Identifying the dominant climate-driven uncertainties in modeling gross primary productivity

Abstract

Accurate simulation of gross primary productivity (GPP) is essential for estimating the global carbon budget. However, GPP modeling is subject to various sources of uncertainties, among which the impacts of biases in climate forcing data have not been well quantified. Here, using a well-validated vegetation model, we compare site-level simulations using either ground-based meteorology or assimilated reanalyses to identify climate-driven uncertainties in the predicted GPP at 91 FLUXNET sites. Simulations yield the lowest root mean square errors (RMSE) in GPP relative to observations when all site-level meteorology and CO₂ concentrations are used. Sensitivity tests conducted with Modern-Era Retrospective Analysis (MERRA) reanalyses increase GPP RMSE by 30%. Replacement of site-level CO₂ with global annual average values provides limited contributions to these changes. In contrast, GPP uncertainties increase almost linearly with the biases in meteorology. Among all factors, photosynthetically active radiation (PAR), especially diffuse PAR, plays dominant roles in modulating GPP uncertainties. Simulations using all MERRA forcings but with site-level diffuse PAR help reduce over 50% of the climate-driven biases in GPP. Our study reveals that biases in meteorological forcings, especially the variabilities at diurnal to seasonal time scales, can induce significant uncertainties in the simulated GPP at FLUXET sites. We suggest cautions in simulating global GPP using climate reanalyses for dynamic global vegetation models and urgent improvements in climatic variability in reanalyses data, especially for diffuse radiation.

Keywords: Climate reanalyses; Diffuse radiation; Dynamic global vegetation model; FLUXNET; Gross primary productivity; YIBs model.