A robust model based on root morphological and anatomical features to distinguish high and low methane emission rice varieties through machine learning approaches

Abstract Rice fields are a major producer of methane, a strong greenhouse gas. However, identifying genetic variation in methane emissions among rice varieties remains challenging. This study applied association rule mining to detect key rice root morphological and anatomical traits influencing methane emissions, validated using a support vector machine. We report models which accurately classified high and low methane-emitting varieties with 98% (morphological) and 94% (anatomical) accuracy. These models effectively distinguished methane emission categories based on intrinsic trait patterns. Machine learning analysis highlighted the top 10 morphological and anatomical traits associated with methane emission levels. High methane-emitting varieties were characterized by lower middle root porosity, base root porosity, average root porosity, root diameter (RDia), and higher S-type lateral root length. Conversely, low methane-emitting varieties exhibited lower root number, tiller number, root dry weight, leaf number, and higher RDia. Anatomically, high methane-emitting varieties showed reduced lacunae number, total stele area, mean metaxylem size, metaxylem number, and metaxylem vessel area. Low methane-emitting varieties, in contrast, had higher percent aerenchyma, total stele area, ratio of total cortical area to root cross-section area, ratio of stele to root cross-section area, and aerenchyma area. The results suggest that the rhizosphere oxygenation role of root porosity and aerenchyma might predominate over the methane transport role.