You are here

Climate Extremes and Machine Learning

October 6, 2017

Big data informs just about everything Ansu Chatterjee does at the University of Minnesota. Along with being an associate professor in statistics, he directs the Institute for Research in Statistics and its Applications.

Currently he’s studying the impact of climate extremes on societies. Droughts are one such extreme, and they are becoming more prevalent around the world. Droughts lead to changes in agriculture and food production. This leads to changes in what people do for a living, which leads to people moving to other places. As a result, societies become less stable—and that leads to even more changes, for better or usually for worse.

Through the lens of big data, Chatterjee is studying this “long chain in which climate plays a part.”

Although his research looks at impacts on humans, Chatterjee relies heavily on machine learning. This area of artificial intelligence gives computers the ability to learn without being explicitly programmed.

“When computers came along, statistical software was written in a way that imitated the process of doing calculations by hand or with a scientific calculator,” he says. “Machine learning takes a totally different approach. The algorithm that you implement is not just an imitation of what you would do if you had a small dataset.”

Chatterjee specializes in resampling techniques, which involve parallel computing—or multiple computers working in unison. The more computers, the more speed and efficiency. This is important, because climate models involve complex dependency patterns.

“The dependency between temperature, wind speed, and precipitation patterns in the Gulf Coast isn’t the same as it is in Minnesota.” Modeling these scenarios requires intensive computations that only machine learning can handle.

That’s where the datasets get so big, taking into account many decades of weather patterns, locations, and more. “Climate is long-term weather. It’s not any given event. And that long-term perspective is changing.”

As they gather and sort through all this data, Chatterjee and his colleagues are careful to distinguish between what’s accurate and what is just hype. With politically-charged topics like climate change, he remains grounded in statistics. “When it comes to the data world—the world of numbers and satellite images and anything a weather instrument collects—there is no misinformation.”