Research Project Summary:

In Climate Science, downscaling is a process of converting global climate model output to high resolution regional data so that it can be used in local scale impact studies. Downscaling can either be dynamical or statistical, dynamic models are often more accurate but tend to be very expensive to run while statistical models are quick to run but struggle with reproducing extreme events. The overarching aim of this project will be to build an efficient machine learning downscaling model that is capable of correctly representing the extremes and can be used in regional impact studies under climate change conditions. Several key themes will run throughout the project, these being: model interpretability, physical/machine learning hybrid models and extreme event predictions. The initial steps will include building a linear model baseline and running a dynamical regional climate model with which to compare performance.