The Bureau of Meteorology has failed in its forecasting of the weather this summer, but not for the reasons you might think.

In September 2023, the Bureau announced the commencement of an El Niño cycle and warned of the potential for a hot, dry summer. It has turned out to be anything but, at great cost to farmers and communities. While opinions differ on whether we are being too hard or too soft in criticising the Bureau, we see around us too much of what behavioural economists call hindsight bias, including criticisms of an expert merely because their predictions turned out to be wrong.

As the great Australian psychologist and analyst of decision-making Professor Leon Mann taught, the more important question is whether the expert based their prediction on the best available information and decision-making processes at the time it was made. It is on this basis the Bureau’s long-standing and obdurate refusal to investigate the potential applications of artificial intelligence (AI) for weather forecasting is a black mark.

Right now, there is a flood of interest in that potential – exemplified by Google’s experimental GraphCast AI and emerging Chinese technology – to the point where the MIT Technology Review has declared, ‘Weather Forecasting is having an AI moment’ and Nature reported that (Google) ‘DeepMind AI accurately forecasts weather – on a desktop computer’. This level of interest is itself a subset of the general heat around AI, which influential tech entrepreneur Mustafa Suleyman has dubbed, ‘The Coming Wave’.

It is with some ironic detachment I note the valiant efforts of IPA senior fellows, Dr Jennifer Marohasy and Dr John Abbot, to alert the Bureau to this potential in meetings as far back as August 2011. The Bureau rejected their suggestions then and at every point since, but the same advice is now coming from organisations who hardly earn the ‘climate denier’ label, which is so loosely thrown around by staff within the Bureau and elsewhere.

Even the World Economic Forum has reported that, ‘AI can now outperform conventional weather forecasting’, and so too the Jeff-Bezos owned Washington Post reported, ‘Google’s AI weather forecast model is surprisingly accurate’.

It is understood the Bureau relies on models that seek to replicate the underlying physics of the Earth’s climate and weather system, running on supercomputers, using a simulation known as ACCESS-S2. AI, in contrast, uses historical data to build an advanced form of statistical analysis. This is ‘intelligent’ on the basis that rather than relying on programming, it is ‘trained’ by detecting the patterns in big data sets such as temperature and rainfall series and uses them to make predictions about the future.

It has the potential to continually recalibrate as new data is generated, rather than relying on programmers in organisations like the Bureau to ‘tweak’ its models. Importantly, AI models can run on desktop computers and produce results far more quickly than the massive supercomputers desired by weather forecasting agencies the world over.

While one might not expect the Bureau to overnight junk one system in favour of the other, and other government weather bureaus in the West are increasingly outliers in refusing to consider the potential of AI, in China the Academy of Sciences and official agencies – always searching for the next source of advantage – have by contrast displayed greater openness. A peer-reviewed journal published by the Academy in 2020 approvingly quoted the same research from Marohasy and Abbot that was so resolutely ignored by our Bureau.

As a potential half-way house, the European Centre for Medium-Range Weather Forecasts has begun publishing forecasts from GraphCast and other AI models on its website alongside its more traditional outputs.

Abbot and Marohasy have over the past decade now published more than a dozen articles in peer-reviewed journals showcasing the advantages of AI for weather and climate forecasting. Some speak to issues of climate change over long time frames – perhaps a ‘red flag’ in the Bureau – but others are focussed just on near-term (three-month) forecasts of rainfall, in which they demonstrate their models outperform official forecasts.

To use the jargon of AI models, their models display greater ‘skill’ in forecasting, using nothing more than a high-performance laptop with commercially available AI software. The most laborious element of the process is developing clean data sets. Marohasy and Abbot soon realised their best forecasts required the development of their own long datasets avoiding the Bureau’s remodelled – and contaminated – homogenised series.

Their dedication to the task in the face of scepticism not only from the Bureau, but also from climate sceptics uninterested in the scope of AI, is a credit to them. Their advice to others looking to use AI for weather forecasting is to be careful that the historical data inputted is genuine and not remodelled.

In January, Scientific American faithfully reported the view from the traditional weather forecasters’ bunker that a fundamental drawback of using big data is that it is upset by discontinuous or chaotic events. This is a red herring. Notably, it echoes what Abbot and Marohasy were told by a Bureau representative circa 2011: that AI/big data could not work because anthropogenic climate change meant the weather system had fundamentally changed and historical data could not be relied upon.

This however misses the point that one can evaluate the ‘skill’ of the competing systems after the event. To refuse outright to consider such promising alternatives – which more than a decade on are sweeping the world – now risks the organisational redundancy of the Bureau.

As the great Anglo-Irish statesman Edmund Burke said, a state (or organisation) without the means of some change is without the means of its conservation. Creative destruction can benefit consumers, but if the outcome in this case is displacement of the Bureau by commercial and Chinese alternatives, it will be its own fault, and a waste of taxpayers’ money.

Get 10 issues for just $10

Subscribe to The Spectator Australia today for the next 10 magazine issues, plus full online access, for just $10.

SUBSCRIBE 10 issues for $10 REGISTER 3 articles a month

Already a subscriber? Log in