Description of local carbon flux from large scale gridded climate data by a dynamic global vegetation model at...

Dynamic Global Vegetation Models (DGVMs) are commonly used to describe the land biogeochemical processes and regulate carbon and water pools. However, the simulation efficiency and validation of DGVMs are limited to varying temporal and spatial resolutions. Additionally, the uncertainties caused by different interpolation methods used in DGVMs are still not clear. In this study, we employ Socio-Economic and natural Vegetation ExpeRimental (SEVER) DGVM to simulate Net Ecosystem Exchange (NEE) flux with large scale National Centers for Environmental Prediction (NCEP) daily climate data as inputs for the years 1997-2000 at 14 Euroflux sites. It is shown that daily local NEE flux on chosen sites can be reasonably simulated, and daily temperature and shortwave radiation are the most essential inputs for daily NEE simulation compared with precipitation and the ratio of sunshine hours. Different running means (1 to 30 days) methods are analysed for each Euroflux site, and the best results of both averaged regression coefficient and averaged slope of regression are discovered by using 5 days running mean method. SEVER DGVM, driven by linearly interpolated daily climate data is compared at the monthly time step with Lund-Potsdam-Jena (LPJ) DGVM, which combines the linear interpolation of daily temperature with stochastic generation of daily precipitation. The comparison demonstrates that the stochastic generation of daily precipitation provides an acceptable fit to local observed NEE, but with a slight decrease in accuracy. Simulation experiments with SEVER DGVM demonstrate that daily local NEE flux inside a grid cell for a region as large as Europe can be modelled by DGVMs, using only large scale climate data as inputs.