Fig. 4. Partial least squares regression (PLSR) statistical modeling to predict sediment CH₄ concentrations.

PLSR analyses tested the ability of different suites of explanatory variables to predict measured sediment CH₄ concentrations in the four cores from 2012 across depths (n = 21); in all models, all measured abiotic variables (except those related to CH₄ concentrations, see Methods) were included as explanatory variables, and biotic variables were added as indicated. Biotic variables included relative abundances of specific OTUs and/or summed OTU abundances grouped by taxonomy or predicted metabolism (as indicated), from 16S rRNA gene amplicon data. a Correlation coefficient (r²) for PLSR models predicting sediment CH₄ using different combinations of explanatory variables. Each bar represents a single underlying data point, with the value of that point indicated by the bar height along the y-axis. b Linear regression of measured and model-predicted sediment CH₄, considering all abiotic variables and methanogen and methanotroph abundances as explanatory variables; error band represents 95% confidence interval; each point is a sample, colored by core. c For the model with the highest r² (rightmost in panel a), all significant explanatory variables are shown (VIP scores > 1, n = 26 significant explanatory variables out of 153 total variables considered). VIP scores show the relative contribution of each variable to the model, with higher VIP scores indicating a more significant contribution. Each bar represents a single underlying data point, with the value of that point indicated by the bar height along the y-axis.