Treatment of model bias in coupled atmosphere-ocean data assimilation

Lead Research Organisation: University of Reading
Department Name: Mathematics and Statistics


It is expected that the change in climate over the next century is likely to lead to many more extreme weather events, which will have significant impacts on society. In order to be able to plan for societal developments policy makers need to understand the likely effects of climate change over the coming decade. This information would be useful in planning projects designed to alleviate the effects of climate change, such as flood defences, as well as for more general projects, such as deciding where to build new housing. Scientists are currently developing methods to predict general weather phenomena over time scales of several years using computer simulations of the atmosphere and ocean. However, whereas many advances have been made in recent years in forecasting on time scales of days to weeks, the science of forecasting on much longer time scales is still in its infancy.

Recent developments in this area suggest that certain parts of the climate system may be predictable on time scales of several years if we can know more accurately the current state of the atmosphere and ocean throughout the world. Data assimilation is the science of combining observations of the atmosphere or ocean with computer simulations in order to be able to determine more accurately the current conditions and so produce a better forecast. It has been widely used in both weather forecasting and ocean forecasting for many years. However in developing predictions on seasonal to inter-annual time scales we need to simulate the evolution of the atmosphere and ocean together. Determining the current atmospheric and ocean states together is made more difficult in particular by two factors. One is that the atmosphere and ocean evolve on very different time scales and this is not very well handled by current methods of data assimilation. The other factor is that the computer models inevitably contain errors, due to our imperfect knowledge, and these errors are exacerbated when we treat the two systems together. In this project we will develop new data assimilation methods to determine simultaneously the state of the atmosphere and oceans using observed data, taking account of both the different time scales in the two systems and of the unknown errors in the computer models. The direct involvement of the European Centre for Medium-range Weather Forecasts in the project will allow a transfer of knowledge to operational practice.

Planned Impact

The science in this proposal will be of great interest to operational agencies that are developing coupled atmosphere-ocean forecasting systems for seasonal to decadal prediction. This includes institutions such as the European Centre for Medium-range Weather Forecasts, who are directly involved in this proposal, and the UK Met Office. The new techniques generated by this project will be able to feed directly into their research and development plans, providing guidance as to how to apply variational data assimilation in coupled systems.

Since the project is designed to develop techniques that can inform operational practice, it has the potential for to provide wide benefits to society. The development of more accurate seasonal to decadal forecasts initialised using data assimilation will allow more detailed guidance to be given to policy makers that must invest in projects on these time scales and will allow more detailed planning to alleviate the negative effects of climate change. Improved prediction over seasonal time scales will also be useful to end users of such forecasts, for example the farming community in parts of the world commonly affected by floods and droughts.


10 25 50
Description Using a simplified model and data assimilation system, we have demonstrated the potential benefits of moving towards coupled data assimilation systems for atmosphere-ocean prediction. In particular we examined two expected benefits, (i) a reduction of the initial imbalance or 'initialisation shock' that occurs when coupled atmosphere-ocean models are initialised using separate data assimilation systems for the atmosphere and ocean; and (ii) the possibility to use near-surface observations to understand information about the atmosphere and ocean simultaneously. We showed that if the assimilation systems are strongly coupled, then these benefits can be seen. When the coupling is only weak, such as in the systems currently being developed for operational forecasting, then some of the benefit can be achieved, but this may be sensitive to the inputs to the system.

We then considered how the presence of model error affects the results. We have shown that when model error is present in a system, then an uncoupled assimilation may produce a better estimate of the state at initial time, but a coupled assimilation scheme can provide an estimate that leads to an improved forecast.

Furthermore, we have developed a new method and allowing for model error within data assimilation. This method is particularly attractive as it avoids the need to calculate directly the statistics of model error that are required in other approaches. However, it does also provide a method for estimating these statistics indirectly.

The work on this grant has led to 3 publications, plus further interaction with both the Met Office and ECMWF.
Exploitation Route The results show the benefits of moving towards coupled data assimilation and justify efforts by operational weather forecasting centres to move in this direction and illustrate the potential gains. In particular, the range of experiments carried out provide some guidance on what are the most important components of a coupled assimilation system. There are indications that moving towards strongly coupled systems would be beneficial.

The method for diagnosing and allowing for model error could be used by forecasting centres. We currently have a small new project to apply this method to an operational system.

The simplified model and assimilation system developed continues to be used by our group to take this work forward. A new project is using the model to develop hybrid assimilation systems for coupled models and is looking at the structure of atmosphere-ocean cross-covariances.
Sectors Environment

Description Our work has been cited by the Met Office in justification of the need to develop a more strongly coupled data assimilation system (doi: 10.1175/MWR-D-15-0174.1).
First Year Of Impact 2015
Sector Environment
Description NOAA grant panel
Geographic Reach North America 
Policy Influence Type Participation in a advisory committee
Description Data assimilation projects
Amount £260,000 (GBP)
Organisation European Space Agency 
Sector Public
Country France
Start 07/2012 
End 06/2014
Description Novel Assimilation Methods for Coupled Atmosphere-ocean Prediction (NAMCAP)
Amount £191,588 (GBP)
Funding ID P107915 
Organisation Newton Fund 
Sector Public
Country United Kingdom
Start 04/2020 
End 03/2021
Title Single column model and assimilation system 
Description A single column atmosphere-ocean model and data assimilation system has been built, based on a column version of the ECMWF IFS atmosphere model and a mixed layer ocean model. 
Type Of Material Computer model/algorithm 
Provided To Others? No  
Impact This tool has allowed the study of different approaches to the coupled data assimilation problem that cannot easily be compared in full size model. The research by our group has been published (Smith et. al 2015) and has been cited by the Met Office in support of their future directions in this area (doi:10.1175/MWR-D-15-0174.1) 
Description ECMWF coupled DA 
Organisation European Centre for Medium Range Weather Forecasting ECMWF
Country United Kingdom 
Sector Public 
PI Contribution Presentation of project results at informal project meetings.
Collaborator Contribution Provision of single column atmosphere and ocean code for use in our project. Initial support to get the code running.
Impact 4 publications.
Start Year 2012
Description ECMWF model error 
Organisation European Centre for Medium Range Weather Forecasting ECMWF
Country United Kingdom 
Sector Public 
PI Contribution A method to estimate model error statistics was developed by my research team in the context of simple models. In this collaboration we took data from ECMWF and began applying the same method to this real data. Work is still ongoing.
Collaborator Contribution ECMWF provided data on which to test the methods.
Impact Work still ongoing - no outcomes so far.
Start Year 2016
Description Met Office coupled DA 
Organisation Meteorological Office UK
Country United Kingdom 
Sector Academic/University 
PI Contribution Visits to Met Office to present results from project.
Collaborator Contribution Met Office provided staff time to comment on results and proposed experiments.
Impact 4 publications. Further grant application for funding
Start Year 2012
Title Single-column coupled atmosphere-ocean assimilation system 
Description Starting from model code developed by ECMWF, we have developed a variational data assimilation system for a single-column coupled atmosphere-ocean model. This includes strongly and weakly coupled assimilation systems, plus the ability to calculate covariance information from ensembles. 
Type Of Technology Software 
Year Produced 2015 
Impact This software has been used in the group to allow fundamental research on coupled atmosphere-ocean data assimilation. This research has been communicated to operational users through conference presentations and peer-reviewed papers, and results are being taken account of in their systems. 
Description NCOF working group on data assimilation 
Form Of Engagement Activity A formal working group, expert panel or dialogue
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Professional Practitioners
Results and Impact Working group on data assimilation of the National Centre for Ocean Forecasting. This involved discussions with practioners (especially Met Office) about how our research results could help them to interpret their own results in a full atmosphere-ocean model and how our results could guide their developments.
Year(s) Of Engagement Activity 2014,2015,2016