Forecasting weather is notoriously difficult and is one that typically requires the use of powerful computers to predict. However, a group of researchers from Google says that the use of machine learning (ML) can allow for high resolution forecasts that are also “nearly instantaneous”.
Machine learning for weather
The work on the use of ML for weather forecasting was detailed in a blog post this week by senior software engineer Jason Hickey at Google Research. Published in a paper titled “Machine Learning for Precipitation Nowcasting from Radar Images”, it was able to generate accurate rainfall predictions at a 1km “resolution” up to six hours ahead of time.
“This precipitation nowcasting, which focuses on 0-6 hour forecasts, can generate forecasts that have a 1km resolution with a total latency of just 5-10 minutes, including data collection delays, outperforming traditional models, even at these early stages of development,” wrote Hickey.
The use of an already-trained model means that making an inference about the weather is computationally cheap and extremely fast, offering the native high resolution of the input data, explains Hickey.
Indeed, the ML method outlined by the team is effectively instantaneous compared to alternative methods. The latter is hindered having to process up to a 100TB of data that adds a computational latency of between 1 and 3 hours. This makes ML superior for very short-term weather forecasting.
Prelude to wider ML use
ML is one of the most popular applications of artificial intelligence (AI) today and relies on algorithms and statistical models that let computer systems make predictions based on patterns and inferences.
The use of ML for weather forecasting might be of interest to logistics firms or even retail businesses that are affected by the weather. More importantly, this research underscores the potential of ML for a wide variety of use cases.
Photo credit: iStockphoto/g4gary