Monday, August 3, 2009 - 3:20 PM

OOS 1-6: Assimilating arctic C flux data into an ecosystem model using the Ensemble Kalman Filter

Edward B. Rastetter1, M. Williams2, Kevin L. Griffin3, B. Kwiatkowski1, G. Tomasky1, M. Potosnak4, Paul C. Stoy2, Gaius Shaver5, M. Stieglitz6, G. Kling7, and John E. Hobbie5. (1) Marine Biological Lab, (2) University of Edinburgh, (3) Lamont Doherty Earth Observatroy, (4) DePaul University, (5) Marine Biological Laboratory, (6) Georgia Institute of Technology, (7) University of Michigan

Background/Question/Methods Are the predictions of a simple ecosystem C flux model that was well corroborated with chamber data consistent with eddy-flux time series data?  Is there information in the eddy-flux time series that can be used to improve the model?  We embedded a simple model of arctic carbon exchange into the Ensemble Kalman Filter (EnKF) and used it to assimilate data from an eddy-covariance tower located in moist tussock tundra on the North Slope of Alaska.  The model predicts net ecosystem carbon exchange based only on leaf area, irradiance, and temperature and has been well corroborated for all the major vegetation types in the low Arctic using chamber-based data.  This is the first application of the model to eddy-covariance data.  We modified the EnKF by adding an adaptive noise estimator that provides a feedback between persistent model-data deviations and the noise added to the ensemble of Monte-Carlo simulations in the EnKF.  Because the leaf area index (LAI) within the tower foot print is difficult to quantify, we augmented the Kalman state vector with a parameter representing LAI and allowed LAI to adapt through time, providing a model-EnKF-based estimate of LAI.  The adaptation of LAI also compensated for latent variables missing from the model structure and thereby allowed us to test the model based on deviations of the LAI trajectory from expectation.

Results/Conclusions On a weekly time scale, the EnKF estimates of LAI followed a general trend expected of canopy phenology.  However, there were also nonrandom, diel deviations in the LAI estimates.  Although the model tracked the eddy-covariance data closely, these deviations indicate an inadequate representation of some latent variable missing from the model.  The latent variable might be associated with daily fluctuations in the metabolism of the ecosystem. For example the deviations are consistent with vapor pressure driven stomatal closure.   The deviations are also correlated with changes in wind direction, which would change the tower footprint.  However, the deviations could equally be associated with responses of the open-path eddy-covariance instrumentation to changes in temperature and humidity.  A distinction among these possibilities could not be made based on the data available.