) and simulated by the 3D fog model (colored areas). Contour lines of modeled liquid water have a spacing of 0.1 g kg−1.
Blogger
del.icio.us
Digg
Facebook
Google
Bookmarks
Linkarena
Mr.
Wong
MySpace
StudiVZ
Twitter
Windows
Live
Xing
Yahoo
Bookmarks