TesseRugged

Global reanalysis data sets such as ERA5 constitute an important backbone for a wide range of topics, most notably including applications related to renewable energy and agriculture, as well as driving fields for climate control simulations. However, due to its resolution of 31 km and even lower resolved uncertainty information, ERA5 lacks details and applicability for regions with heterogeneous terrain or renewable energy applications. 

To increase the ERA5 spatial resolution in a step-wise manner for the whole globe without the need for large computational resources, parameter-wise downscaling with statistics and/or machine learning using a higher resolved reanalysis data set as target/proxy is a promising approach. For such a purpose, the CERRA data covering Europe with a spatial resolution of 5.5 km (and 11 km for the ensemble) is ideal. 

Here, we implemented a model output (baseline) approach, and two deep learning approaches for post-processing and downscaling using residuals. 

Follow the developments on GitHub