Stochastic Multiscale Climate Models

Current high-performance computers do not allow climate models to capture all relevant processes through a higher resolution or better process description. High-resolution ocean models use already near eddy-resolving horizontal resolutions (e.g. 0.1°) but many important processes such as upper ocean turbulence and sub- mesoscale eddies, are not adequately captured. In practice, ocean models need to be simulated for multiple millennia to give meaningful results for the current climate, which can only be reached with low-resolution models.

To overcome this dilemma a low-resolution model can be integrated over multiple millennia and next the solution is interpolated to start the high-resolution model at an interesting era. In this project we developed a Python interface to do so. From theory we know that interpolation errors and missing physics will be quickly recovered by the high-resolution model. However, the closer the low-resolution solution is to the one of the high-resolution model, the faster the recovery process will go. Therefore, we try to improve the low-resolution model by combining its outcomes with those learned from data generated by the high-resolution model. We tested this idea successfully for a simplified ocean model and the next step will be to add it to the advanced models.