HR: 0800h
AN: H41A-0400    [Abstracts]
TI: Bayesian Inversion of Soil-Plant-Atmosphere Interactions for an Oak-Savanna Ecosystem Using Markov Chain Monte Carlo Method
AU: * Chen, X
EM: chenxy@ce.berkeley.edu
AF: Department of Civil and Environmental Engineering, University of California, Berkeley, CA 94720 United States
AU: Rubin, Y
EM: rubin@ce.berkeley.edu
AF: Department of Civil and Environmental Engineering, University of California, Berkeley, CA 94720 United States
AU: Baldocchi, D D
EM: baldocchi@nature.berkeley.edu
AF: Department of Environmental Science, Policy and Management, University of California, Berkeley, CA 94720 United States
AB: Understanding the interactions between soil, plant, and the atmosphere under water-stressed conditions is important for ecosystems where water availability is limited. In such ecosystems, the amount of water transferred from the soil to the atmosphere is controlled not only by weather conditions and vegetation type but also by soil water availability. Although researchers have proposed different approaches to model the impact of soil moisture on plant activities, the parameters involved are difficult to measure. However, using measurements of observed latent heat and carbon fluxes, as well as soil moisture data, Bayesian inversion methods can be employed to estimate the various model parameters. In our study, actual Evapotranspiration (ET) of an ecosystem is approximated by the Priestley-Taylor relationship, with the Priestley-Taylor coefficient modeled as a function of soil moisture content. Soil moisture limitation on root uptake is characterized in a similar manner as the Feddes­_ model. The inference of Bayesian inversion is processed within the framework of graphical theories. Due to the difficulty of obtaining exact inference, the Markov chain Monte Carlo (MCMC) method is implemented using a free software package, BUGS (Bayesian inference Using Gibbs Sampling). The proposed methodology is applied to a Mediterranean Oak-Savanna FLUXNET site in California, where continuous measurements of actual ET are obtained from eddy-covariance technique and soil moisture contents are monitored by several time domain reflectometry probes located within the footprint of the flux tower. After the implementation of Bayesian inversion, the posterior distributions of all the parameters exhibit enhancement in information compared to the prior distributions. The generated samples based on data in year 2003 are used to predict the actual ET in year 2004 and the prediction uncertainties are assessed in terms of confidence intervals. Our tests also reveal the usefulness of various types of soil moisture data in parameter estimation, which could be used to guide analyses of available data and planning of field data collection activities.
DE: 1813 Eco-hydrology
DE: 1818 Evapotranspiration
DE: 1852 Plant uptake
DE: 1866 Soil moisture
DE: 1869 Stochastic hydrology
SC: Hydrology [H]
MN: Fall Meeting 2005