Driving Digital Innovation in AI is a series of stories about how Penn State’s Institute for CyberScience researchers are studying machine learning and AI to make these technologies more useful and effective, as well as investigating the impact of AI on people and society.

You can’t do anything about the weather, but with a little help from artificial intelligence (AI), meteorologists and atmospheric scientists say they may be able to better predict how our weather patterns might change in the future, and that could lead to more informed decisions on infrastructure, smarter allocation of emergency services, and a better understanding of how climate change may affect long-term weather patterns, according to a Penn State researcher.

Melissa Gervais, assistant professor of meteorology and atmospheric science and Institute for CyberScience co-hire, said that she is increasingly incorporating machine learning, which is a branch of AI, as a tool to help her own research studying future changes in weather patterns with climate change. According to Gervais, the sheer amount of data involved in analyzing these weather patterns is making AI a perfect research partner.

Until now, meteorologists would­­­­­ need to painstakingly look at as many weather maps as possible to understand and classify different weather patterns, for example, the locations of high-pressure ridges and low-pressure troughs. Researchers like Gervais are now analyzing a large number of climate model forecasts over a longer period of time. Simply looking at these maps and charts might give the scientists a better way to visualize weather scenarios, but the process not only is time-consuming, it also is becoming impractical, considering the rapid growth of data. 

Gervais now uses a type of AI called "self-organizing maps," which might be considered a computerized version of the pattern recognition, similar to how weather forecasters today analyze weather patterns. This technology mimics how the human brain makes decisions but can take the loads of weather data to identify weather patterns. Gervais said this technology can show how weather patterns are fluctuating – or, variability – as well as give better insights into how it might change in the future.

“If it’s going to become more extreme, more variable, that’s important for us to know,” Gervais said. “If it’s going to be wetter with more extreme rainfall in some places and dryer with more droughts in others, that's important for us to know. This kind of work can tell us what is underlying the mean change and help us to understand why these changes are occurring.”

Better predictions may mean better preparedness

Understanding how weather patterns are changing can do more than just help you decide whether you need an umbrella when you leave for work in the morning, according to Gervais. If AI can improve predictions of variability, scientists can prepare people and policy-makers for problems that might arise as the climate changes in their specific regions.

“We could use rainfall as an example because there's a paradigm that the wet get wetter and the dry get drier, which may hold during climate change,” said Gervais. “If you're not used to all that extra rainfall and don't have the right infrastructure for handling that amount of water, this could result in flooding that can lead to loss of human life and destruction of property. In contrast, other parts of the country could get that amount of water all the time – but if they’re used to it they may be better equipped to handle it."

Determining how climate change will affect certain areas could also ease public health issues, simply because it can raise awareness about shifts in weather variability that might help people adapt to the oncoming change. For people who live in cool cities, for example, knowing that there is an increased probability of experiencing hotter summers and extreme heat waves may give them time to install air conditioning.

"It's all about what you're used to and that's one of the big problems: People in different cities and in different locations are adapted to certain weather conditions,” said Gervais. “It's not always that the new conditions are inherently more dangerous -- sometimes they are -- but if they are outside of your experience, you're just not adapted to it."

Gervais said that scientists are only tapping the surface of how AI and machine learning can be used in meteorology and climate science. She expects AI to play a more important role in her own work.

“There are so many things that we can do with it that we haven't even started to think about,” said Gervais. “In my case, I'm just starting to explore using it for prediction.”

Her research group is also working on new ways of using this technique to improve climate predictions on decadal – 10 year – time scales.   

A mix of challenges and opportunities remain for meteorologists and atmospheric scientists, according to Gervais. Deep learning is a more advanced type of AI that scientists are beginning to use to explore weather and climate; however, the inner workings of the method make it difficult to use in science.

“The biggest challenge with deep learning for a scientific application at this stage is that it's basically a black box,” said Gervais. “We don't really know what is happening inside and that's not the mission for a scientist. The mission is understanding and so we really hope we can pull it apart and understand how it is working.”

She believes that the next stage for using AI in science will require computer scientists and scientists from other fields to come together to explore how deep-learning technology actually works.