Reconciling inventory, tower, and remotely-sensed carbon estimates across northern Wisconsin through model-data fusion

Background/Question/Methods

The fundamental questions about how terrestrial ecosystems will respond to climate change are straightforward and well known, yet a small number of important gaps separate the information we have gathered from the understanding required to inform policy and management. A critical gap is that no one data source provides a complete picture of the terrestrial biosphere, and therefore multiple data sources must be integrated in a sensible manner. Process-based models represent an ideal framework for this synthesis, but to date model-data synthesize has only made use of a subset of the available data types, and remains inaccessible to much of the scientific community, largely due to the daunting ecoinformatics challenges. The Predictive Ecosystem Analyzer (PEcAn) is an open-source scientific workflow system and ecoinformatics toolbox that manages the flow of information in and out of regional-scale terrestrial biosphere models, facilitates formal data assimilation, and enables more effective feedbacks between models and field research.

Herein we demonstrate an application of the PEcAn system coupled with the Ecosystem Demography (ED2) model to synthesize the forest inventory data, eddy covariance tower fluxes, and LIDAR, RADAR, and MODIS remote sensing in northern Wisconsin over 15 years.

Results/Conclusions

Compared to tower flux measurements, ED2 had the lowest error at the Willow Creek (mature hardwood) and Sylvania (old growth) towers, was intermediate at WLEF (heterogeneous), and highest at Lost Creek (wetland). Compared to stand inventory data, ED2 predicts overall growth and mortality rates well but slightly underestimates understory growth, thought this as little impact on the overall carbon budget. LIDAR and RADAR data had the most impact on constraining biomass estimates for early-successional and low-stature vegetation, while inventory data had the largest impact for constraining mature and overmature stands. Prior to correcting data sets for spatial and temporal autocorrelation, flux data had the greatest impact on the carbon budget uncertainties. We developed and applied an novel spectral-based correction that accounts for autocorrelation at the 30 min, daily, and annual time scales. For all data types there is a diminishing return for collecting more of the same data, and a small amount of data on a new process (e.g. soil respiration, sap flux) can have a large impact on carbon estimates. Overall, at the site level soil carbon data had a modest information contribution, due to fine-scale heterogeneity, while large-scale uncertainties in soil carbon remain a major source of uncertainty.