Climate-driven uncertainties in modeling terrestrial energy and water fluxes: a site-level to global-scale analysis



We used a land surface model constrained using data from flux tower sites, to analyze the biases in ecosystem energy and water fluxes arising due to the use of meteorological reanalysis datasets. Following site-level model calibration encompassing major vegetation types from the tropics to the northern high-latitudes, we repeated the site and global simulations using two reanalysis datasets: the NCEP/NCAR and the CRUNCEP. In comparison with the model simulations using observed meteorology from sites, the reanalysis-driven simulations produced several systematic biases in net radiation (Rn), latent heat (LE), and sensible heat (H) fluxes. These include: (i) persistently positive tropical/subtropical biases in Rn using the NCEP/NCAR, and gradually transitioning to negative Rn biases in the higher latitudes; (ii) large positive H biases in the tropics/subtropics using the NCEP/NCAR; (iii) negative LE biases using the NCEP/NCAR above 40°N; (iv) high tropical LE using the CRUNCEP in comparison with observationally derived global estimates; and (v) flux-partitioning biases from canopy and ground components. Across vegetation types, we investigated the role of the meteorological drivers (shortwave and longwave radiation, atmospheric humidity, temperature, precipitation) and their seasonal biases in controlling these reanalysis-driven uncertainties. At the global scale, our site-level analysis explains several model-data differences in the LE and H fluxes when compared with observationally derived global estimates of these fluxes. Using our results, we discuss the implications of site-level model calibration on subsequent regional/global applications to study energy and hydrological processes. The flux-partitioning biases presented in this study have potential implications on the couplings among terrestrial carbon, energy, and water fluxes, and for the calibration of land–atmosphere parameterizations that are dependent on LE/H partitioning.