AI model deciphers biochar’s role in reducing soil greenhouse gases

Applying biochar to soil is a recognized strategy for combating climate change, primarily by locking away carbon for long periods. Yet, its broader impact is complex; under different conditions, biochar can either suppress or unexpectedly release other potent greenhouse gases like nitrous oxide and methane from the soil. This inconsistency has been a significant barrier to its widespread adoption. A new set of predictive models developed by researchers Beatriz A. Belmonte, Raymond R. Tan, and their colleagues at the University of Santo Tomas and De La Salle University brings clarity to this issue. The team created a system to predict how soils will respond to biochar, offering a way to tailor its application for maximum climate benefit.

From Data to Decision Rules

Instead of conducting new field experiments, the scientists leveraged existing knowledge by compiling a vast dataset from 91 published research papers. This dataset included key variables like biochar pH and carbon-to-nitrogen ratio, soil properties, and production conditions like pyrolysis temperature. Using an approach called Rough Set-Based Machine Learning (RSML), the team trained their models to find hidden patterns within this data. The distinct advantage of this method is its transparency; it generates a series of simple, human-readable "if-then" rules, sidestepping the "black box" problem common to many other AI techniques and allowing for validation against known scientific principles.

Decoding Greenhouse Gas Responses

The resulting rule-based models demonstrated strong predictive power for nitrous oxide (N₂O) and methane (CH₄) fluxes. For N₂O, a powerful greenhouse gas, the models determined that applying biochar to soils with a high carbon-to-nitrogen ratio can effectively suppress its release. For CH₄, the models revealed that applying alkaline biochar with a high C/N ratio over shorter durations reduces emissions, likely by fostering a soil environment that favors methane-oxidizing microbes. The models also correctly predicted that very high biochar application rates could trigger methane release, confirming the mechanistic plausibility of the AI-generated rules.

Identifying Key Uncertainties

While the models for N₂O and CH₄ were robust, the models for carbon dioxide (CO₂) did not perform as well when tested against a separate validation dataset. This outcome suggests that the dynamics of CO₂ flux following biochar application are influenced by more intricate interactions that were not fully captured by the available data. This finding is valuable, as it precisely identifies a knowledge gap and highlights the need for further investigation into the factors controlling soil CO₂ emissions. The model's performance on CO₂ prediction signals a clear direction for future experimental work.

A Practical Tool for Climate Action

"Biochar holds immense promise for carbon dioxide removal, but its real-world effectiveness hinges on getting the application just right," states corresponding author Beatriz A. Belmonte. "Our work provides a clear, rule-based framework to guide these decisions. Instead of a black box, we developed an interpretable model that tells farmers and policymakers why certain conditions lead to better outcomes. This moves us from trial-and-error to a targeted strategy for maximizing climate benefits and avoiding unintended GHG emissions."

The Path Forward

The developed models serve as a foundational decision-support tool that can be continually refined. The authors propose that expanding the dataset and including additional variables, such as the experimental setting (laboratory versus field), will enhance the models' predictive accuracy. Future iterations could also be designed to predict other critical outcomes, including biochar's stability in soil and its effect on crop productivity. This work provides an essential guide for calibrating biochar application to local conditions, paving the way for its more effective and reliable use as a global negative emission technology.

Corresponding Author: Beatriz A. Belmonte

Original Source: https://doi.org/10.1007/s44246-024-00153-w

Contributions: Beatriz A. Belmonte performed the study conceptualization, development of methodology, formal analysis, writing of original draft, revision and editing of the manuscript. Jesus Gabriel A. Flores, Cristine L. Mestizo, and Patricia Nicole B. Rafer contributed in the development of methodology, data processing (software), and writing of the original draft. Michael Frances D. Benjamin contributed in the validation, and review and editing of the manuscript. Kathleen B. Aviso contributed in the development of methodology, validation, and review and editing of the manuscript. Raymond R. Tan contributed in the study conceptualization, development of methodology, validation, and review and editing of the manuscript.