Environmental Linkages of Hourly Net Ecosystem Exchange (NEE) Across Diverse Ecosystems

We investigated the potential similarities in linkages between the environmental variables and the terrestrial net ecosystem exchange (NEE) of CO₂ across 58 North American sites. Observations of hourly NEE and the concurrent environmental variables during 2000-2015 were obtained for all sites from the AmeriFlux network, representing six diverse ecosystems (wetland, cropland, grassland, evergreen, deciduous and mixed forest). Pearson correlations, principal component analysis, and factor analysis were employed to investigate the interrelations and underlying grouping patterns among NEE and the environmental drivers. The relative linkages of NEE with the environmental variables were then reliably estimated by resolving multicollinearity through partial least squares regression (PLSR) modeling. Further, cross-validations were conducted with the ecosystem-specific ensemble PLSR model parameters to assess the similarities of the linkages across the different ecosystems. The analysis led to the identification of four biophysical process components of NEE, in general. The 'radiation-energy' component had the strongest linkages with NEE. The 'temperature-hydrology' component had moderate linkages, whereas the 'aerodynamic' (wind and friction velocities) and 'ambient atmospheric CO₂ concentrations' components were, respectively, weakly and moderately linked to NEE. Comparable cross-validations with the ecosystem-specific ensemble PLSR parameters provided further evidence of the underlying similarity in the environmental linkages of NEE across ecosystems. The analysis provided important insights into the potential cross-ecosystem generalizability of environmental parameters in terrestrial carbon flux models. The identified linkages and major process drivers would further guide the development of low-dimensional models for a robust prediction of ecosystem CO₂ fluxes.