부산대학교 도서관

Land–atmosphere carbon and water exchanges have large uncertainty in terrestrial biosphere models (TBMs). Using observations to reduce TBM structural and parametric errors and uncertainty is a critical priority for both understanding and accurately predicting carbon and water fluxes. Recent implementations of the Bayesian CARbon DAta–MOdel fraMework (CARDAMOM) have yielded key insights into ecosystem carbon and water cycling. CARDAMOM estimates parameters for an associated TBM of intermediate complexity (Data Assimilation Linked Ecosystem Carbon – DALEC). These CARDAMOM analyses – informed by co-located C​​​​​​​ and H2O flux observations – have exhibited considerable skill in both representing the variability of assimilated observations and predicting withheld observations. CARDAMOM and DALEC have been continuously developed to accommodate new scientific challenges and an expanding variety of observational constraints. However, so far there has been no concerted effort to globally and systematically validate CARDAMOM performance across individual model–data fusion configurations. Here we use the FLUXNET 2015 dataset – an ensemble of 200+ eddy covariance flux tower sites – to formulate a concerted benchmarking framework for CARDAMOM carbon (photosynthesis and net C exchange) and water (evapotranspiration) flux estimates (CARDAMOM-FluxVal version 1.0). We present a concise set of skill metrics to evaluate CARDAMOM performance against both assimilated and withheld FLUXNET 2015 photosynthesis, net CO2 exchange, and evapotranspiration estimates. We further demonstrate the potential for tailored CARDAMOM evaluations by categorizing performance in terms of (i) individual land-cover types, (ii) monthly, annual, and mean fluxes, and (iii) length of assimilation data. The CARDAMOM benchmarking system – along with the CARDAMOM driver files provided – can be readily repeated to support both the intercomparison between existing CARDAMOM model configurations and the formulation, development, and testing of new CARDAMOM model structures.