Data Assimilation

Data assimilation helps us make the most of our observations and our models by combining them in a mathematically optimal way. This helps us to provide more accurate and reliable model predictions, and learn more about how the Earth system works, using models to interpret and extend different types of observational data (reanalysis).

Data assimilation (DA) often involves the integration of data from a variety of sources, including satellite measurements, ground-based observations, and computer models. DA is a key ingredient for environmental model forecasting. It helps us identify how to improve models of the Earth system by identifying mismatches between observed data and the model predictions.

At NCEO, we are developing cutting-edge techniques in DA to help strengthen scientific capacity in the UK. Our DA group, one of only a few large academic DA research groups in Europe, develops new mathematical data science approaches and bespoke DA systems for a broad range of science and applications.

The DA framework has enabled us to enhance process understanding (e.g. the first DA-based soil moisture maps for the UK and Africa with improved soil parameters), deliver new reanalysis products (e.g. methane), optimize observing networks (e.g. autonomous ocean gliders) as well as contribute advances in uncertainty quantification which improved forecast accuracy at operational weather centres around the world. DA-enhanced models give more accurate predictions, for better decision-making across a broad range of sectors, from farming to flying.

Data assimilation is often thought of as a way of keeping a model ‘on track’ by constantly correcting it with fresh observations.

Professor Sarah Dance
NCEO Divisional Director of Data Assimilation based at the University of Reading.

Key research groups

NCEO DA staff are spread across four institutions:

• University of Reading
• Plymouth Marine Laboratory
• University of Leeds
• University of Edinburgh

We work closely with researchers at operational centres like the Met Office and the European Centre for Medium-range Weather Forecasts (ECMWF) to improve on the approximations in their data-assimilation systems for future weather and climate modelling.

We collaborate with the Alfred Wegener Institute, Bremerhaven on the development of our shared data assimilation community software PDAF, and with UKCEH on the development of data assimilation systems for the UK community land-surface model JULES.

Research priorities

• Assimilation into models of the entire Earth system, across a range of space- and time scales, including proper treatment of processes at the interfaces between Earth-system components (e.g. land-atmosphere).
• Uncertainty quantification and machine learning: speeding up computations using modern deep learning approaches, while quantifying and improving accuracy and physical fidelity.
• Partnerships and software: collaborations with academic and end-user communities, contributing to community software, ensuring that our work is used and useful in a fast-changing environment.

Our current work includes

• Improving the representation of uncertainties in data assimilation systems used for high resolution chemical transport modelling and reanalysis.
• Designing new hybrid DA-Machine Learning approaches approaches for coupled ocean-biogeochemistry, for better monitoring and prediction of coastal algal blooms and hypoxic events.
• A collaboration with the Met Office, developing new metrics of observation influence for regional ensemble hazardous weather prediction.
• Developing joint-state parameter estimation systems for the JULES land-surface model with applications to hydrology, in collaboration with UKCEH.