ECMWF: Scientist for Machine Learning
ECMWF, Shinfield Park, UK
Closing date: 22 October 2020
ECMWF has embarked on an exciting new initiative to explore the use of artificial intelligence and machine learning in applications of numerical weather predictions. To learn more about the application of machine learning in the weather and climate domain and at ECMWF, please have a look at the webpage of the Machine Learning Seminar Series at ECMWF (https://www.ecmwf.int/en/learning/workshops/machine-learning-seminar-series) or the ESA-ECMWF machine learning workshop which is planned for October
As part of this effort, ECMWF is coordinating the MAchinE Learning for Scalable meTeoROlogy and climate (MAELSTROM) EuroHPC project to fund this position. This Scientist position will be in the Physical Processes team in the Research department at ECMWF. The successful candidate will apply their skills, knowledge and expertise to help achieving the goals, and complete the deliverables, of the MAELSTROM project.
The main focus will be on the development of machine learning emulators for some of the parameterisation schemes of ECMWF’s Integrated Forecast System (IFS). The use of deep learning to emulate some of the parametrizations used to represent subgrid atmospheric processes, such as radiation or clouds, promises a significant reduction of computing cost and improvements in portability of the models to heterogeneous hardware, and could potentially lead to improvements in predictive skill if savings are reinvested into higher resolution or model complexity. This project builds on previous studies which have successfully used neural network emulators within weather and climate simulations. However, the use in a forecasting system for operational weather predictions, such as the IFS, will require more complex machine learning solutions and training data of higher quality than what has been used so far. The successful candidate will also explore the use of such emulators within the data assimilation framework.
The Scientist will work in close collaboration with other teams across the organisation and strong communication skills are essential.
Main duties and key responsibilities
- Diagnosing training data in form of input/output pairs of physical parametrisation schemes from simulations with the IFS at high resolution
- Publishing the training data in a user-friendly form for use as benchmark dataset for MAELSTROM
- Developing customised machine learning solutions for the emulation of parametrisation schemes
- Reintroducing machine learning solutions into the IFS source code and evaluate model fidelity in coupled forecast and data assimilation experiments
- Testing machine learning emulators for data assimilation applications
- Contributing to reports, and dissemination and training activities of the MAELSTROM project
The successful candidate will be recruited at the A2 grade, according to the scales of the Co-ordinated Organisations and the annual basic salary will be £60,590.64 net of tax. This position is assigned to the employment category STF-PL as defined in the Staff Regulations.
Full details of salary scales and allowances are available on the ECMWF website at www.ecmwf.int/en/about/jobs, including the Centre’s Staff Regulations regarding the terms and conditions of employment.
Starting date: 1 January 2021, or as soon as possible thereafter.
Length of contract: 30 months, subject to available funding with possibility of extension.