Neural Network Learns How to Forecast Extreme Weather Events
Scientists from Rice University have developed a deep learning system that accurately predicts extreme weather events, like heat waves or winter storms, up to 5 days in advance.
In the article “Analog forecasting of extreme-causing weather patterns using deep learning” published in the Journal of Advances in Modelling Earth Systems on February 3, research team has reported that capsule neural network CapsNets “outperform simpler techniques such as convolutional neural networks and logistic regression”.
Despite all technological advances, current science can not fully provide scientists with an understanding on how extreme weather events are developed, therefore NASA-supported research team opted to use AI that has itself identified patterns, explained Pedram Hassanzadeh from the Rice’s department of mechanical engineering.
Solving a pattern recognition problem, scientists have opted to teach capsule neural network on analog maps, that became almost obsolete in weather forecasting in the 1950s.
Hassanzadeh believes that their network can assist current weather forecast system and help early warning systems:
“Computationally, this could be a super cheap way to provide some guidance, an early warning, that allows you to focus NWP [Numeral weather prediction, as opposed to analog methods] resources specifically where extreme weather is likely.”
As Future Time previously reported, Google researchers also have developed an AI system that outperforms current solutions for short-term weather predictions.
Image credit: NASA