It eases forecast data preprocessing, statistical downscaling through a selection of machine learning techniques and result validation. It provides a flexible module for any modelling scheme requiring tailored climate inputs, enables scalability and applicability to other domains, resolutions, and datasets.
Release Notes
The development of the downScaleML Python package is currently underway, and it is being piloted on a virtual machine hosted at Eurac Research.
The current version of the downScaleML package comprises two distinct workflows: one for preprocessing the input data and another for performing downscaling. In the current developmental stage, the downscaling component identifies the appropriate model and parameters through a grid-search process.
Future Plans
For upcoming package updates, our objectives include:
- Implementing the most effective downscaling machine learning models to process real-time seasonal forecast data from ECMWF, with a focus on downscaling climate variables, particularly temperature and precipitation, while emphasising climate extremes.
- Containerizing the entire package environment using Docker and establishing a streamlined pipeline for managing input and output data through openEO processes.