Effects of ocean coupling on weather forecasts

Kristian S. Mogensen
Tim Hewson
Sarah Keeley
Linus Magnusson


Experiments have shown that using interactive ocean and sea ice components in ECMWF’s Integrated Forecasting System (IFS) can significantly improve sea-surface temperature predictions in Europe and, as a result, predictions of near-surface air temperature.

With the operational implementation of IFS Cycle 45r1 on 5 June 2018, all forecasts issued by ECMWF are based on a coupled model with interactive ocean and sea-ice components. This coupled model exchanges data from the atmospheric model with the ocean/sea-ice models and receives back information about the dynamic evolution of the sea-surface temperature (SST) and sea-ice concentration. A recent ECMWF Newsletter article (No. 154, winter 2017/18 issue) showed that SST coupling is important for the prediction of large-scale tropical cyclones, but there are other situations where this coupling also matters. Here we show two examples of how coupling can improve the prediction of SST in the seas surrounding Europe, and we present the repercussions on atmospheric variables. All results shown here are from research done in preparation for the introduction of coupling in ECMWF’s high-resolution forecasts (HRES) in 45r1.

Physical mechanisms

Aside from advection of sea water, there are multiple physical mechanisms that can cause substantial in situ changes to SST:

  1. Cooling via an upward sensible and/or latent heat flux. This cooling is enhanced when the overlying air is, in relative terms, very cold and/or very dry, and also when winds are strong.
  2. Heating via insolation. This is enhanced when winds are light, because there is a reduction in mechanical mixing, allowing heating to be confined to the upper ocean layers.
  3. Cooling by mechanical mixing. This is particularly effective when a very shallow layer of relatively warm water at the top of the ocean overlies much cooler water below (e.g. following an insolation-related heating episode as in 2). The converse, heating by mixing, tends not to occur because it requires, beforehand, a stratification with cold water above warm water, which is generally unrealistic as it is unstable (although salinity variations can complicate the picture).

The most powerful of these processes is probably (3), with changes of as much as 10ºC in 24 hours having been recorded in the Adriatic during a Bora wind event. The coupled model should be able to represent
all of the above processes, although the depth of the uppermost ocean model layer, 1 m, could be a limiting factor. The first example below illustrates processes (2) and (3) whilst the second example illustrates process (1).

Mistral case. Predicted SST from experimental runs with ECMWF’s coupled model starting from 00 UTC on 10 June 2017 valid at 00 UTC on 15 June (top left) and at 00 UTC on 17 June 2017 (top right). The other panels show the observed and predicted mean sea level pressure, wind speed, 2-metre temperature (2mT) and SST at the location of two buoys marked A and B in the top panels.

Effect of summer insolation and Mistral on SST

During June 2017, there were long periods of warm weather over the Gulf of Lion interspersed with several Mistral events (strong cold winds blowing from the south of France into the Gulf of Lion). An example of a simulation with the new coupled HRES configuration can be found in the first figure. After the onset of the Mistral, predicted SST is reduced by about 2ºC in large parts of the Gulf of Lion. To get a feel for how realistic the simulation is, we have compared it with measurements from two moored buoys deployed by Météo-France (black diamonds in the top panels). The agreement between the predicted SST and the observed SST is remarkably good up to 8 days for the eastern buoy (marked ‘A’ in the map) and even better for the western buoy (marked ‘B’ in the map). It is worth noting that during the first five days of the forecast the SST at buoy B is steadily increasing (with a diurnal cycle superimposed), but as soon as the wind starts to pick up the SST drops rapidly, for a period of 48 hours, before increasing again. Buoy A shows a similar steady increase in SST during the first 8 days, but the model fails to capture the rapid increase in wind from day 8 to day 9, and because of this it also fails to predict the concurrent decrease in SST.

Comparing the coupled simulation with an uncoupled simulation (bottom panels), we see that mean sea level pressure (MSLP) and wind speed are similar, but the predicted 2-metre temperature is quite different. Observations agree much better with the coupled model output. It is worth bearing in mind that the uncoupled simulation only changes the SST in accordance with date-based climatological trends, so it is not surprising that it fails to capture changes in 2-metre temperature originating from changes to SST.

Cooling of North Sea SST

During the winter of 2018, there was an extended period when exceptionally cold air blew in from the east across the North Sea. The second figure shows an example of the predicted SST at day 8 from a coupled HRES simulation (top left) together with an uncoupled simulation (top right) and the corresponding analysis (bottom left). In this case the coupled simulation predicts the cooling in the German Bight quite well, whereas the cooling in the Skagerrak between south Norway and Denmark is too large. If we compare with observations from a moored buoy (bottom right, diamond in maps), we can see that the predictions for the German Bight are in remarkable agreement with the observations.

North Sea case. Predicted SST with mean sea level pressure (contours) and wind (barbs) for 8-day coupled (top left) and uncoupled (top right) forecasts with corresponding analysis (bottom left). The bottom-right panel shows the SST evolution of the coupled and uncoupled forecast at the location of a moored buoy (marked with a diamond in the other plots) together with the observed SST.


The two examples presented here both show that the SST reacts to meteorological fields in physically sound ways. The Mistral case also shows that this can lead to improvements in the prediction of 2-metre temperature. Work will continue to improve the understanding of the processes involved in interactions between the atmosphere and the ocean (including sea ice) to further improve our coupled modelling capabilities in accordance with ECMWF’s Strategy.