Seasonal patterns of CO2fluxes in Amazon forests: Fusion of eddy covariance data and the ORCHIDEE model

Abstract

[1] In some regions of the Amazon, global biogeophysical models have difficulties in reproducing measured seasonal patterns of net ecosystem exchange (NEE) of carbon dioxide. The global process-based biosphere model Organizing Carbon and Hydrology in Dynamic Ecosystems (ORCHIDEE) used in this study showed that a standard model parameterization produces seasonal NEE patterns that are opposite in phase to the eddy flux data of the tropical evergreen forest at the Tapajos km 67 site (Brazil), like many other global models. However, we optimized several key parameters of ORCHIDEE using eddy covariance data of the Tapajos km 67 site in order to identify the driving factors of the seasonal variations in CO2 flux in this tropical forest ecosystem. The validity of the retrieved parameter values was evaluated for two other flux tower sites in the Amazon. The different tested optimization scenarios showed that only a few parameters substantially improve the fit to NEE and latent heat data. Our results confirm that these forests have the ability to maintain high transpiration and photosynthesis during the dry season in association with a large soil depth (Dsoil = 10 m) and a rooting system density that decreases almost linearly with depth (Hroot = 0.1). Previous analyses of seasonal variations in eddy covariance fluxes indicated that higher GPP levels were reached in the dry season compared to the wet season. Our optimization analysis suggests that this pattern could be caused by a leaf flush at the start of the dry season increasing the photosynthetic capacity of the canopy. Nevertheless, the current model structure is not yet able to simulate such a leaf flush, and we therefore suggest improving the ORCHIDEE model by including a specific phenology module that is driven by light availability for the tropical evergreen plant functional types. In addition, our results highlight both the potential and the limitations of flux data to improve global terrestrial models. Several parameters were not identifiable, and the risk of overfitting of the model was illustrated. Nevertheless, we conclude that these models can be improved substantially by assimilating site level flux data over the tropics.