...
The next figure shows the 3-day accumulated precipitation in the box 44N-45N, 18E-20E from different initial dates (from different resolutions). In the deterministic forecasts a hint of the extreme rainfall started to appear on 6 May. The T319 forecast was very consistent from 6 May 12UTC (probably with a large portion of change). We do not yet have any good gridded precipitation data set, but from the synop stations in the area the observed precipitation for the period should be 170-200 mm.
3.3 ENS
On 14th May in Serbia and Bosnia heavy rain between 40 and 130 mm just in 24h was recorded. From the few stations in Serbia displayed on the plot below we can see that 24-hour accumulations exceeded the 99th percentile of the observed 15-year climate. The most extreme precipitation was observed in Loznica, a station in western Serbia at the border with Bosnia where 129 mm were recorded on 14th May whilst the climate maximum is just 68.4 mm. Comparing the forecast with observations we can notice that 7-days in advance the tail of the forecast consistently predicted extreme values of the rainfall but the forecast was uncertain and therefore EFI values were positive but not high whilst SOT was positive indicating a possibility of an extreme event. Approaching the day of interest forecast CDFs became more extreme ending up with really extreme CDF in the most recent forecast on the day of the extreme rainfall. We can also notice that the forecast underestimated the rainfall totals even for the most extreme members but at the same time such an extreme rain as observed in Loznica seem locakized. All the other stations around measured rainfall amounts between 40 mm and just above 100 mm which is close to the rainfall in the most extreme ensemble members.
...
The plots below show CDF for 3-day precipitation 13 May 06UTC to 16 May 06UTC for Loznica (point for 44.3N, 19.1E), for different centres. The observed value for the period was 213 mm. Here UKMO outperformed all other centres. NCEP seems to have less spread for this point than the other dentres.
Gallery includeLabel tigge_cdf_3d_loz sort comment title TIGGE CDF for Loznica (3-day)
...
The plot below shows the CDF for 14 May 06UTC to 15 May 06UTC for Beograd (left 108 mm observed) and Loznica (right, 110 mm observed). On this short lead time only NCEP got high precipitation amounts for Beograd, while UKMO had a bad forecast (due to model spin-up or something else?). For Loznica CMC had the worst performance.
...
To conclude, the results are quite mixed comparing different centres. The outstanding results are the high precipitation for 3-days from UKMO for Loznica.
The figure below shows the evolution of probabilities for >60 mm for the 3-day period 13 May to 16 May for the box 44-55N, 18-20E for ensemble forecasts from different initial dates (x-axis). We have to bear in mind here that one cannot evaluate a probabilistic forecast for one event, but the plot gives an indication of the strength of the signal. For the shortest forecasts, UKMO has the strongest signal, as expected from the CDF above. This could be due to a wet bias in their model, but it has to be evaluated further. NCEP had the bad forecast for 12 May 00UTC, which is also present in the ensemble.
ECMWF started to pick up a signal on 6 May together with UKMO. UKMO had a drop in the probabilities the day after, connected to a bad forecast on the European scale. One thing worth to mention is that all but one ECMWF HRES, control and T319 forecasts (as seen in HRES section) had the event from 8 May 00UTC and onwards (total 30 forecasts, 97% probability), but the ensemble probabilities for all but the last ensemble (13 May 00UTC) had probabilities below 80%. It could be interesting to investigate if it could be diagnosed further.
3.6 EFAS
Gallery includeLabel efas_severe sort comment title EFAS
...
- Good early detection of the large scale cut-off low and extreme precipitation (in EFI)
- Early response in the EFAS system
- Precipitation amounts clearly underestimated for Beograd. ECMWF worse than other centres.
- Only the very last forecast captured the amount over Beograd.
- ECMWF ensemble too diffuse for this case?
6. Additional material
Information sheet from JRC/EFAS 16 May
...