Simulating climate with a synchronization-based supermodel

The SPEEDO global climate model (an atmosphere model coupled to a land and an ocean/sea-ice model with about 250.000 degrees of freedom) is used to investigate the merits of a new multi-model ensemble approach to the climate prediction problem in a perfect model setting. Two imperfect models are generated by perturbing parameters. Connection terms are introduced that synchronize the two models on a common solution, referred to as the supermodel solution. A synchronization-based learning algorithm is applied to the supermodel through the introduction of an update rule for the connection coefficients. Connection coefficients cease updating when synchronization errors between the supermodel and solutions of the “true” equations vanish. 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.

Complex numerical codes are being used to predict the behavior of real-world phenomena like the climate or the economy. In this study, we demonstrate that predictions can be improved by forming an ensemble of inter-connected different imperfect climate models that synchronize on a common solution, referred to as the supermodel solution. This supermodel solution depends on the connections. The connections are trained using observations of the truth such that the supermodel minimizes synchronization errors when nudged to trajectories of the truth. 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.