...
2. Description of the event
The evaluation of the cold temperatures rainfall will focus on 24-hour rainfall on 16 April 00UTC - 17 Aprl 00UTC in a 0.25x0.25 box centred on Dubai (centre point 25.2N, 55.3E).
...
The plots below show analyses of T850 and z500 from 30 March 13 April to 5 17 April.
3. Predictability
...
Signs of a rainfall extreme was seen from 6 April - 10 days before the event in the ensemble (median passed the model climate maximum).
ecPoint
16th April
Below are show 95th percentile values of 12h rainfall from ecPoint, for global view, to put the Middle East forecasts for 16th, when, in particular, Dubai/UAE were hit, into perspective for that day. The local view is also shown, to highlight absolute values. Note first two things: 1. these are for a specific 12h period (12h being the ecPoint default at the moment) of 06-18UTC, whilst others have generally quoted 24h rainfall values (the event lasted more than 12h overall I think), 2. the 95th percentile is shown (so a 1 in 20 exceedance prob for a given site), but any integer percentile value could be referenced (from 1 to 99), or indeed I could have shown exceedance probabilities.
These plots re-emphasise the stunning long range predictability for the event, at least in the ECMWF IFS. At 8 days out not only was this region clearly seeing the strongest signal for heavy (point) rainfall anywhere in the world, but this is in a normally arid area. These patterns were evidently maintained as the event approach, with absolute values of the percentiles going up somewhat.
Day 8 ecPoint forecast (95th percentile) for UAE floods (16 April)
Day 4 ecPoint forecast (95th percentile) for UAE floods (16 April)
Day 2 ecPoint forecast (95th percentile) for UAE floods (16 April)
As above row, but zoomed in versions
14th April
Regarding events of 14th, which caused 19 fatalities in Oman (compared to 1 reported for the UAE event on 16th) the plots below show the ecPoint 99th percentile from leads of 8, 4 and 2 days in advance. We don't have many measurements, but very dramatic floodwater videos, and data we do have, suggest that 100-500mm very probably fell in a short period of time in hills/mountains north of the crosshairs marker on the plots. Here there is again a signal well in advance (though at a lower level than for 16th events given that this is the 99th percentile). This signal grew markedly in forecasts from 4 to 2 days in advance. For comparison, the last plot below shows the equivalent 99th percentile from the raw ENS output (=max in the ENS here). As is usual for a high percentile, raw ENS values are generally notably less than ecPoint, and much more noisy. Just locally the raw ENS can give a much higher value than ecPoint, if for example there is a marked outlier, as shown by the CDF example.
Day 8 ecPoint forecast (99th percentile) for Oman floods (14 April) - crosshairs denote the centroid of the casualty reports (the wadi rises northwards from here)
Day 4 ecPoint forecast (99th percentile) for Oman floods (14 April)
Day 2 ecPoint forecast (99th percentile) for Oman floods (14 April)
Day 2 RAW ENS forecast (99th percentile) for Oman floods (14 April)
CDF example for where ENS > ecPoint for 99th percentile.
3.4 Monthly forecasts
3.5 Comparison with other centres
...