Weather forecasting is not just predicting whether if it’s going to rain tomorrow. It also involves long-term predictions that are useful for policy and conservation. For instance, predicting when the next hurricane or rainstorm is going hit is especially useful.

However, predicting the weather on an hourly basis, known as now-casting. is also incredibly useful. But, now-casting is harder than long-term forecasts as they are susceptible to a lot of variable factors.

Hence, an AI trained by machine learning might be better equipped for the job. And that’s exactly what the folks at Google’s DeepMind have done. Using AI, researchers say they were able to accurately predict when and where it’s going to rain within the next 90 minutes.

The AI was developed by DeepMind in collaboration with the UK’s meteorological office and can predict short-term weather more accurately than conventional methods.

DeepMind, based in the UK is a subsidiary of Alphabet – Google’s parent company – and has achieved numerous high-profile successes previously with neural networks, including training it to play the notoriously difficult game Go and to investigate how protein’s fold.

In a paper published in the journal Nature, researchers outline their new prediction model called the Deep Generative Model of Rainfall or DGMR to predict the chances of rain within the next one to one-and-half hours.

DGMR uses the insane computational capabilities of neural networks to weed out vagueness within the machine-learning process, leading to highly reliable and accurate predictions.

Cover Image: Shutterstock