Cross-site evaluation of eddy covariance GPP and RE decomposition techniques

Eddy covariance flux towers measure net exchange of land-atmosphere flux. For the flux of carbon dioxide, this net ecosystem exchange (NEE) is governed by two processes, gross primary production (GPP) and a sum of autotrophic and heterotrophic respiration components known as ecosystem respiration (RE). A number of statistical flux-partitioning methods, often developed to fill missing NEE data, can also be used to estimate GPP and RE from NEE time series. Here we present results of the first comprehensive, multi-site comparison of these partitioning methods. An initial test was performed with a subset of methods in retrieving GPP and RE from NEE generated by an ecosystem model, which was also degraded with realistic noise. All methods produced GPP and RE estimates that were highly correlated with the synthetic data at the daily and annual timescales, but most were biased low, including a parameter inversion of the original model. We then applied 23 different methods to 10 site years of temperate forest flux data, including 10 different artificial gap scenarios (10% removal of observations), in order to investigate the effects of partitioning method choice, data gaps, and intersite variability on estimated GPP and RE. Most methods differed by less than 10% in estimates of both GPP and RE. Gaps added an additional 6-7% variability, but did not result in additional bias. ANOVA showed that most methods were consistent in identifying differences in GPP and RE across sites, leading to increased confidence in previously published multi-site comparisons and syntheses. Several methods produced outliers at some sites, and some methods were systematically biased against the ensemble mean. Larger model spread was found for Mediterranean sites compared to temperate or boreal sites. For both real and synthetic data, high variability was found in modeling of the diurnal RE cycle, suggesting that additional study of diurnal RE mechanisms could help to improve partitioning algorithms. (C) 2007 Elsevier B.V. All rights reserved.