THE CLIMATE KELPIE BLOG: Better forecasts with the move from ACCESS-S1 to ACCESS-S2

Posted by BCG on 29th March 2022

ACCESS-S version 2 (ACCESS-S2) is a major upgrade to the Bureau of Meteorology’s seasonal prediction system. It went live late last year, replacing the previous version, ACCESS-S1.

ACCESS-S supports the Bureau’s multi-week and seasonal climate outlooks and serves several external customers. It also provides input data for the seasonal forecast component of the new Australian water outlook service – driving a water balance model to produce seasonal forecasts of soil moisture, evapotranspiration and runoff.

“The development of ACCESS-S2 was a huge collaboration across the research and development, computing and customer-facing sections of the Bureau,” said Oscar Alves, the Earth System Modelling Section Head of the Bureau’s Research Program. “It also involves an important linkage with the United Kingdom Meteorological Office (UKMO).”

How does ACCESS-S work?

At the core of ACCESS-S is a coupled climate model. A climate model is a computer simulation of the real world, based on the laws of physics. A coupled model is when different sub-models are linked. ACCESS-S brings together global models of the atmosphere, ocean, land and sea-ice.

However, more than just the model is needed. A forecast needs to start with the best estimate of the current state of the atmosphere, ocean and land. The process of getting observations and putting them into the model for the start of the forecast is called data assimilation – a process that is nearly as complicated as the model itself.

All the latest observations from ships, satellites, ground stations etc. are used to construct a picture of what the ocean, land and atmosphere look like today. The observations need to be quality-controlled and put in a format that can be used by the coupled model.

The model can then freely evolve to make the forecast. But, the Bureau doesn’t just run one forecast. There are uncertainties in the way the weather can evolve. So, several equally likely outcomes are run—this is called an ensemble of forecasts.

Also, for seasonal prediction, a set of forecasts need to be run retrospectively for a period in the past. These are known as hindcasts. They are used to determine the accuracy of the forecast system, as well as to calibrate the real-time forecasts.

Finally, the forecast data needs to be post-processed to produce forecast products, like those on the Bureau’s seasonal outlook webpages.

So really, ACCESS-S is much more than just a model.

Figure 1: Key components of a seasonal prediction system. ACCESS-S is a dynamical (physics based) global modelling system. It comprises: (1) a data assimilation system (combining data gathered through observations with model output for the start a forecast); (2) a coupled atmosphere–ocean–land–ice model; (3) a strategy for running forecast ensembles (i.e., more than one forecast); (3) data post-processing to generate forecast products; and (4) a hindcast set to evaluate the performance and calibrate the real-time forecasts. Source: BoM

What are the main differences between ACCESS-S1 and ACCESS-S2? What does it mean for seasonal prediction?

The major change between ACCESS-S2 and ACCESS-S1 (version 1) is in the data assimilation – in particular the ocean and land starting conditions for the forecasts.

“Changing the data assimilation can change the behaviour of the forecast system, sometimes substantially,” said Bureau Principal Research Scientist, Debbie Hudson. “In ACCESS-S2 we now have an in-house data assimilation system to produce ocean initial conditions, which we previously relied on from the UKMO. By doing it ourselves, our forecasts are more timely which in turn means they’re more accurate for our customers,” said Debbie.

There are also indications these ocean initial states are better in ACCESS-S2, and this improves the forecasts of El Niño–Southern Oscillation (ENSO), particularly during autumn. “This is a pleasing result, since it is a challenging time of year for predictions, with all models experiencing some difficulty in predicting the development of ENSO, ” said Debbie.

Figure 2: Correlation skill for NINO3 sea-surface temperature anomalies (which is an indicator of ENSO) from ACCESS-S1 (red) and ACCESS-S2 (blue) for forecasts that start in May. The x-axis is the forecast lead time in months. Higher correlation values for ACCESS-S2 indicate higher skill. Source: BoM

“We have also improved the way soil moisture is initialised and made it more realistic in ACCESS-S2,” said Debbie. “This has led to a clear improvement in the accuracy of maximum temperature forecasts over northern, eastern and western Australia in ACCESS-S2 in the first month of the forecast and in the winter period.”

Figure 3: Correlation skill of forecasts of monthly maximum temperature anomalies (differences from usual) for the month of May from (a) ACCESS-S1, (b) ACCESS-S2, and (c) the difference in the skill between ACCESS-S2 and ACCESS-S1. Forecasts were initialised on 1st May for the period 1990-2012. The warmer colours in a) and b) indicate higher skill. The red colours in c) indicates where ACCESS-S2 is better than ACCESS-S1. Source: BoM

Some of the other ACCESS-S2 changes include:

  • A longer hindcast set – 38 years compared to 23 years – which is better for forecast calibration and improves estimates of the skill
  • Additional outlooks of extreme climate conditions for end users arising from the Forewarned is Forearmed project
  • Enhanced post-processing, which provides more data outputs to users and will make the addition of tailored forecast products easier in the future.

ACCESS-S supports several seasonal prediction projects. “We have major projects with the agriculture sector, such as Forewarned is Forearmed and the Northern Australia Climate Program. These projects play a key role in our research to understand and improve the forecast system. This ultimately leads to more accurate and useful forecasts for our customers,” said Debbie Hudson.

For more information contact:

Debbie Hudson,

Leave a Reply