Modeling tree stem methane

Tree stems are increasingly recognized as important components of methane (CH4) production and consumption, yet their contribution to the terrestrial CH4 budget remains highly uncertain due to sparse observations and challenges in upscaling in-situ measurements. Here, we explore the modeling of tree stem CH4 (TSCH4) fluxes across wetland and upland ecosystems using multi-source gridded environmental datasets. We first compiled 13,547 TSCH4 flux observations from 90 published studies, supplemented by in-situ measurements from two sites in 2025. A feature importance analysis is applied to identify key structural and environmental drivers of TSCH4 fluxes in wetland and upland ecosystems. Random forest regression models were developed using five key predictors identified through the feature selection. The models were trained using a combined 10-fold cross-validation to evaluate TSCH4 fluxes for uplands and wetlands separately. Results suggest that tree structural traits (e.g., sampling height, diameter at breast height) and subsurface soil variables (e.g., soil temperature and moisture at deep soil layer (< 100 cm), as well as belowground biomass) are the key predictors of TSCH4 fluxes. The models show reasonable predictive skill, with a mean coefficient of determination (R2) of 0.65 and a root mean square error (RMSE) of 3.26 nmol CH4 m–2 s–1 for wetlands, and a mean R2 of 0.64 and an RMSE of 0.70 nmol CH4 m–2 s–1 for uplands. The models generally capture the seasonal variability of TSCH4 fluxes in both wetland and upland ecosystems, with higher and delayed peaks in wetlands than in uplands. The models demonstrate the potential for upscaling TSCH4 by linking in-situ measurements with gridded environmental predictors, improving our understanding of the role of tree stems in the terrestrial methane budget.