Several of our Climate Dynamics and Prediction researchers just published a new article in Climate Dynamics, under the lead of François Counillon (Nansen Center). They showed how reducing model biases can improve seasonal prediction for the tropical Atlantic Ocean.
Climate models are very useful and have been in use for some decades. They help us understand the past, current, and future climate. How well these models represent climate has improved vastly in the past years and using methods such as data assimilation is contributing to the advancements being made. Standard climate models couple ocean and atmospheric dynamics to be able to simulate phenomena like the El Niño Southern Oscillation. Such phenomena are behind the skill in seasonal predictions. However, all models, including coupled models, suffer from biases that decrease their ability to perfectly predict climate. François Counillon, Noel Keenlyside, and Yiguo Wang (all NERSC), along with other Bjerknes Centre co-authors, set out to tackle biases in the tropical Atlantic Ocean region using an innovative modelling approach. Their article’s title is “Relating model bias and prediction skill in the equatorial Atlantic”.
The tropical Atlantic Ocean is a region in which coupled Earth system models traditionally suffer from large biases: The temperature of the sea surface, the amount of rain, and winds are not modelled correctly with standard models. These shortcomings lead to forecasts not being reliable enough, which is problematic in this region – where the societal impact of climate variability is large.
To address the problem, Counillon and his colleagues made use of “anomaly coupling”, an innovative technique developed in 2018 by two of the co-authors of this paper, Toniazzo and Koseki. This technique makes it possible to limit the known biases in the region effectively, and by doing so, the authors were able to assess how much better the model performed in predicting on a seasonal scale without the large biases.
What Counillon and his co-authors did is a comparison between the performance of the standard version of the Norwegian Climate Prediction Model (NorCPM) and a version including the “anomaly coupling” technique.
Their findings indicate the following:
- Correcting the biases through “anomaly coupling” makes the model perform much better – as good as the best models of the North American multi-model ensemble.
- The technique makes it possible to achieve good and reliable predictions up to half a year ahead for the secondary peak of interannual variability that occurs in November-December.
- The results are improved, because the model better mimics the true dynamic, and as such can make a more efficient use of the observations to produce the initial condition from which predictions are started. By improving the model, we improve the data assimilation method.
Lead author François Counillon says the following about the importance and limitations of their findings: “While we wait for models to improve and cure the very large bias in the tropical Atlantic, we can make use of innovative methods to mitigate these biases and greatly enhance the quality of our prediction. Still, this study also suggests that model bias is not the sole explanation of the poor prediction skill in particular for the Atlantic Niño that peaks in summer. However, it has opened new avenues on how to further improve our predictions that will be investigated further within EU-TRIATLAS.”
Keenlyside is excited about another approach he and Counillon are also working on in the ERC STERCP project: “We are building supermodels by combining models interactively to reduce errors even further and in a more dynamically consistent way. I believe our approach can revolutionize the way ensemble predictions will be done in the future.”
