We analyzed errors and uncertainties in time-integrated eddy correlation data for sites in the Amazon. A well-known source of potential error in eddy correlation is through possible advective losses of CO2emissions during calm nights. There are also questions related to the treatment of low frequencies, non-horizontal flow, and uncertainties in, e.g., corrections for tube delay and frequency loss, as well as the effect of missing data. In this study, we systematically explore these issues for the specific situation of flux measurements at two Amazon forest sites. Results indicate that, for this specific environment with tall forest and tall towers, errors and uncertainties caused by data spikes, delay corrections, and high-frequency loss are small (<3% on an annual basis). However, sensitivities to the treatment of low frequencies and non-horizontal flow can be large, especially if the landscape is not homogeneous. Given that there is no consensus on methodology here, this represents an uncertainty of 10–25% on annual total carbon uptake. The other large uncertainty is clearly in the nighttime fluxes. Two different ways to evaluate the validity of these fluxes resulted in at least a 100% difference of annual totals. Finally, we show that uncertainty (standard errors) associated with data gaps can be reduced to <0.5 Mg·ha−1·yr−1 if data are covering at least half of the time, with random spread. Overall uncertainty, on annual CO2 fluxes, excluding the nighttime dilemma, is estimated at ±12% (central Amazon site) to ±32% (southwest Amazon site). Additionally, the nighttime uncertainty is of similar magnitude as the time-integrated fluxes themselves.
