Those of us of a certain age may have noticed a gradual improvement in the accuracy of weather forecasting over the years. That, of course, is no coincidence as we have indeed become much better at atmospheric modellings, as we have improved in every data-driven field of knowledge.

Not only do meteorologists have a deeper understanding of the science behind weather patterns thanks to continuing academic research over the years but they also have access to much richer pools of data. The Application of AI algorithms to hunt for previously undetected patterns in growing data lakes is now taking things one step further.

Deepmind, the UK-based Google sister company that is one of the world’s leaders in the field of AI, has developed an algorithm that is proving itself a cut above anything else out there when it comes to “nowcasting” of near term weather patterns.

Meteorologists have been quite good at predicting general weather for up to a few days in advance for some time. If the weather forecast provided on Monday says Wednesday is likely to be 17 degrees Celsius with low wind, some cloud cover and scattered showers, there’s a high chance that is pretty close to what will happen on Wednesday.

And we’re getting better and better at accurately predicting the weather further in advance with powerful computers modelling the atmosphere to predict its evolution. Where the Met Office and other weather forecasters still struggle most is short term predictions for how the weather will unfold over the next couple of hours.

For many sectors of the economy, accurate “nowcasting” would be hugely valuable. The construction industry would be one as well as anyone planning outdoor events. Companies like Uber and Deliveroo would also have some prior warning of an hour of rain about to hit and be able to alert drivers working flexible hours a busy period is coming. Accurate near term weather intel would also be helpful to emergency services who have to deal with the fallout of extreme weather events.

A paper recently published by Deepmind researcher Shakhir Mohamed in the scientific journal Nature explains how AI technology he and colleagues developed proved itself more accurate at nowcasting than other models currently available as much as 89% of the time.

The Deepmind algorithm offers an incremental rather than revolutionary step forward in the accuracy of nowcasting but the very nature of AI is that it improves over time as it continues to learn from a growing pool of data. Together with further improvements to the algorithm itself, the research project could, within a few short years, lead to a major advancement in the accuracy of weather forecasting for the couple of hours ahead.

One of the core techniques used by nowcasting is to look at where weather events like rain are now and taking into account factors like the Earth’s rotation, wind speed and other variables, estimate where it will move over the next minutes and hours. Modern big data resources mean we are constantly gathering more and more data on how rain and other localised weather events move and strengthen or weaken over time. With each new day of data, Deepmind’s algorithm will continue to improve.