June brings with it a few things – the peak of the mango season and the official arrival of summer. Through what was a pleasant June, citizens in the city were treated to showers and even a hailstorm in certain parts of the city. While long-time residents of the city are used to drastic swings in the weather and temperature over the course of a day, it is interesting to investigate if technology can play a role in better predicting the weather. While hailstorms offer a welcome respite from the heat, they can also cause damage. More severe storms can do a lot more than just dent a car.
This ability to better predict the weather is important in the Indian context, a country where a healthy monsoon signifies a strong economy. This is one sector which is also gaining significance importance from a governance standpoint in the country. The Ministry of Earth Science (MoES) launched a high-performance computer (HPC) system named Mihir earlier this year, with the aim of improving India’s weather forecasting. With this, India occupies the fourth position, after the United Kingdom, Japan, and the USA, in the list of dedicated capacity for HPC resources for weather and climate use.
According to the World Economic Forum along with PwC and Stanford Woods Institute for the Environment, disruptive technologies, when harnessed right, could help build solutions to overcome challenges pertaining to weather resilience.
Specific to hailstorms, a few short minutes can lead to battered crops, dented cars, and even accidents, resulting in significant losses. In a talk at NVIDIA’s GPU Technology Conference, David Gagne II, a postdoctoral fellow at the National Center for Atmospheric Research, delved into the topic a little further, “Because hail is so damaging, we want to be able to forecast it more reliably so people can take shelter and protect their property.” Gagne is among a group of scientists at NCAR who are investigating how GPU-accelerated deep learning can be used to more accurately forecast the likelihood of hail, while also predicting where it will fall and how large it can be.
The problem with current hail forecasts
First, how exactly does hail form? When upward air currents from thunderstorms are strong and carry water droplets well above freezing temperatures, these droplets become hailstones, which grow in size as more water freezes to them. Beyond a certain weight, the hailstones become too heavy for updrafts and they fall to the ground.
Meteorological departments and other scientists have explored and currently, have multiple ways of predicting storms. However, according to Gagne, most of these models have shortcomings that result in false alarms and, worse, missed storms. Another avenue that scientists have explored are machine learning-based forecasts.
“Machine learning can produce reliable severe weather forecasts but it struggles with learning spatial patterns,” Gagne said. These patterns show what areas in a city will be affected by rain or hail.
The potential of AI
However, in contrast to machine-learning based forecasts, using a deep learning model allows for easier integration of spatial patterns, time, and physical understanding of conditions finds a paper by Gagne and other scientists published in the American Meteorological Society Journal.
Application of modern AI techniques to high-impact weather forecasting improves the ability to sift through the deluge of big data to extract insights and accurate, timely guidance for human weather forecasters and decision-makers. Artificial Intelligence can also reveal new knowledge in data representations such as Doppler radar maps, the multi-coloured maps viewers see in weather forecasts on TV.
“I wonder if deep learning can look at these images and see what meteorologists see, or if it sees something different,” Gagne said.
Predicting hail that damages
Gagne and his team have used GPUs and the TensorFlow deep learning framework to train models that predict hail roughly an inch in diameter. “That’s the size where it will ding your car or mess up your roof,” Gagne said.
So far, Gagne’s team’s deep learning models have produced fewer false alarms and have higher accuracy. The implications of this are far-ranging – people have the chance to move to protected areas and secure property, and this could even allow airports to reroute air traffic, Gagne said.
Gagne and other scientists are also using deep-learning models to help predict the type of expected precipitation and storm lifetimes. All this, so rain can be enjoyed in the best way – on a balcony with a book and cup of tea!
Vishal Dhupar is NVIDIA’s Managing Director in South Asia.
(Disclaimer: The views and opinions expressed in this article are those of the author and do not necessarily reflect the views of YourStory.)