Bring an umbrella next month: Chinese scientists claim they can forecast rain weeks in advance

PUBLISHED : Wednesday, 12 August, 2015, 7:00am
UPDATED : Wednesday, 12 August, 2015, 6:58pm

A new weather forecasting model developed by Chinese researchers purports to detect heavy rainstorms weeks before they take place, challenging the "butterfly effect" thought to make such prediction impossible.

The scientists' algorithm can predict heavy rain up to 30 days in advance "without false alarm or omission", according to a paper published in the journal Advances in Atmospheric Sciences.

This should be impossible however, due to the "butterfly effect".

American meteorologist Edward Lorenz coined the term in 1961 in a key work on chaos theory. Lorenz showed how in an unpredictable and chaotic system such as weather, tiny events – like the flapping of a butterfly's wings – can have dramatic effects, such as a hurricane.

Under Lorenz's theory, no accurate weather forecast can be made more than two weeks in advance, because tiny disturbances can quickly lead to major fluctuations in weather patterns, throwing the predictions off.

With the rapid increase in computing power and data gathering capability – helped in part by global research on climate changes – scientists have hoped to overcome Lorenz's problem and extend weather forecasting to weeks and even months.

Bigger and bigger super computers have been employed to run larger and larger models, but so far efforts have largely been in vain.

Currently, no weather authorities will provide accurate forecasts beyond 10 days, and most smartphone apps limit themselves to less than a week.

The new Chinese study, led by Xia Zhiye of the Chinese Academy of Sciences' Institute of Atmospheric Physics in Beijing, took the opposite approach to previous, globe-spanning studies.

Using local and regional data, Xia's team examined a hundred heavy rain events around the world in the past few decades, including in China, Japan, India, North America, Mexico and the United Kingdom.

In every case, the model was able to detect signs of the upcoming storm around 10 to 30 days in advance of it hitting.

The unprecedented long range and accuracy of their method came from a "small data" strategy and tight focus, the researchers said.

When running global models on super computers, the uncertainty caused by minor disturbances could be further magnified by taking too many things under consideration. In other words, the bigger the net, the more butterflies there are to catch.

The Chinese team said each local weather system should be examined individually, as each has its own personality and particular quirks. The historical data collected by a village station is far more useful for predicting that area's future weather than information from all across the globe.

The new algorithm is not perfect however, the researchers said.

For instance, it can only predict heavy rain, and would require far more variables – and a corresponding increase in sophistication – to detect other extreme weather such as hailstorms or hurricanes.