Retrospective multiyear climate forecasts from the Climate Analysis Forecasting Ensemble System

Australia is a climate exposed nation where communities and industries are impacted by the combined influences of large internal variability coupled with increasingly more intense extreme events like drought, floods and bush fires driven by anthropogenic forcing. Many commercial (e.g. Primary producers) and government entities (e.g. Local, State and Federal) want to include climate information in their annual to decadal business plans. Recognising the enormous potential benefit of climate forecasting on the annual to decadal timescale to Australia, CSIRO has initiated the Decadal Climate Forecasting project (DCFP). Useful climate information on these time scales will help Australia better handle climate variability and extremes. Our project is using Pawsey systems to both improve annual to decadal global climate forecasts and to investigate their utility. The goal of the project is to help Australia become more resilient to climate variability and climate extremes.

Principal investigator

Richard Matear
Magnifying glass

Area of science


Systems used


Applications used

MOM5 with the sea ice, atmosphere and ocean biogeochemistry
Partner Institution: CSIRO Ocean and Atmosphere | Project Code: 0315

The Challenge

Delivering better climate forecasts on the annual to decadal climate time scale is a recognised grand challenge in climate science and a focus of the World Climate Research Program. This project will make a valuable contribution to this international effort by using the capabilities Pawsey provides to help generate novel sets of annual to decadal climate forecasts and to apply innovate analysis to extract utility from these forecasts.

The Solution

We have developed the Climate Analysis Forecast Ensemble (CAFE) system (O’Kane et al. 2021a and b) and we are using the system to explore the benefits to the forecast skill of:
1) more sophisticated ways to assimilate ocean, sea ice and atmospheric observations to determine the initial climate state;
2) novel ways to generate the perturbations of the climate state for delivering multiyear ensemble forecasts.

Both approaches are well developed, and we recently published our 96-member ensemble climate reanalysis product (CAFE60) to characterise the climate since 1960 (O’Kane et al. 2021a and b). We have used a subset (10-members) of CAFE60 as initial conditions for decadal climate forecasts run every Nov 1 from 1960. The forecast dataset has now been supplement by 96-member decadal forecasts every November 1 and May 1 from 1980 to 2020. These additional forecasts provide a large ensemble with a wealth of information to characterise extremes in the current climate (e.g. Squire et al., 2021 submitted). Squire et al. showed the extreme bushfire weather of 2019-2020 was a 1 in 200 year event but not the most extreme event we simulated in our large ensemble of forecasts (Figure 2).

The Outcome

Using the nationally important Pawsey infrastructure we have enhanced the CSIRO contribution to the World Meteorology Organisation (WMO) annual to decadal climate forecasting effort ( CSIRO, with the support of Pawsey, is the only recognised centre from the Southern Hemisphere and the only centre that has provided a large ensemble of forecasts (96-members) to this international effort. As a WMO forecasting centre, Australia research is connected to international leaders in climate forecasting. Further, it will enable CSIRO to redirect some of the Northern Hemisphere efforts to climate prediction features of interest to Australia

The forecast dataset is already being used to assess regional climate extremes (Tozer et al., 2020) that may help Hydro Tasmania better manage their water resources. Further, the research team is now exploring how to use our forecast products to improve stock management in the Eastern Tuna and Billfish Fishery.

Most of our effort to date has been directed at producing the forecast datasets, and we are increasingly directing our effort towards extracting useful information from unique datasets.

As part of our contribution to the WMO annual to decadal predictions the following outlook was provided (Figure 1).

Executive Summary of the WMO annual to decadal predictions made on Nov. 1 2020:
• Annual mean global (land and sea) mean near-surface temperature is likely to be at least 1°C warmer than preindustrial levels (defined as the 1850-1900 average) in each of the coming 5 years and is very likely to be within the range 0.9 – 1.8°C
• It is about as likely as not (40% chance) that one of the next 5 years will be at least 1.5°C warmer than preindustrial levels and the chance is increasing with time
• It is very unlikely (10%) that the 5 year mean annual global near-surface temperature for 2021-2025 will be 1.5°C warmer than preindustrial levels
• The chance of at least one year exceeding the current warmest year, 2016, in the next five years is 90%
• Over 2021-2025, almost all regions, except parts of the southern oceans and the North Atlantic are likely to be warmer than the recent past (defined as the 1981-2010 average)
• Over 2021-2025, high latitude regions and the Sahel are likely to be wetter than the recent past
• Over 2021-2025 there is an increased chance of more tropical cyclones in the Atlantic compared to the recent past
• In 2021, large land areas in the Northern Hemisphere are likely to be over 0.8°C warmer than the recent past
• In 2021, the Arctic (north of 60°N) is likely to have warmed by more than twice as much as the global mean compared to the recent past
• In 2021, southwestern North America is likely to be drier than the recent past whereas the Sahel region and Australia are likely to be wetter

References cited:
O’Kane, T. J. et al. CAFE60v1: A 60-year large ensemble climate reanalysis. Part II: Evaluation. Journal of Climate 1–62 (2021) doi:10.1175/JCLI-D-20-0518.1.

O’Kane, T. J. et al. CAFE60v1: A 60-year large ensemble climate reanalysis. Part I: System design, model configuration and data assimilation. Journal of Climate 1–48 (2021) doi:10.1175/JCLI-D-20-0974.1.

Squire, D. T. et al. Likelihood of unprecedented drought and fire weather during Australia’s 2019/2020 megafires. submitted to Nature Climate Change, 17.

Tozer, C. R. et al. Assessing the representation of Australian regional climate extremes and their associated atmospheric circulation in climate models. Journal Of Climate 33, 1227–1245 (2020

Figure 1. Rainfall outlook for 2021 based on forecasts initialised on November 1, 2020. CSIRO and the ensemble mean forecasts predict wetter than average conditions in most of Australia, which is how the year is progressing
• Figure 2: Likelihoods of exceedance. a Likelihoods of exceeding values of FFDIDec. The observed 2019 value of FFDIDec and the associated likelihood of exceeding this value are shown with dashed lines. b As in a, but for DI. c Likelihoods of simultaneously exceeding FFDIDec and DI. The observed 2019 values of FFDIDec and DI are shown with dashed black lines and the likelihood of exceeding both these values together is shown by the text in the upper right. Dashed white lines show contours of a two dimensional kernel density estimate using the forecast data (levels at 2e-05, 1.8e-04, 3.4e-04, 5e-04, enclosing approximately 97%, 68%, 39%, 13% of the data, respectively). d As in c but showing only values of FFDIDec and DI that are more extreme that the observed 2019 values. In all panels, likelihoods are calculated using model data over the period 2014-2023. Note that here “exceeding” a value of DI is defined as having a smaller magnitude since lower values of DI are indicative of increased susceptibility to fire.