Fig. 5: Final SEM structure for annual NEE, GPP and R_e fluxes.

a, The hypothesized pathways linking explanatory variables to ecosystem carbon fluxes show the standardized path coefficients of the final SEM with significance levels (****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05, exact P values in Supplementary Table 10). The tests are two-sided, with no adjustments for multiple comparisons. The arrow thickness indicates the magnitude of standardized effect sizes, the black arrows and text indicate direct pathways, the double-headed grey arrows and text indicate the residual correlations and the blue arrows and text indicate the mediation pathways supported by the bootstrap analysis. The SEM was fitted to complete cases for all variables (605 site years, 84 sites). b, The bootstrap-derived estimates shown are means and 95% percentiles from 1,000 replicates for the direct and indirect effects of GS length and ALAN on NEE. c, The outputs from a leave-one-out analysis show the change in marginal (lighter-coloured symbols and lines) and conditional (darker-coloured symbols and lines) R² (ΔR²) for NEE (gold), GPP (blue) and R_e (magenta) relative to the full model after systematically removing each exogenous predictor (Supplementary Table 11).