In climate science, we will use datasets collected from existing external projects as well as public datasets to build prediction models that out-perform existing analytical and simulation models. Given enough information about the current state, analytical and simulation models can help us to understand complex processes in the climate and predict future trends. However, for reasons of computational tractability, existing models are often artificially limited in terms of the number parameters they support, and existing simulations are computationally expensive, requiring large scale simulations. We will design and validate Deep Neural Networks (DNNs) that provide comparative predictions with massively reduced computational complexity.

We will build on the experience we have from the FAST MCP. These include using Machine Learning techniques to construct radiation parameterization from already calculated profiles in reanalysis to speed up calculations in global models. We are also working on relating turbulent surface fluxes to flow properties using observations, also with the aim to improve and speed up global model calculations. New ideas include using ML techniques to find different processes influencing Arctic cloud formation and dissipation by observations in terms of in-situ and satellite data in combination with high-resolution reanalysis data and possibly large-eddy simulation data.