Skip to main content

Climate Analysis Forecast Ensemble System

Leader: Terry O’Kane

Activity that combines the coupled modelling with the ensemble data assimilation and forecasting to provide the Climate Analysis and Forecast Ensemble System (CAFES).  The CAFES provides  a research tool to characterise the climate state over the recent past, to investigate the predictability of the climate system and to provide multi-year climate forecasts.

Skilful  multi-year prediction skilful multi-year forecasts of relevant climate observables relies on the development of coupled ocean atmosphere climate models in combination with advanced forecast methods underpinned by modern data assimilation methods and a rigorous mathematical framework. The factors that enable these advances at this point in time are the availability of modern subsurface ocean observations, knowledge of the relevant ocean modes of variability that are predictable, and the technical means to track the growth of predictable modes and reduce forecast errors. Advances in understanding have highlighted some of the key processes in the ocean upon which decadal predictability depends, and the models now feature high enough spatial resolution to simulate these processes with sufficient fidelity. Ensemble forecasts provide a basis to not only estimate the mean climatic state in the future but also the uncertainty in the climate system enabling informed strategic decisions that exploit knowledge of climate variability over multiyear to decadal scale.

A model ensemble is a set of forecast runs that differ from a control simulation by the introduction of small perturbations into each ensemble member forecast.  An initialised forecast will eventually depart from the true system as errors in the forecast grow.  Such errors can arise due to model biases or due to uncertainties in estimating the initial state. The rate at which the forecast diverges from the truth is a key indicator of transitions between climate regimes – e.g. periods of flood or drought. The requirement of a successful ensemble prediction system is to track and slow the growth of error modes ensuring more skillful forecasts.  This is a mathematically complex task but one that is now possible given new theoretical advances and the adaption of methods from numerical weather prediction. The procedure relies on generating an ensemble of forecast scenarios by perturbing about the predictable modes in the subtropical and midlatitude oceans where predictability on decadal timescales resides. The predictable modes will be quantified in testing of the coupled model at this stage of the project.  Runs will also be carried out to determine the advantages gained through use of the ensemble system and to determine the ensemble size and type required.  This work will answer questions about what can be predicted skillfully, and for how long into the future. Answering these questions tells us what can be realistically expected from the forecast system.