Supermodels for improved weather and climate forecasts (Speaker Frank M. Selten)

Speaker: Frank M. Selten (R&D Weather and Climate Modelling)

Sometimes model A makes a better forecast, sometimes model B. Combining forecasts in multi-model ensembles generally improves skill. Here we demonstrate that the skill can be further improved by allowing models to exchange information during the forecast.

We use a global climate model in a perfect model setting. The model with standard parameter values is regarded as the truth, imperfect models are created by perturbing parameter values. We dynamically combine models by introducing connection terms into the model equations that nudge the state of one model to the state of every other model in the ensemble, effectively forming a new dynamical system with the values of the connection coefficients as additional parameters. For appropriately chosen connections, the models synchronize on a common solution that depends on the values of the connection coefficients. This solution is referred to as the supermodel solution. During a learning phase, the supermodel is nudged to an observed trajectory and the connection coefficients are adjusted by update rules that depend on the synchronization error. The connection coefficients cease to update when the synchronization is perfect. These final connection coefficients define the supermodel.

Different supermodel solutions, but with equivalent performance, are found depending on the initial values of the connection coefficients during learning. The supermodels have a climatology and a climate response to a CO2 increase in the atmosphere that is closer to the truth as compared to the imperfect models and the standard multi- model ensemble average, showing the potential of the supermodel approach to improve climate predictions.

This is the first time that the potential of the supermodel approach is demonstrated in the context of a complex global climate model. Due to its computational efficiency, the synchronization-based learning approach is applicable to state-of-the-art climate models with millions degrees of freedom and historical observations of the Earth’s global climate system. In cooperation with University of Bergen we are currently working on the construction of a supermodel based on the EC-Earth, ECHAM and NorESM climate models.