Fig. 2: Optimal North American and biome-specific estimates of the temperature sensitivity of ecosystem respiration for individual models and the model ensemble.

Adjustments to model-specific estimates of the activation energy are determined by maximizing consistency with observed atmospheric CO₂ variability (Methods). a–d, Relationships between adjustments to model-specific estimates of the activation energy needed to maximize consistency with observed atmospheric CO₂ variability (ΔE_a; vertical axis) and the original estimates of activation energy (E_a; horizontal axis) for models in the MsTMIP v2 (pink), TRENDY v6 (brown) and FLUXCOM (orange) model ensembles in the North American domain (a), croplands (b), evergreen needleleaf forests (c) and deciduous broadleaf and mixed forests (d). Note that only models for which the explanatory power of simulated GPP (\({R}^{2}_{{{{\rm{GPP}}}}}\)) exceeds that of shortwave radiation (\({R}^{2}_{{{{\rm{SW}}}}}\)) are included (Extended Data Fig. 6), leaving two models in the MsTMIP ensemble and one model in the TRENDY ensemble excluded (Methods). The grey dashed lines represent the best orthogonal distance regression (ODR) fit between ΔE_a and E_a estimates across the three ensembles of models (N = 29), with light grey shading indicating the 95% prediction interval. The optimal E_a (\({\hat{E}}_{{{{\rm{a,opt}}}}}\)) corresponds to the point where the ODR fit line crosses ΔE_a = 0 eV (that is, no adjustment to E_a is needed) and is indicated using a green dashed line and listed for each biome in the corresponding panel.