An Amazon basin-wide intercomparison of ecosystem land surface models and flux observations: Results from the LBA-MIP

Amazon forests play an important and complex role in the global cycles of water and carbon, and important advances have been made in understanding Amazon processes in recent years. However, reconciling modeled mechanisms with observations across scales remains a challenge. To better address this challenge, we initiated a Model Intercomparison Project for the ‘Large-Scale Biosphere Atmosphere Experiment in Amazonia’ (LBA-MIP) to integrate modeling and observational studies for improved understanding of Amazon basin water and carbon cycling. Here, we report on expanded analysis from this project, which used the network of meteorological data from Amazon tower sites (the BrasilFlux network) in forest and converted lands to drive a suite of land surface ecosystem models that simulate energy, water and CO2 fluxes. We focus here on controls on gross primary photosynthesis (GPP) and evapotranspiration (ET), and in particular on how well models capture the observed diurnal and seasonal cycles across sites. We intercompare stomatal (via available soil moisture), phenological, and available energy controls on GPP and ET across models and observations. We found that available energy is the dominant control on ET across the Amazon basin, and in assessing mechanisms responsible for differences in model performance with respect to ET, we found that soil moisture storage capacity was an important factor in ability of models to match observed seasonal cycles. We evaluated model performance with respect to GPP in terms of light use efficiency as well as leaf phenology, where observations were available. Finally, we discuss important ancillary measurements at flux tower sites (e.g., deep soil moisture profiles, LAI, and litterfall) necessary to improve our ability to empirically discriminate among different model mechanisms used to match observed seasonal patterns.