Get more content like this in your Inbox monthly!
Our newsletter consists of curated articles from our top authors.
Google's new data-driven physics-free approach provides almost instantaneous weather predictions.
Weather predictions are very crucial for our day to day lives, and yet instantaneous weather forecast remained a holy grail for meteorologists and weather agencies across the globe despite the advancement in satellite imaging and super computers. However, as with many other fields, machine learning is proving to be a game-changer in weather prediction as well.
A recent paper published by researchers at Google titled "Machine Learning for Precipitation Nowcasting from Radar Images" explains how Google's new data-driven physics-free approach provides localized instantaneous weather predictions. According to the paper, instead of resorting to gigantic supercomputers to crunch the weather data, Google is treating weather prediction as sort of an image-to-image translation problem. The researchers then used convolutional neural networks (CNNs) for current state-of-the-art in image analysis, to produce weather forecast.
The paper presents "new research into the development of machine learning models for precipitation forecasting that addresses this challenge by making highly localized 'physics-free' predictions that apply to the immediate future."
The most prominent advantage that machine learning brings to weather forecasting is "that inference is computationally cheap given an already-trained model, allowing forecasts that are nearly instantaneous and in the native high resolution of the input data. 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," wrties Jason Hickey, Senior Software Engineer, Google Research in a blog post.
Presently, the major challenges in weather forecasting are the lack of computational resources and unequal measurements of weather data parameters across the globe.
"Computational demands limit the spatial resolution to about 5 kilometres, which is not sufficient for resolving weather patterns within urban areas and agricultural land. Numerical methods also take multiple hours to run," explains Hickey. By contrast, "nowcasting is especially useful for immediate decisions from traffic routing and logistics to evacuation planning."
When compared to other top weather forecasting models, Google's ML models found to have produced superior results, according to the research paper.
Source: Google AI Blog
Researchers can now detect backdoor attacks
DeepMind's AlphaFold marks major milestone in predicting protein structure