Status:Ongoing analysis Material from: Linus, Ahmed, Rebecca, Tim H
1. Impact
A severe weather event that affected the Maldives on Sunday 31st December. A low pressure area formed west of the Maldives late 30th Dec and resulted in very heavy rainfall over the country, especially the central region and capital island Male'. Hulhule station (VRMM), Central Maldives, reported a total of 179.7mm within 24 hours for 31st December 2023. This broke the previous record of 175.9mm on 23 Dec 1977. The main city island Male' was completely flooded and was dewatered using pumps. National Disaster Management Authority reported 140 households were assisted during the flood response and 42 people from 8 households are currently in temporary shelters.
2. Description of the event
The evaluation will focus on the 1-day rainfall on 31 December inside a 1x1 degree box centred on 4N, 73E.
Satellite picture below showing the convective setup.
The timeseries below shows the observed rainfall over the main city (Male).
The measured amount between 0 and 12UTC on 31st was about 178mm.
3. Predictability
3.1 Data assimilation
3.2 HRES
The plots below show 24-hour precipitation (31 December 00UTC - 1 January 00UTC) in concatenated short forecasts (first plot) and ENS control forecasts with different lead times. The box marks 0.5x0.5 degree centred on Male'. While all forecast plotted here had some rainfall, it was first the shortest forecast (from 31 December 00UTC) that captured the severe rainfall.
3.3 ENS
The plots below shows EFI and SOT for 1-day precipitation (31 December 00UTC - 1 January 00UTC) from different initial dates. An early signal was present for having a wetter than normal day but it was only the shortest ensemble that indicated the extreme around Male.
The plot below shows the forecast evolution plot for 24-hour precipitation (31 ecember 00UTC - 1 January 00UTC) a 0.5x0.5 box centred on Male. Concatenated 6-hour forecasts - green dot, ENS control –red, ENS blue box-and-whisker, Model climate – cyan box-and-whisker. Ensemble mean as black diamonds. Triangle marks the maximum in the model climate based on 1800 forecasts. While the wetter-than normal signal is visible also here, the extreme was only captured in the last forecast.
ecPoint
Figures in the table below illustrate ecPoint performance for this case.
On the CDFs we see the standard wet tail elongation that characterises ecPoint post-processing. X-axis values for the blue ecPoint line show that at these short-medium-range lead times rainfall of order 178mm/12h was low probability, but not out of the question. Bias correction at the gridbox scale (green versus red lines) acts to slightly reduce the totals, except at very short leads when it does little to change them. For large rainfall totals in the forecast one tends to see more of a bias correction to lower values. The mitigating factor here is very probably the consistently rather high CAPE values (denoted by values of typically 3 in the fourth digit of the weather types on the CDF tables). Meanwhile one can see that the model is diagnosing a mix of convective and large scale rainfall; in some members it is much more convective, in others it is much more large scale. This is denoted by the values 4 and 1 respectively in the first digit of the weather types. Mixed types with high CAPE also tend to exhibit less of a reduction through bias correction.
On the map plots, showing probabilities of gauge measurements exceeding 100mm/12h, we see that as the event approached the signal slowly but surely increased. And there is not much jumpiness. It would be wrong to criticise the forecasts because the highest probabilities are not over Male. This all suggests that these products could have been used as the basis for early warnings of extreme rainfall in parts of the Maldives. One caveat is that we have not looked at a lot of cases - if such signals occur regularly in this area the false alarm count would be high. Nonetheless global verification suggests the probabilities should be fairly reliable. Note also that not one ENS member showed >100mm/12h at Male in the raw model output. This also highlights the benefits, for localised extreme prediction, of ecPoint post-procressing.
A final point to note is that the final forecast, for T+0-12, could not have been used for actual forecasting or warning issue because it would become available too late. It is included just for completeness.
| DT & lead time: | 0Z 28th (T+72-84) | 12Z 28th (T+60-72) | 0Z 29th (T+48-60) | 12Z 29th (T+36-48) | 0Z 30th (T+24-36) | 12Z 30th (T+12-24) | 00Z 31st (T+0-12) |
CDFs for 12h rainfall for Male: Valid 00-12UTC 31st when about |
|
|
|
|
|
|
|
Probabilities of point rainfall (as measured by a gauge) exceeding 100mm in 12h, from ecPoint: X on the first plot is Male. |
|  |  |  |  |  |  |
3.4 Monthly forecasts
The plot below below shows the weekly mean precipitation anomaly for 25 December - 31 December. All lead-times here showed a wet anomaly around 10-15N, and the forecast from 16 Decmber had already a very strong anomaly.
Also the seasonal forecast from 1 November predicted a wet season for the region around the Maldives.
The plot below shows MJO forecasts from 31 December and 26 December. A strong MJO event was present in the western Indian Ocean around the day of the extreme.
The plot below shows tropical wave identification in the analysis OLR on 31st December, also indicating a large area of convection associated with the MJO (purple) and with a large Kelvin wave (grey & dashed), alongside total 24-hour precipitation.
The plot below shows the tropical wave identification in the extended-range control forecast produced 1 week ahead on 24th December, indicating the convection associated with all 3 wave types in the area. The HRES forecast did not contain any MJO signal at this lead time. Extended-range control forecasts produced before 20th December did not show any MJO signal in this area.
The plots below show the wave-filtered OLR MJO signal from the tropical wave identification, for the forecasts produced on 4th January (showing analysis fields for time of the event), and for 25th December (right)
3.5 Comparison with other centres
4. Experience from general performance/other cases
5. Good and bad aspects of the forecasts for the event
- Early signal for wet weather, probably related to the MJO
- In raw model output the extreme around Male appeared only in the last forecast before the event
- ecPoint post-processed output highlighted the potential for an extreme event several days in advance, and maintained and slowly grew that signal as the valid date approached.