A13L-3341:
Parameter estimation and data assimilation with the Community Land Model (CLM) to upscale net CO2 fluxes from plot to catchment scale
Monday, 15 December 2014
Hanna Post1, Timothy J Hoar2, Jasper A Vrugt3, Xujun Han1, Roland Baatz1, Kumbhar Pramod4, Harry Vereecken1 and Harrie-Jan Hendricks Franssen1, (1)Agrosphere Institute (IBG-3), Forschungszentrum Jülich, Jülich, Germany, (2)NCAR, Boulder, CO, United States, (3)University of California Irvine, Irvine, CA, United States, (4)École polytechnique fédérale de Lausanne, CH-1015, Switzerland, Lausanne, Switzerland
Abstract:
The Community Land Model CLM version 4.5 (CLM4.5) was applied to simulate the net ecosystem exchange of CO2 (NEE) for the 2454 km2 Rur catchment located in the border region of Belgium, Germany, and the Netherlands. NEE was measured within this catchment by six eddy covariance (EC) towers located on different land use sites. To reliably determine NEE patterns within the catchment, and taking into account uncertainty in observations, meteorological forcings, model parameters and initial conditions, we applied two different model-data fusion methods: (1.) The parameter estimation problem was solved using an adaptive Markov Chain Monte Carlo (MCMC) method. The eight parameters estimated had been selected through sensitivity analysis. We tested the effect of different lengths of the model calibration period (4x3 months versus 1 year) and measurement averaging intervals (30 min. versus 6 hourly). Parameter estimation results were strongly influenced by initial model states and varied for the different seasons. For the one year runs parameter uncertainty was larger than for the three months runs. Posterior pdfs of parameters for three months periods differed significantly from the one year run. The parameter estimation results were evaluated for an additional verification period of one year. (2.) After model calibration, we assimilated leaf area index (LAI) and eddy covariance data into CLM using the Ensemble Kalman Filter (EnKF). Only model states (e.g. LAI, leaf nitrogen) were updated. We found that calculated LAI and other state variables for c3-grass and prognostic c3-crops in CLM are highly sensitive to meteorological input data. We assume this is due to the phenology of c3-grass and c3-crop in CLM which is strongly based on threshold values for e.g. temperature and precipitation to initiate onset and offset. Adding parameter uncertainty noticeably affected the data assimilation results.
