Posted 28 June 2012

Forecasting

Neural networks outperform meteorologists

Artificial intelligence (AI) software is outperforming the bureau of meteorology in Australia, by accurately predicting long-term rainfall – AI could soon provide useful daily forecasts from a weather station near you.

AI-enabled neural networking technology has proved a more accurate means of forecasting weather events in 16/17 locations in Queensland Australia, including the rainfall that led to the Ivanhoe dam bursting, researchers from Central Queensland University found. Study co-author Dr John Abbott said an AI enabled system has combined historical records of temperature and rainfall with climate data such as El Niño Southern oscillation, to forecast monthly unseasonal rainfall. The technology could also be used to forecast temperature and, potentially provide daily forecasts.

“This is a much simpler, cheaper method. We are already a little ahead of the bureau of meteorology here in Australia, and we think that in further work there is every reason to believe we can improve on that”, Dr Abbott said. “One off events might be more difficult to analyse and predict using neural network technology. If you're thinking of something like a tornado, that's going to be a lot more difficult because you have limited data to draw on.” Dr Abbott also said that countries such as China and Malaysia were already using the technology for accurate daily forecasts.

The fact that this neural network technology has been applied to meteorological prediction successfully means that the use of such technologies could become a standard practice worldwide – regrettably this may put some meteorologists out of business. Dr Abbott also said that AI enabled modelling cast doubt on CSIRO backed research that found carbon dioxide levels in the atmosphere influenced rainfall patterns.