Madeira Flooding (25 Dec 2020)
Intensive and persistent precipitation, caused by stationary flow anchored to orographic effects, led to floods and landslides on the north coast of Madeira Island during Christmas 2020, resulting in significant impacts on the local population and heavy financial costs. Weather radar observations recorded precipitation estimates of up to 500 mm within 24 hours, near Ponta Delgada and Boa-ventura (north coast), while values of 222.7mm were also observed in Porto Moniz, near the northwestern coast of the island.
As reported by IPMA [1], between December 24th and 26th, 2020, several mountain convergence processes [2] were established over the island (and its surroundings), resulting in the persistent anchoring of convection to the orography, in a quasi-stationary process driven by the orography interaction with the East/Southeast unconditionally unstable atmospheric flow associated with the upward branch of a depression valley. Mesovortices were identified, locally intensifying the upward currents. One should note that the dynamical or convergence locus (identified by the local weather radar) were not necessarily co-located with the orography features.
[1] Moreira, N. et al., 2023, Precipitação forte e persistente na costa norte da Ilha da Madeira, Internal Technical Report, IPMA.
[2] Kirshbaum, D.J., Adler, B., Kalthoff, N., Barthlott, C., Serafin, S., 2018, Moist Orographic Convection: Physical Mechanisms and Links to Surface-Exchange Processes., Atmosphere, 9(3):80.
The reference runs were performed using CY43T2_bf11 of AROME Dynamical Adaptation from ARPEGE, at 2.5km resolution with 60 vertical Levels.
We then used DEODE (OD-DT), v0.4.0, to run at a finer resolution of 500m with 90 vertical levels using CY48T3 of AROME from IAL package, using IFS-HRES as coupling model. In both experiments no DA was applied.
Both models were run on ECMWF ATOS machine.
The reference runs with AROME CY43T2 at 2.5 km resolution used the operational domain MAD delimited by the 34.75ºN and 31ºN paralells and by the 19ºW and 14.8ºW meridians (orange shaded area on the panels below).
The DEODE Runs were performed on a smaller domain focusing on a region around the archipelago of Madeira centered at latitude 39ºN and longitude 8.5ºW with an area of 1080x960 square cells of 500x500m, roughly delimited by the 30.6ºN and 34.9ºN paralells and by the 20ºW and 14ºW meridians (blue shaded area on the panels bellow).
The panels below illustrate the domains vs. horizontal resolution used in each experiment:
The finer resolution employed in the DEODE runs allows, for example, for a better description of the island's orography. DEODE maximum height is closest than AROME's maximum height to Pico Ruivo (~1861m), which is the highest elevation on Madeira Island, located on the East central part of the island.
This event was characterized by massive amounts of precipitation in December 25 2020, reaching estimated accumulations values larger than 400 mm of rain in some locations. In order to test the capabilities of DEODE, we performed 48h forecasts in cycles of 12h starting from 24/12/2020 at 00:00 and compared them with the 48h forecasts preformed using CY43T2 of AROME in the same region.
Each 48h forecast run alone used 16 full nodes at a total of 4096 CPUS for about 5 hours with a cost of about 225000 SBU.
Although negligible in terms of SBU usage, the Marsprep task is the second one in terms of runtime, taking an average of two hours for each forecast run.
The preliminary analysis of the results was conducted by comparing the forecasts obtained with the two NWP configurations, executed with lead times of 48 and 24 hours and starting at 00UTC. As reference, some conclusions from [1] were used.
DEODE horizonal 10m wind fields have been filtered to the AROME mesh size. Over land, the 10m wind field shows more spatial variability in direction and intensity on the DEODE simulations (left panels), although upstream of the island the atmospheric flow exhibits similar patterns in both DEODE and AROME simulations.
Forecast from 24dec 00:00:00
Forescast from 25dec 00:00:00
Around the PBL height, the interaction of the atmospheric flow with the orography is still visible over land leading to convergence patterns over the adjacent Atlantic which differ between the two simulations. Upstream the flow exhibits similar patterns in both simulations.
Forecast from 24dec 00:00:00
Forescast from 25dec 00:00:00
The vertical wind illustration is presented as a series of sequential 2-hour forecasts, at 700 hPa. DEODE simulations illustrate the convergence line lead by the orographic effects on the atmospheric flow, as suggested on the horizontal wind illustrations. The convergence line prevails during the 24 hours of 25 December 2020. One should note also the convergence line is more discontinued in the 24h lead time run than on the 48h lead time run.
Forecast from 24dec 00:00:00
| 08:00 | 10:00 | 12:00 |
|---|---|---|
 |
 |
 |
Forescast from 25dec 00:00:00
| 08:00 | 10:00 | 12:00 |
|---|---|---|
 |
 |
 |
The 24h acumulated precipitation estimated from Porto Santo/Pico do Espigão (PS/PE) weather radar observation, is illustrated in the panel below. It shows the image RAIN24 (mm), 00-24UTC, valid at 00UTC of 26 december 2020. Without the hail correction, the maximum value is 523.7 mm (494mm, after the hail correction) at the north coast of the Island (illustration extracted from [1]).
The panels below illustrate the 24h accumulated precipitation simulated by the reference and the hectometric runs. DEODE seems to forecast high precipitation values more consistently for lead times of 24 and 48 hours than AROME. However, in none of the settings or runs the maximum values reached those estimated by the weather radar.
Forecast from 24dec 00:00:00
Forescast from 25dec 00:00:00
This link contains all the plots from other timesteps, togheter with some movies
An objective validation should take place, using as reference local conventional observations.
https://github.com/pardallio/dcmdb/tree/develop/cases/madeira_2020