by Amos Lawless, April 2025

Why is the ocean relevant for weather forecasts?

The weather we experience each day is influenced not just by what is happening throughout the atmosphere, but also by what happens in the oceans. For example, the formation of storms and hurricanes over the oceans depends greatly on the transfer of heat and moisture from the ocean to the atmosphere. Hence, to try to improve weather forecasts, many operational weather forecasting centres are now using coupled atmosphere-ocean models for prediction, with a more realistic representation of the physical interactions between the atmosphere and ocean. This means that we need to understand the current states of both the atmosphere and ocean to be able to make a forecast.

Coupled data assimilation

To estimate the atmospheric and oceanic states, we can combine observations of both systems with our model predictions using data assimilation techniques. Operational centres have been using data assimilation for the atmosphere and ocean separately for many years. However, building an assimilation system for the coupled atmosphere-ocean system is not an easy task, with technical and scientific challenges. Hence, current assimilation systems use what we call ‘weak’ coupling. In this approach, atmospheric observations help estimate only the atmospheric state, and ocean observations help estimate only the oceanic states. We then combine these two estimates [1,2].

The weakly coupled approach is sub-optimal, because we know that physical quantities near the surface are linked. For example, we expect a strong link between the sea-surface temperature and the air temperature near the surface. Hence a sea-surface temperature measurement (for example, from satellites) also provides information about the temperature of the air above. The question is, can we make better use of near-surface observations in our assimilation systems by including some coupling information in the assimilation?

Understanding the connection between the atmosphere and ocean

To better use such observations, we must understand how the uncertainties in the atmospheric and oceanic forecasts are related.  We can quantify this relationship using the statistical measure of correlation. For example, suppose that we have a measurement of sea-surface temperature that is warmer than our model forecast. This would tell us that our model is too cold and we should increase the sea-surface temperature in the model. If there is a positive correlation between the uncertainties in our sea-surface temperature and air temperature, this would imply our air temperature is also too cold, and we should increase the air temperature in the model. Hence, by knowing the correlation, the measurement of the sea-surface temperature allows us to deduce information about the air temperature.

These kinds of correlations are currently not represented in data assimilation systems, and their inclusion would require a huge effort. To decide whether to undertake the challenge, we must understand if such correlations exist, and are strong enough to matter, in real models. This was the focus of some recent work in our group [3]. In collaboration with the Met Office, we used their coupled model to estimate these correlations on timescales relevant for numerical weather prediction. In Figure 1 we show our calculated daily mean correlation between sea-surface temperature and air temperature. As expected, we see that it is positive almost everywhere and has large correlation of above 0.4 in places.

Figure 1. Daily mean correlation between anomalies in sea-surface temperature and 2m atmospheric temperature on 15 December 2019. Figure taken from [3].
We also found that correlations vary spatially, with the time of day and with the current weather situation. For example, in Figure 2 we show the correlation between sea-surface temperature and the atmospheric 10m wind. Over the Indian Ocean we see large negative correlations, associated with the diurnal heating at the surface. They are present only at the surface and disappear during the night. However, over the North Atlantic we see a positive correlation associated with a low-pressure system. We find correlations between the wind and the ocean temperature down to depths of over 100m.

Figure 2. Daily mean correlation between anomalies in sea-surface temperature and 10m wind on 15 December 2019. Figure taken from [3].

What next?

Our results show that correlations exist between the uncertainties in our atmosphere and ocean forecasts. These correlations are not only strong in places, but exhibit spatial, temporal and weather-dependent variations. These findings suggest that accounting for such relationships could significantly enhance coupled data assimilation systems. However, how to do this in practice remains a complex challenge for future research.

References

[1] Browne, P., de Rosnay, P., Zuo, H., Bennett, A., Dawson, A. (2019). Remote Sensing 11, 234.

[2] Lea, D. J., Mirouze, I., Martin, M. J., King, R. R., Hines, A., Walters, D. and Thurlow, M. (2015). Mon. Wea. Rev, 143, 4678-4694.

[3] Wright, A., Lawless, A.S., Nichols, N.K., Lea, D. and Martin, M. (2024), Assessment of  short-range forecast error atmosphere-ocean cross-correlations from the Met Office coupled NWP system. Quart. J. Royal Met. Soc., 150, 2783-2797.