Introduction and history of development
Seasonal prediction at the Bureau of Meteorology is based on a coupled ocean-atmosphere model, a data assimilation system and a strategy for generating forecast ensembles.
Why coupled models, data assimilation and ensembles?
Seasonal prediction depends on variability driven by slow-processes in the climate system, particularly the ocean. Successful seasonal forecasts are often related to a model’s ability to reproduce and predict the slowly changing ocean state (e.g. associated with El Niño) and how this interacts with the atmosphere. It is changes in the ocean that have the biggest influence on Australia’s climate a season ahead. The use of coupled atmosphere-ocean dynamical models for seasonal prediction is now commonplace in major international operational centres.
When we start a forecast, we first have to provide the coupled model with the most up-to-date observed conditions of the ocean, land and atmosphere. We use all the latest observations from ships, satellites, ground stations etc. to construct a picture of what the ocean, land and atmosphere look like today. Observations, like those from Argo floats in the ocean, need to be quality-controlled and blended with other observations in a format that can be used by the coupled model. This process is called ‘data assimilation’.
Once observed data has been assimilated, a picture of how the state of the ocean, land and atmosphere might evolve into the future is then generated using the coupled model - the forecast. The coupled model includes component models of the atmosphere, ocean and land surface. These models are what we call “dynamical” models; they are based on mathematical equations representing the laws of physics, including for example, Isaac Newton’s laws of motion. They are essentially computer simulations of the weather, and land and ocean processes. Dynamical atmospheric models are what currently provide us with our daily weather forecasts.
One of the benefits of coupled models is that many forecasts can be produced - this is called an ensemble. Ensembles provide us with the most likely outcome of an event, as well as an indication of the level of uncertainty in a given forecast. If the forecasts in an ensemble are all close together (e.g. 90% of them are forecasting that an event will occur), then we can have confidence in the forecast. If they all differ significantly they can tell us that there is considerable uncertainty in the future and they give us the range of possibilities. This is very important for seasonal prediction because there is considerable uncertainty in seasonal predictions - some due to natural uncertainty such as the ‘chaos’ factor (which will always exist, even with a perfect model) and some due to model and observational errors.
Unlike existing statistical forecasting systems, coupled models are not limited by historical relationships and can forecast a new set of climatic conditions. For example, because they simulate the real world they have the potential to predict how the impacts of one El Niño might be different to those of another. The outputs of coupled models are also appropriate for feeding into downstream applications models, such as crop or hydrology models.
The other nice thing about using dynamical models is the potential for providing forecasts across a whole range of timescales - a concept often referred to as seamless prediction. Forecasts and warnings on a range of weather and climate timescales, from hours to years into the future, have the potential to inform a variety of strategic and tactical decisions for many different sectors of society.
Compared to weather forecasting, coupled model seasonal forecasting is still in its infancy. Still, great potential lies ahead. By continuing collaborative work and investing in further improving our system we will reap the full benefits of coupled models and provide Australia with a seasonal forecasting capability second to none.
Development of coupled-model seasonal forecasting at the Bureau
The first version of the Bureau’s coupled-model forecast system, called the Predictive Ocean Atmosphere Model for Australia (POAMA), was developed in a joint project involving the former Bureau of Meteorology Research Centre (BMRC) and former CSIRO Division of Marine Research, with support provided by the Climate Variability in Agriculture Program (CVAP), a consortium of rural research and development corporations managed by Land and Water Australia. The core of the research was carried out by scientists from the Oceanography Group at BMRC and scientists from the Oceans and Climate Group at CSIRO Marine Research. This first version went operational in October 2002 and produced routine forecasts of El Niño conditions.
A new version, POAMA-1.5, replaced POAMA-1 as the Bureau’s operational dynamical seasonal prediction system in September 2007. Operational products consisting of various Sea Surface Temperature (SST) indices for the Pacific and Indian Oceans were made available on the National Climate Centres web site. Another set of operational products focussed on the prediction of extreme ocean temperature in the Great Barrier Reef (for warnings of coral bleaching) and were made available on the Bureau’s Oceanographic Services web site. POAMA-2 (version P24) replaced POAMA-1.5 in 2011 and was the first operational pseudo multi-model system, using three different versions of the atmospheric model in the generation of the forecast ensembles. The forecast products mentioned above were upgraded to be based on this new system. Seasonal forecasts from this system were produced twice monthly.
A new version of POAMA-2, version M24, was then developed in order to facilitate the generation of multi-week forecasts - POAMA became a seamless multi-week to seasonal forecast system. From 22 May 2013, this version became the new official Bureau seasonal forecast model. The Bureau’s monthly and seasonal climate outlook is based on forecasts from this new system (as are the products mentioned previously). In addition to the official Bureau products, a large suite of experimental forecast products are available on the POAMA research web site for research and trial purposes.
Successive versions of POAMA have demonstrated improved capability to predict Australian climate and the important drivers, such as the El Niño Southern Oscillation.
Correlation skill of forecasting anomalies in the Eastern Pacific (the NINO3 region) for month-3 of the forecasts for all forecasts in 1982-2006 from POAMA-1, POAMA-1.5 and POAMA-P24 (blue). There is clear improvement in skill with subsequent system versions. The level of skill is compared with international systems: The European Centre for Medium-Range Weather Forecasts (ECMWF) System 3; the Japanese SINTEX system; the USA’s National Centers for Environmental Prediction CFSV1 and CFSV2 systems.
Forecasts of SST anomalies in the Eastern Pacific (the NINO3 region) for the winter and spring of 2005 from the POAMA-1 real-time system (left), and the same forecast done retrospectively with the currently operational version of POAMA, version 2 (right). The green line in the right plot shows the observed SST. Clearly POAMA-2 has done a better job of predicting this event.
Reliability diagrams of the probability that a) rainfall, b) maximum temperature and c) minimum temperature in the first season of the forecast is in the upper tercile (lower tercile for minimum temperature) for POAMA-1.5 (green), POAMA-P24 (blue) and POAMA-M24 (red) based on all start months in 1980-2006 and model grid boxes over Australia. The closer the line is to the diagonal, the better the reliability. The current operational version of POAMA (M24) is the most reliable.
The Bureau’s seasonal prediction system continues to be developed. The next version will be based on ACCESS (Australian Community Climate and Earth System Simulator), already in use for seven-day weather forecasts and climate change projections. ACCESS consists of state-of-the-art modelling software developed by the Bureau and CSIRO in collaboration with Australian universities and several international organisations. The seasonal prediction component of ACCESS is expected to be a significant improvement over the current operational model. One of the most obvious improvements will be forecasts with more regional detail. At the moment the spatial resolution of the forecasts is about 250km. In the next version, the resolution will be about 60 km in the Australian region. At this resolution the model is able to, for example, differentiate between the climates of western and eastern Tasmania and better represents the Great Dividing Range, which plays a key role in the spatial distribution of rainfall. Increased resolution will also improve the representation of important large-scale climate drivers, like ENSO, potentially leading to better multi-week and seasonal forecast accuracy over Australia.
ACCESS-S, “S” for seasonal prediction, will replace the name POAMA to reflect the use of the ACCESS modelling infrastructure for all weather and climate applications by the Bureau of Meteorology. The next version of the POAMA system will be called ACCESS-S1.
POAMA development and milestones: 2002-present