Terrestrial ecosystems are greatly impacted by global warming, according to Prof. Yiqi Luo, integrative plant science. In a July 2023 study, Luo utilized various data synthesis models to highlight the importance of soil in carbon sequestration, which is the process of capturing and storing carbon dioxide from the atmosphere.
According to Luo, soil is one of the most important carbon storage systems among terrestrial systems. The microbial processes in soil are the driving determinants of its carbon use efficiency, a ratio between how much carbon is released back into the atmosphere and how much is used for growth and, thus, absorbed from the air and stored away.
Until recently, however, the data assimilation models and machine learning used to measure sequestered carbon have been relatively unreliable. Without a dependable way of acquiring these measurements, it may become difficult for the carbon market —a trading system where companies can buy and sell credits that allow them to emit a certain amount of greenhouse gasses each year — to sustain itself.
“For example, you could have one pound for an apple but your scale is so uncertain that it reads 10 pounds,” Luo said. “The uncertainty is so huge and that is the major issue here. Our scale to measure carbon sequestration is just not accurate.”
Luo mentioned how carbon measurements in soil have historically resulted in a low R-squared value — a numerical value from zero to one that indicates how well the independent variable can predict the outcome — suggesting that the measured carbon levels cannot be explained by various carbon cycle processes most the time and cannot be applied in actuality.
In the past, Luo utilized a process-guided deep learning and data-driven modeling approach that combined machine learning, big data, earth system models and data model fusion techniques. The process, also known as PRODA, evaluated the efficacy of microbial carbon use efficiency compared to six other factors — plant carbon inputs, carbon input allocation, non-microbial carbon transfer, substrate decomposability, environmental modifications and vertical transport. .
PRODA found that microbial CUE is roughly four times more effective for carbon storage than other factors because the microbes found in soil use a much higher percentage of the absorbed carbon for growth rather than metabolism. This method allows carbon to be stored underground rather than reemitted into the atmosphere.
Luo also noted that PRODA generated an R-squared value of 0.54, which — although higher than previous techniques — still needed improvement, according to Luo.
He transitioned to developing a newer technique called Biogeochemistry Informed Neural Network that generates R-squared values at around 0.60 and even sometimes 0.70.
A neural network is a subdivision of machine learning that is based on the human brain and mimics the firing of neurons between transmitters. It is a way of teaching computers to develop on their own and limit the amount of human intervention needed to carry out specific tasks. Neural networks — like humans — are able to learn and evolve by themselves.
Luo hopes to further research and develop BINN in order to disseminate this technology to other agricultural scientists
With the support of the United States Department of Agriculture and $60 million in funding, Luo believes that machine learning methods can continue to improve the accuracy and reliability of carbon measurements, as well. Especially as artificial intelligence gains an increasing amount of traction, Luo said that AI — machine learning in particular — has the potential to serve as an incredibly powerful tool to learn from big data and gain a better understanding of carbon sequestration and carbon removal.
Although Luo focused more on methods to process big data and improvements in statistical accuracy, he has also worked alongside a team of directors and faculty members who all researched about manure management and worked with underrepresented, minority farmers to offer support and advice through all stages of farm development.
Currently, Luo is researching specific factors that influence microbial CUE and, in particular, the effect of atmospheric nitrogen deposition on microbes and how well they utilize and absorb carbon.
Luo hopes to improve confidence in carbon measurements and push for more effective carbon sequestration policies.
“These results provided more realistic and accurate microbial parameters,” Luo wrote in his 2023 study. “Which could reduce the uncertainty in long-term soil carbon response modeling to climate change.”
Madison Kim can be reached at [email protected].