Diagnosing Model Error in Canopy-Atmosphere Exchange Using Empirical Orthogonal Functions

Darren Drewry & John Albertson

Duke University

Multi-layer canopy process models (MLCPMs) have been established as tools for estimating local-scale canopy-atmosphere scalar (carbon dioxide, heat and water vapor) exchange as well as testing hypotheses regarding the mechanistic functioning of complex vegetated land surfaces and the interactions between vegetation and the local microenvironment. These model frameworks are composed of a coupled set of component submodels relating radiation attenuation and absorption, photosynthesis, turbulent mixing, stomatal conductance, surface energy balance and soil and subsurface processes. Submodel formulations have been validated for a variety of ecosystems under varying environmental conditions. However, each submodel component requires parameter values that are known to vary seasonally as canopy structure changes, and over shorter periods characterized by shifts in the environmental regime. The temporal dependence of submodel parameters limits application of MLCPMs to short-term integrations for which a specific parameterization can be trusted. We present a novel application of empirical orthogonal function (EOF) analysis to the identification of the primary source of MLCPM error. Carbon dioxide (CO2) concentration profiles, a commonly collected and underutilized data source, are the observed quantity in this analysis. The technique relies on an ensemble of model runs transformed to EOF space to determine the characteristic patterns of model error associated with specific submodel parameters. These patterns provide a basis onto which error residual (modeled - measured) CO2 concentration profiles can be projected to identify the primary source of model error. Synthetic tests and application to field data collected at Duke Forest (North Carolina, USA) are presented.

Last Update: March 25, 2005