Forecasting climate

Forecasting Australia's climate

To help Australians plan ahead, we provide regular long-range forecasts. They extend beyond 7-day weather forecasts to indicate weather and ocean behaviour for weeks to seasons ahead.

To deliver long-range forecasts, we use:

• Australian and international data collected from different levels of the atmosphere and ocean (observations)
• a sophisticated computer model of the Earth.

This model runs on our supercomputer. It's like a virtual planet Earth, simulating behaviour of the atmosphere, ocean, land surface and sea ice.

The observations feed into the model, so it represents the oceans and atmosphere as they are now. Observations for Australia are collected through our observation network.

We run the model forward in time to forecast likely conditions ahead.

Australia's climate model – ACCESS-S

Our climate model for weekly to seasonal and longer-range forecasts is called the Australian Community Climate Earth-System Simulator – Seasonal (ACCESS-S).

ACCESS-S is a state-of-the-art dynamical (physics-based) forecast modelling system. It's a collaboration between the Bureau and the United Kingdom Meteorological Office.

Based on grids

Like similar models, ACCESS-S divides the atmosphere and oceans into grid boxes. The grids are both vertical and horizontal.

The grid boxes for the atmosphere and land in the Australian region are about 60 km square. At this size, the model can represent the different climates of the Great Dividing Range and Australia's eastern seaboard.

For the ocean in the Australian region, the grids are about 25 km square. This means the model can distinguish small-scale currents and eddies.

As the model steps forward in time, it calculates how much heat, moisture and wind move into and out of each grid box. This takes into account the way the atmosphere interacts with the ocean and land surfaces.

Producing long-range forecasts

Each long-range forecast from ACCESS-S is based on 99 individual forecasts. To produce these, the model is run 99 times with slightly different initial conditions.

This allows us to account for differences between our estimate of the current state of the atmosphere and oceans, and the real world. It also helps us capture a range of likely scenarios.

This process is called ensemble forecasting. It's why most of our long-range forecast information is provided as probabilities. For example:

• If 80 of the 99 forecasts predict above average rainfall, then the chance of above average rainfall is close to 80%.
• If only 50 of the 99 forecasts are wetter than average, then the forecast shows a close to 50% chance of above average rainfall.

Climate model accuracy

To measure the accuracy of the model, we run it over a period in the past – usually about 30 years.

We compare forecasts made using historical observations – called hindcasts – with what actually happened. Hindcasts are produced using a similar process to forecasts.

We use as long a period as we can to see how the model behaves in different situations. This helps capture climate events such as El Niño and La Niña that don't happen every year.

But there are some limits when using historical records. For example, detailed ocean satellite observations only became available when the satellite era began in the 1980s.

Improving long-range forecasts

Some of the factors that will help us provide more accurate long-range forecasts include:

• advances in science
• progress in supercomputing – ACCESS-S needs vast computing power to complete its calculations and provide constant updates
• improvements in the quality of observations, how they are collected and fed into the model.

Many processes in the atmosphere, such as thunderstorms and clouds, happen at scales much smaller than the current grid boxes in our model. As we shrink the grid box size, the resolution is greater and forecasts have more local detail.

It's like looking at a road map at the regional level, showing highways and main roads, then zooming into your town or suburb, with every street showing.

Using international climate models

When forecasting months and seasons ahead, we also look at climate models from other meteorological agencies. This includes models from:

• meteorological agencies in Japan, the United States, the United Kingdom, France and Canada
• a collaboration between more than 30 European nations. It's called the European Centre for Medium-Range Weather Forecast (ECMWF) model.

Climate models from around the world have common elements. They also differ, particularly in how they calculate smaller-scale weather processes. Combining the results from various models can sometimes be more accurate than relying on any one model.

This is relevant when predicting indicators for the major climate drivers months in advance.

We include an average of the international models for the major climate driver indicators in the Climate Model Summary on our current website.