Friday, August 7, 2009 - 10:50 AM

COS 119-9: Carbon cycling at in a temperate evergreen forest: a multi-model data-model fusion analysis at Howland, ME

David J.P. Moore, King's College London, Andrew D. Richardson, University of New Hampshire, Dan Ricciuto, Oak Ridge National Lab, and David, Y. Hollinger, US Department of Agriculture Forest Service.

Background/Question/Methods

The eddy covariance method and automated respiration chambers, are being used to obtain continuous measurements of different C fluxes, and ancillary ecological measurements (e.g., biomass inventories) provide valuable information on C pools. Starting in 1996 . eddy covariance measurements of net ecosystem exchange and evapo-transpiration, chamber measurements of soil respiration, as well as periodic measurements of leaf area index, litterfall, soil respiration, and standing biomass have been quantified at the spruce-dominated Howland Forest AmeriFlux site. We conducted a multi-model data-model fusion experiment, using the available data to constrain the parameterization of three ecosystem models; the Simplified Photosynthesis and Evapo-transpiration (SIPNET) model, the Data Assimilation Linked Ecosystem Carbon (DALEC) model and the Local Terrestrial Ecosystem Carbon model (LOTEC). Model parameters were optimized, conditional on four years (1997-2000) data and best estimates of data uncertainty, using a modified Metropolis algorithm, a Monte-Carlo Markov Chain technique and the Ensemble Kalman Filter.  

Results/Conclusions

We compare the strengths and weaknesses of three ecosystem models in (1) reproducing observed carbon fluxes from the forests at different time scales (e.g., diurnal cycles vs. interannual variation) and (2) extracting process level information from different data streams in isolation and in combination (e.g., model performance when only eddy covariance measurements are used to constrain the model vs. when all available data streams are used).  The optimized models are then run forward (2001-2004) and model predictions evaluated against additional flux and pool measurements made at the site.  Although conditioning the models using eddy covariance data improved estimates of short term variation this was in part driven by unrealistic behaviour in Carbon pools. Longer term processes were only constrained by incorporating LAI, litterfall and soil respiration, reducing uncertainty in model predictions and carbon cycling over inter-annual time scales remains challenging.