CAFE: CSIRO’s Decadal Prediction System

Disentangling the combined, and sometimes competing effects, of anthropogenic warming due to increasing levels of greenhouse gases, and natural variations over the next few seasons to a decade, on the climate system is a grand challenge. Natural variability on interannual timescales such as El Niño occur due to the combined effects of fast atmospheric (stochastic) processes forcing equatorial ocean thermocline processes. On decadal timescales, large-scale ocean heat content anomalies evolve slowly providing a potential mechanism for predictability years in advance. In near term climate prediction, an ensemble of climate models is initialised about our best estimate of the current climatic state. The different members are used to sample uncertainties in our knowledge of the current climate and their evolution gives an estimate of the reliability of the forecast. Initialization to observations is required to predict internal variability whereas specifying our best estimate of future changes in anthropogenic sources of greenhouse gases and aerosol concentrations and projected changes in solar irradiance and volcanic aerosols is required to capture the forced response.

The Climate Analysis Forecast Ensemble system (CAFE) developed at CSIRO was founded on using a large ensemble (96 members) Kalman filter. While most climate forecast systems are initialised by relaxing to independently generated analyses of ocean temperature and salinity, and atmosphere analyses of winds, temperature and surface pressure, CAFE was one of only a few systems to use data assimilation to initialise the ocean, atmosphere and sea-ice directly to observations. By adjusting the initial model atmosphere to ocean observations and visa-versa, a process known as strongly coupled data assimilation, there is the further potential for better balanced forecasts avoiding large forecast errors due to “initialization” shock. However, such an approach requires a large ensemble of forecasts to accurately determine the covariations of the respective physical domains – at a significant computational cost. The CAFE system was the first to apply advances in strongly coupled data assimilation and to employ a very large ensemble, enabling a probabilistic approach to forecasting the near term climate analogous to current state of the art weather prediction systems. Updated decadal forecasts, along with verification of previous forecasts, are available from https://hadleyserver.metoffice.gov.uk/wmolc/

References

• T.J. O’Kane et al (2020) CAFE60v1: The CSIRO Climate re-Analysis and Forecast Ensemble system, Version 1 (in prep)

• P.A. Sandery, T.J. O’Kane, V. Kitsios and P. Sakov (2020) State estimation of the climate system with the EnKF using variants of coupled data assimilation Monthly Weather Review, 148, pp2411-2431, https://doi.org/10.1175/MWR-D-18-0443.1)

• T.J. O’Kane, P.A. Sandery, D.P. Monselesan, P. Sakov, M.A. Chamberlain, R.J. Matear, M.A. Collier, D.T. Squire & L. Stevens (2019) Coupled data assimilation and ensemble initialization with application to multi-year ENSO prediction, Journal of Climate, 32, 997—1024 doi.org/10.1175/JCLI-D-18-0189.1