Evaluate the seasonal cycle and interannual variability of carbon fluxes and the associated uncertainties using modeled and observed data

We assessed the performance of the Carnegie-Ames-Stanford Approach - Global Fire Emissions Database (CASA-GFED3) terrestrial carbon cycle model in simulating seasonal cycle and interannual variability (IAV) of global and regional carbon fluxes and uncertainties associated with model parameterization. Key model parameters were identified from sensitivity analyses and their uncertainties were propagated through model processes using the Monte Carlo approach to estimate the uncertainties in carbon fluxes and pool sizes. Three independent flux data sets, the global gross primary productivity (GPP) upscaled from eddy covariance flux measurements by Jung et al. (2011), the net ecosystem exchange (NEE) estimated by CarbonTracker, and the eddy covariance flux observations, were used to evaluate modeled fluxes and the uncertainties. Modeled fluxes agree well with both Jung's GPP and CarbonTracker NEE in the amplitude and phase of seasonal cycle, except in the case of GPP in tropical regions where Jung et al. (2011) showed larger fluxes and seasonal amplitude. Modeled GPP IAV is positively correlated (p < 0.1) with Jung's GPP IAV except in the tropics and temperate South America. The correlations between modeled NEE IAV and CarbonTracker NEE IAV are weak at regional to continental scales but stronger when fluxes are aggregated to >40°N latitude. At regional to continental scales flux uncertainties were larger than the IAV in the fluxes for both Jung's GPP and CarbonTracker NEE. Comparisons with eddy covariance flux observations are focused on sites within regions and years of recorded large-scale climate anomalies. We also evaluated modeled biomass using other independent continental biomass estimates and found good agreement. From the comparisons we identify the strengths and weaknesses of the model to capture the seasonal cycle and IAV of carbon fluxes and highlight ways to improve model performance.