PECAn, a workflow management tool for real-time data assimilation and forecasting: Evaluation of a switchgrass (Panicum virgatum) cropping system

Background/Question/Methods   Ecosystem models use physiological traits and environmental data to model land-air carbon exchange. Such models can integrate heterogeneous ecological and environmental data, but rarely account for the contribution of each input to the overall uncertainty in model output. Quantifying uncertainty in each model input and propagating uncertainty through model ensembles directly estimates overall uncertainty in model output. Identification of data that would most effectively constrain uncertainty can also be directly inferred by processing output. Real-time data assimilation and forecasting benefits from a method that is transparent, scalable, and adaptable. A tool that manages these steps while tracking the provenance of data inputs, model parameters, output, and post-processing would facilitate evaluation, reproducibility, and adaptation of the workflow.

Results/Conclusions   We present such a workflow management tool, the Predictive Ecophysiological Carbon flux Analyzer (PECAn). PECAn is an open source utility that encapsulates 1) acquisition of meteorological inputs, 2) synthesis of physiological trait data as the posterior distribution of a Bayesian meta-analysis, 3) sampling trait meta-analysis posterior distributions to parameterize ensembles of an ecophysiological model, Ecosystem Demography, version 2 (ED2), 4) probabilistic forecasts,5) postprocessing to constrain forecasts and model parameters with field, meterological, eddy flux, and spectral data, and 6) provenance tracking. This system provides a detailed analysis of the past and present ecosystem functioning that seamlessly transitions into forecasts. We demonstrate an application of PECAn to an intensively monitored and data-rich plot of a perennial biofuel crop, switchgrass (Panicum virgatum). This particular context will be used to highlight PECAn's general utility to support model-data integration, ecological research, and decision support.