Time for climate models to be linked
GroundCover™ Issue: 111 | Author: Sarah Cole
The key to improving Australia’s seasonal (long-range) climate forecasts may lie in finding a consensus among different climate models, says leading climatologist Dr Roger Stone.
Four months ago, Dr Roger Stone was asked to work out when new dynamic climate forecasting systems were useful for seasonal forecasts, and when they may have been less useful.
New forecasting systems are based on general circulation models (GCMs), and Dr Stone’s initial assessment has looked at the past six to seven years of forecasts. His task is to assess usefulness for the grazing industry, but his analysis also has crucial implications for grain growers.
He is exploring answers to the following questions.
- How well do different forecasts run at different times of the year?
- When do they not do so well?
- How much better are predictions than chance?
“My analysis looks at consistency,” Dr Stone says. “When rainfall was predicted to be above or below the median for three months, how often was it right for each small region of Australia’s grazing lands?”
Working in a team at the University of Southern Queensland in the International Centre for Applied Climate Science, funded by Meat and Livestock Australia, Dr Stone has compared six years of forecasts from three different models – a total of 72 forecasts:
- the Bureau of Meteorology’s Predictive Ocean Atmosphere Model for Australia (POAMA) – a computer model that predicts the weather using occurrences in the oceans and atmosphere;
- forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF); and
- GloSea forecasts from the UK meteorological office.
Dr Stone’s team has also compared the usefulness of statistical forecasting based on the Southern Oscillation Index (SOI) (Figure 1). This system analyses the patterns of the phases of SOI and has operated through Queensland Government agencies for 20 years.
The key to how well the forecasts performed has been linked to whether or not a strong El Niño–Southern Oscillation (ENSO) signal existed.
The best forecasts have occurred as El Niño or La Niña conditions start and continue.
The ENSO describes the back-and-forth climate cycle between El Niño and La Niña.
El Niño conditions generally mean below-average rainfall over much of eastern Australia, and La Niña conditions often result in above-average rainfall across Australia.
“All of the forecasts did their best predicting rainfall patterns during an ENSO year in what we call ‘core ENSO periods’, especially for eastern Australia,” Dr Stone says.
“This is basically during definite El Niño or La Niña conditions.”
For instance, the POAMA model made excellent predictions for northern and eastern Australian in 2010, 2011 and 2012 – clear ENSO years with very strong La Niñas, but predictions were relatively unreliable in 2013, when indications wavered back and forth.
This year, all climate models indicate an El Niño is coming. If this occurs, then climatologists anticipate that quantitative forecasts of rainfall and temperature for many regions of Australia will have high accuracy, no matter what the forecasting model.
The real challenge, scientists say, are the ‘middle years’ such as 2013, where there is no clear or sustained signal either way. Weak ENSO years appear to have no signal at all in terms of rainfall.
Dr Stone emphasises that his initial assessment has just covered the past six years. He says that to obtain a more robust assessment of seasonal climate forecast value in Australia, a more comprehensive assessment of forecast outputs (linked to key farming decisions) over the past 15 to 20 years is needed.
While Dr Stone’s findings might suggest it is only worth switching on each forecast model when it is more likely to do a good job, climate researchers are already pointing to a new approach; one that combines the models. This is a considerable technical challenge but there is agreement among researchers that it is the next step to achieving increased accuracy and reliability.
Forecasters say they can approach this multi-model forecasting in three different ways:
- adjusting the outputs of computer models in relation to previous weather records (the hybrid approach);
- using the models that work best for each region at particular times of year (the consensus approach); or
- adjusting computer models for specific weather indicators such as rainfall (the bridging approach).
Scientists from major research providers such as the Bureau of Meteorology (BoM), CSIRO and universities are now considering how to bring together multiple models.
Dr Oscar Alves, from the Centre for Australian Weather and Climate Research, is keen to collaborate with other agencies on this effort to increase the accuracy of seasonal forecasting.
“BoM and CSIRO are convinced that there are mutual benefits in working together, in both climate prediction and in applications for users,” he says.
Dr Mark Howden, from the CSIRO Climate Adaptation Flagship, believes the time has come in the development by different agencies of various forecasting models to now move collaboratively towards a combined model.
The science manager of the GRDC-supported Managing Climate Variability (MCV) program, Simon Winter, says that the idea to bring together several models from different sources is an opportunity to deliver better outcomes using the latest science and techniques.
The next thing to work out, he says, is what such a national approach will look like.
MCV is bringing together relevant research organisations to look at the potential for a collaborative project on improving climate forecasting specifically for agriculture.
CSIRO’s Dr Howden is quietly confident: “We’re looking forward to the process, and how it can add to existing systems.
Australian climatologists aren’t starting with a blank sheet of paper – we already have a whole raft of existing forecasting systems. It’s about supplementing the work we’ve already done.”
Dr Oscar Alves
03 9669 4835
Dr Roger Stone
07 4631 2736
Dr Mark Howden
02 6246 4118
02 6281 5257
02 6754 3389
02 6743 4263
GRDC Project Code MCV00008
Region North, Overseas, South