NASA Career Award winner uses Blue Waters supercomputer to mine crop yield data

Assistant professor Kaiyu Guan’s work builds off his previous research with satellite and earth system modeling. Prior to coming to Illinois, his PhD and postdoc work focused mostly on how rainfall and other components of the hydrological cycle control plant growth in tropical forests, savannas and farms.

He used some brand new capabilities of satellites to look at previously unviewable wavelengths in the electromagnetic spectrum, challenging intuition and finding that photosynthesis rates in tropical forests were actually higher during dry seasons, which is further confirmed by field-level measures.

“I found this out with the fluorescence wavelength. It’s a very tiny signal but is directly linked with photosynthesis,” said Guan, who holds a joint position with the National Center for Supercomputing Applications (NCSA) as a Blue Waters Professor. “Previous satellites just couldn’t detect that small signal.”

There are other wavelengths that satellites can look at now, such as thermal and passive/active microwaves, and Guan says this information has not been fully used for monitoring crops in the U.S. Corn Belt. He aims to change that.

Using thermal satellite sensors, he can detect water stress in crops, and through insights from field work, extrapolate the effects on crop yield. Using microwave sensors, he can estimate biomass yield from space uninhibited by weather—potentially to 10-meter accuracy.

“Very few have looked at crop growth with the satellite data that I’m using—they’re only using visual and near-infrared,” Guan said. “They’re very focused on one technology. I’m using a portfolio of various satellite data that’s available from NASA and the European Space Agency.”

With these technologies at his disposal, he is using Blue Waters to make a Corn Belt-wide dataset every day during next year’s growing season to monitor crop growth. Blue Waters is being used now to determine how the different sources of data can be integrated together to create the most useful dataset possible. He’s collaborating with NCSA Faculty Fellow and Assistant Professor of Computer Science Jian Peng to accomplish this with deep learning.

“The dataset will do two things: estimate crop type and crop yield. There’s never been a product like that for academics—so it’s very exciting,” Guan said.

In addition to generating this dataset, he’s working with NCSA Postdoc Fellow Bin Peng to use process-based crop models to show how crops in the past have responded to climate change, and how they are going to respond into the future.

The model they’re using pulls the best from two families of models in crop sciences: agronomic and earth system. Guan says that both types of models “naïvely” leave out or over simplify critical information. In the case of Earth System models for example, crop growth is usually too simplified. In the case of agronomic models, the so-called “light use efficiency approach” often leads to inaccurate photosynthesis rates.

“Both families of models have their pros so we combine those together to improve how we simulate critical processes for crop growth,” he said. “And then on top of that, we will integrate new types of data from these satellites to achieve a better simulation of crop yield under climate change.”

Disclaimer: Due to changes in website systems, we've adjusted archived content to fit the present-day site and the articles will not appear in their original published format. Formatting, header information, photographs and other illustrations are not available in archived articles.