Seamless Ensemble Prediction
Seamless Ensemble Prediction is about quantifying and understanding uncertainties in climate projections on a range of time scale (monthly to seasonal, decadal and long-term). Where it is feasible and justified, this involves making probabilistic projections from observations and ensembles of climate model simulations, so decision-makers can use them in risk-based approaches for planning their adaptation and mitigation strategies.
We cannot be certain about future climate change because:
- some variations in climate are inherently unpredictable (internal variability);
- we evaluate climate model output using measurements which have errors (observational uncertainty);
- we only have plausible storylines of how anthropogenic emissions might evolve (emissions uncertainty);
- we have limited computer resources and an imperfect knowledge of the Earth system, so climate models have to approximate some of the key processes that affect climate change (modelling uncertainty). By understanding our climate model across a range of time scales, we can learn about errors in our climate model and factor this information into the projections.
Therefore, there is no single best estimate, only a range, of future climate change.
To explore the modelling uncertainty, we generate ensembles of climate projections where each realisation differs from the other members by altering the approximations of the key climate processes within plausible bounds. These ensembles, together with output from other climate models, are used to understand and quantify uncertainties in climate projections on different space and time scales. When appropriate, statistical methods are used to combine our ensembles with observations and climate projections from other international models to provide probabilistic climate projections of local climate on (multi)decadal time scales. The observations are used to weight each model variant according to its ability to simulate the observed changes. The probabilities measure the credibility of any given level of climate change. Planners can then assess the risk of exceeding a threshold of climate change they are vulnerable to, and use this information in deciding adaptation and mitigation strategies.
In 2009, our probabilistic climate projections were published for the UK, called UKCP09, and for limited climate variables for Europe as part of the ENSEMBLES project. Part of our role is to work with some users of the data to evaluate whether the product meets their needs, and if not, to factor these missing needs into future products.
- To run ensembles to explore the uncertainty in climate modelling and to understand what drives predictability.
- To develop new experimental designs that allow the key uncertainties to be sampled.
- To combine these ensembles with observations and information from other climate models to produce probabilistic climate projections on different time scales.
- To work with the users of the probabilistic information to understand their needs and to develop new prediction systems that meet these needs and new
ways to represent the uncertainty inherent in climate prediction.
- In 2009, we produced climate projections for UKCP09 and ENSEMBLES. Currently writing up the methodologies that underpinned these two products.
- In 2010, we extended the UKCP09 product by adding information on wind speed, fog, lightning, and using storylines which are snapshots of future climate that help users understand their climate vulnerabilities, in particular for nationwide adaptation.
- Analysis of a recently completed ensemble that simultaneously explores the uncertainty in four different components of the Earth System (atmosphere/land, ocean, sulphur cycle, and carbon cycle), understanding how the uncertainties interact with each other.
- Develop statistical methods that allow us to provide more multivariate information and use the observations more effectively.
- (Long-term) Work towards merging projection methods on monthly to seasonal, decadal, and centennial time scales or at least understand how predictions on shorter time scales can inform longer term projections. The first stage is to run an ensemble of the latest climate model with prescribed SSTs to understand which model parameters control predictability.