Climate-driven uncertainties in modeling terrestrial gross primary production: a site level to global-scale analysis

We used a land surface model to quantify the causes and extents of biases in terrestrial gross primary production (GPP) due to the use of meteorological reanalysis datasets. We first calibrated the model using meteorology and eddy covariance data from 25 flux tower sites ranging from the tropics to the northern high latitudes and subsequently repeated the site simulations using two reanalysis datasets: NCEP/NCAR and CRUNCEP. The results show that at most sites, the reanalysis-driven GPP bias was significantly positive with respect to the observed meteorology-driven simulations. Notably, the absolute GPP bias was highest at the tropical evergreen tree sites, averaging up to ca. 0.45kgCm(-2)yr(-1) across sites (ca. 15% of site level GPP). At the northern mid-/high-latitude broadleaf deciduous and the needleleaf evergreen tree sites, the corresponding annual GPP biases were up to 20%. For the nontree sites, average annual biases of up to ca. 20-30% were simulated within savanna, grassland, and shrubland vegetation types. At the tree sites, the biases in short-wave radiation and humidity strongly influenced the GPP biases, while the nontree sites were more affected by biases in factors controlling water stress (precipitation, humidity, and air temperature). In this study, we also discuss the influence of seasonal patterns of meteorological biases on GPP. Finally, using model simulations for the global land surface, we discuss the potential impacts of site-level reanalysis-driven biases on the global estimates of GPP. In a broader context, our results can have important consequences on other terrestrial ecosystem fluxes (e.g., net primary production, net ecosystem production, energy/water fluxes) and reservoirs (e.g., soil carbon stocks). In a complementary study (Barman etal., ), we extend the present analysis for latent and sensible heat fluxes, thus consistently integrating the analysis of climate-driven uncertainties in carbon, energy, and water fluxes using a single modeling framework.