What's happening in farm country reflects the promise and pitfalls of fighting climate change. While the United Nations has notedagriculture's potential to collect and store carbon, it contributes 10 percent of the United States' greenhouse gases, and 25 percent across the globe. In Minnesota, agricultural emissions have held steady in recent decades, and are now the second-biggest source of carbon releases, according to a study released last year.
New research from the university suggests machine learning could help scientists, and eventually farmers, do a better job of tracking that carbon. Two authors of the study said the knowledge would help determine which farming changes will work best. It's crucial knowledge at a time when the federal Inflation Reduction Act earmarked $20 billion for farmers to implement "climate-smart" practices on their land.
"It's easy to verify, OK, this farmer has planted a cover crop, that farmer has [stopped tilling], but the outcome impact on carbon is totally different," said Zhenong Jin, a researcher in the Department of Bioproducts and Biosystems Engineering, and one of the authors of the study. The study was produced by the university's federally funded AI-CLIMATE center, which is dedicated to tracking carbon on farms and in forests.
It seems counterintuitive that agriculture can create planet-warming gases, because plants absorb carbon dioxide as they undergo photosynthesis. But under the right conditions, organic matter can break down in the soil and release that carbon dioxide back into the atmosphere. Fertilizer can also release nitrous oxide into the air, a lesser known climate pollutant that is 273 times more potent than CO2, according to the EPA.
Doing less to disturb the soil by planting perennial or temporary cover crops and reducing plowing and tilling should help keep carbon in the ground. But there is significant debate among researchers about what methods work best.
The new study, published in the journal Nature Communications, reports that using AI proved significantly more accurate than other models.
Vipin Kumar, a professor of computer science and engineering and also a study author, said researchers sought to cut down on "hallucination," or risks that AI would invent inaccurate results. They did that by feeding the AI data from an existing model used to estimate the movement of carbon.
The technology researchers used "is not that far from Chat GPT," Kumar said.
While it's one challenge to better measure agricultural emissions, another challenge will be recruiting farmers to reduce them.
A bevy of programs in recent years have promised payments, often per-ton of carbon stored, to entice producers to change farming practices. Last August, when visiting the Minnesota State Fair, U.S. Agriculture Secretary Tom Vilsack noted income streams already available and others on the horizon — from carbon credits to sustainable aviation fuel — aim to help families stay on farms at a time when 50 percent of producers draw off-farm income.
Richard Conant, a professor in the Department of Ecosystem Science and Sustainability at Colorado State University, said the methods proposed in the paper could be useful to outline national policy on where to best use money to keep carbon in the soil of farm fields. Conant was not part of the study.
"I love this approach," said Conant, who has also used AI to track the movement on nitrous oxide.
But he most often relies on soil samples in his own research. He said even an AI model will need more information from on the ground. In general, he said, "We're operating in kind of a data-poor environment."
©2024 StarTribune, Distributed by Tribune Content Agency, LLC.
