| Summary: | We present an application of the Ensemble Kalman Filter (EnKF) to assimilate eddy covariance data from several sites in the arctic and test a simple model of tundra carbon dioxide exchange. We modified the EnKF to allow adaptive noise estimation by providing a feedback from the model-data residuals (innovations) to the Monte-Carlo noise added to the ensemble of model simulations. Leaf area index (LAI) was allowed to vary as an estimated parameter within the EnKF; variation in LAI accounts for seasonal phenology, wind driven changes in the tower footprint, and, most importantly, for latent or unaccounted variables missing from the model. We then use the EnKF to assimilate the data into the PLIRTLE model of arctic ecosystem-atmosphere CO2 exchange (Shaver et al. 2007, J. Ecol. 95:802-817). Filtering the data with the modified EnKF improved estimates of CO2 exchange both by filtering out noise in the eddy covariance data and by compensating for biases associated with deficiencies in the model. Accounting for latent variables in the EnKF, but without the adaptive noise estimation, improved the filter performance slightly over the unmodified EnKF. Adding the adaptive noise estimation without accounting for latent variables resulted in very high levels of model noise, which allowed the filter to track the data virtually without flaw, but did not filter out obvious noise in the data stream. The best performance was achieved when both latent-variable accounting and adaptive noise estimation were added to the EnKF. Finally we used the modified EnKF to test the PLIRTLE model. We found the trends in the estimates of LAI were associated with seasonal phenology. However, the EnKF also produced a diel pattern in the LAI estimates for some sites that was clearly not random. This pattern is indicative of some unaccounted variable missing from the PLIRTLE model. We hypothesize that the mechanism missing in the PLIRTLE model, at least for one Alaskan site, may be stomatal closure driven by low atmospheric humidity.
|
|---|