The bureau has warned that its long-term climate forecast capability may have peaked and could even have started to decline.

Speaking at the first public hearing of a federal inquiry into long-term forecasting yesterday, the bureau's acting chief climatologist Michael Coughlan said that as the climate changed, so did the margin for error when using traditional statistical modelling.

Dr Coughlan said while Australia had 100 years of statistical data to call on for long-term forecasts, changes in climate over the past two decades had affected the accuracy of forecasts covering periods of one to several months in advance.

"The problem is that most statistical forecasting is based on a stable climate where your average is a good representation of what the climate in the future is going to be," he said.

"(But) when it comes to a representation of the climate we have now, we don't have 100 years of records. We've only got about 20 years of records and that's not reliable enough."

Dr Coughlan said funding was needed for greater "super computing capacity", allowing long-term forecasters to conduct more complex tasks, as well as more people power.

Head of climate analysis David Jones told the panel that the bureau needed to make the leap from statistics-based forecasting to model-based forecasting.

"The world is changing and it's very difficult to find pasts that look like the present," Dr Jones said.