Hereon at the ECMWF Code for Earth
Posted by Dr. Martin Ramacher, and Dr. Johannes Bieser
For last year’s edition of the ECMWF Code for Earth innovation programme, Dr. Johannes Bieser (Matter Transport and Ecosystem Dynamics) and Dr. Martin Ramacher (Chemistry Transport Modeling) successfully proposed and mentored a project on downscaling regional pollutant concentrations to urban scales using Machine Learning techniques.
Code for Earth is an innovation programme run by the European Centre for Medium-Range Weather Forecasts (ECMWF). Its aim is to drive innovation and open-source developments in the Earth sciences community – supporting developments in weather, atmosphere, and climate, Copernicus and Destination Earth (DestinE). Since 2018, each summer, selected developer teams work together with experienced mentors from ECMWF on innovative projects. These projects are related to the broad scope of activities at ECMWF, including data science, weather, climate or other earth sciences, and more.
For the 2024 edition, for the first time, the Code for Earth opened the programme for external partners and due to connections made during the 2023 Chaos Communication Congress (youtube video) Johannes Bieser and Martin Ramacher were invited to create and mentor a challenge alongside Miha Razinger (ECMWF), that fits into the nexus of ECMWF and Copernicus.
The goal of the proposed challenge was to develop an application capable of improving (downscaling) the resolution of regional-scale pollution models to urban-scale concentrations for urban areas in Europe. This idea was born out of need and interest to improve and compare existing urban-scale pollutant Chemistry Transport Modelling approaches at Hereon with more generic or simpler approaches to achieve urban air pollution concentrations and also to explore possibilities to use Machine Learning for this purpose.
The challenge was accepted by a team of three enthusiastic early career scientists with different background in Meteorology, Machine Learning, Engineering and Software Development: Eloi Codina Torras, Adrià Fenoy and Òscar Hernández Rodríguez from MeteoSim (Barcelona, Spain). Their approach to tackle the challenge was based on a Machine Learning application that considers a wide range of geo-referenced open-source datasets (land use, measurements, model results) to achieve the desired high resolution map of pollutant concentrations.
Although time was short and the goal ambitious, project results look very promising and have been presented at the final Code for Earth event in Reading, UK (youtube video). The experts at ECMWF expressed interest for a continuation of this project due to similar approaches being proceeded by ECMWF and Copernicus. Moreover, the project has now been submitted to the Helmholtz AI unit to explore the possibilities of similar ML approaches for the same purpose and eventually integrating such methods into our departments‘ toolboxes.
Johannes and Martin have already been invited to propose and mentor another challenge for this year’s edition of the ECMWF Code for Earth programme. If you are interested in participating in the programme as a project team that works on one of the proposed challenges, stay tuned. Successful applications are granted 5000€ on completion of a challenge.
