Representing Sub-Plot Canopy Heterogeneity Improves Model Prediction of Net Ecosystem Exchange in a Mixed-Deciduous Forest
Abstract:Canopy density and composition may vary within an eddy covariance tower’s footprint in response to small-scale topographic features, biotic interactions such as herbivory, local disturbances, etc. We are investigating how different representations of canopy heterogeneity influence predictions of net ecosystem CO2 exchange in a mixed-deciduous forest by an age/plant functional type structured ecosystem model.
Our study area is located at the University of Michigan Biological Station (UMBS) where two eddy covariance towers and periodic tree censuses provide a rich long-term record of ecosystem structure, weather, and carbon uptake. Meteorological measurements collected at the US-UMB AmeriFlux tower served to force, optimize, and evaluate the Ecosystem Demography model version 2 (ED2), while tree census information was used to initialize ED2.
To test the influence that representing canopy heterogeneity has on model-tower agreement, we ran a set of ED2 site-level simulations with an increasing number of sub-grid patches. The first simulation, which we call ‘aggregated’, had one large patch explicitly containing all trees. The aggregated canopy represents a case where different size cohorts of each plant functional type are distributed homogeneously throughout the plot with uniform stem density. Six other simulations represented patch-level canopies with varying degrees of heterogeneity, ranging from 5 to 64 sub-plot patches; each patch represented from one to several of the 0.1 ha tree census plots.
A preliminary comparison of the aggregated and the 20-plot heterogeneous simulations showed that including patch-level heterogeneity in the canopy description improved model prediction quality. For example, compared to the single-patch, aggregated simulation, including 20 sub-plot patches improved model bias in the estimated accumulated 5-year net ecosystem exchange from 17% to 5%, which is smaller than our tower observation uncertainty.
As a result of this study, we will identify methodologies that lead to better simulation designs by allowing the user to employ sub-plot census information to evaluate how much detail regarding canopy heterogeneity is needed to most accurately describe the total carbon flux observed in a forest.