Drivers of Long-Term Variability in Amazon Forest Carbon Fluxes

Abstract

Amazon forests are potentially an important sink of atmospheric carbon dioxide (CO2). However, the present magnitude of net CO2 uptake by intact Amazon forests and the controls upon seasonal, yearly, and decadal variability in their net CO2 sequestration and emission are not well understood. Across space and time, changes in meteorology, phenology, and disturbance can all affect the net ecosystem-atmosphere exchange of carbon (NEE). Forest NEE can be estimated using direct atmospheric eddy covariance measurements, but systematic measurement biases can make accurate estimates of the long-term direction and magnitude of the net carbon balance challenging. We present a 3.5-year update to a previously published 4-year record of hourly eddy covariance-derived NEE from an evergreen Eastern Amazon forest. To address biases resulting from subjective quality control (QC) of NEE data, we introduced QC criteria that were naïve to final NEE values. We also corrected poor calibrations that resulted in physically unrealistic CO2 concentrations, recovering 9% of all eddy covariance data before QC data removal (4.5% after removal). The new 3.5-year raw NEE data record had 74% coverage over all hours, compared with 78% coverage in the previous 4-year data record. The criteria produced annual NEE that were consistent with previously reported totals for the first 4 years. To partition influences upon interannual variability in NEE, we use a statistical model to represent NEE at our Eastern Amazon forest site as a constant response to changing meteorology and phenology throughout a decade. Our model simulated mean annual NEE well, with exception to the first year of our data record in 2002, during which the total residual emission was 1.2 MgC ha-1 because photosynthesis was anomalously low. We confirmed that a severe drought occurred in 1998, four years prior to our data record. We hypothesize that this drought caused a persistent disturbance through 2002. We estimated tree mortality of 12%-31% of aboveground biomass following the 1998 drought-disturbance, but there was no evidence that 21st century droughts disturbed this region of evergreen Amazon forest. Our results suggest that legacy effects of drought may impact photosynthesis via partial damage to still-living trees. Systematic biases in the total magnitude of NEE are typically associated with low-turbulence conditions at night. We developed a two-step bias correction, which makes different assumptions than the conventional friction velocity (u*) filter correction, assumptions that are supported by prior measurements of subcanopy advective divergence loss of nighttime flux. We applied both the convention correction and our novel correction to our Eastern Amazon site, an additional Amazon forest site, and a temperate forest site. Plausible u* filters led to implausible decadal carbon budgets at both Amazon forest sites. Conversely, our novel correction eliminated the dependence of nighttime NEE on u*, produced estimates for the mean advective lost nighttime flux that agreed with measurements made during growing seasons at two sites, and produced plausible decadal carbon budgets at both Amazon forest sites. Our NEE bias correction is a necessary alternative to the u* filter, but requires year-round growing seasons and multi-year data records.