Predictions of Diurnal Cycles of Net Ecosystem Exchange (NEE) across Global FLUXNET Sites using Scaling-based Empirical Modeling

The diurnal cycles of hourly net ecosystem exchange (NEE) were predicted across terrestrial ecosystems by employing a scaling-based empirical modeling procedure. The diurnal NEE cycles - representing different days and sites - were scaled using the corresponding day- and site-specific reference observations as the scaling parameters, transforming different cycles into a common dimensionless diurnal cycle. The scaled cycle was then parameterized by applying an extended stochastic harmonic algorithm (ESHA). Hourly NEE cycles during 1991-2014 from 190 FLUXNET sites incorporating a diversity of ecosystem types and climates were utilized for model parameter estimations, calibrations (training), and validations (independent testing). The estimated model parameters over the calibration days and sites exhibited spatiotemporal robustness under two broad climate classes: polar and non-polar. The predicted NEE using the site-specific, temporally ensemble model parameters from a reference NEE observation measured between 11 A.M. to 2 P.M. demonstrated good model fitting efficiency (Nash-Sutcliffe Efficiency, NSE = 0.58-0.94). Further, predictions based on the spatiotemporally aggregated ecosystem-specific (averaged for different ecosystems under polar and non-polar climate types) and generalized (averaged across ecosystems under the polar and non-polar climate types) parameter sets also showed acceptable accuracies and efficiencies. The predicted NEE for the validation sites were then compared with the FLUXNET gap-filled data at both hourly and annual scales. The developed models can be utilized for robust estimations and predictions of the diurnal cycles of hourly NEE at diverse terrestrial ecosystems across the globe.