J5.5 Estimating realistic canopy-level atmospheric values from low resolution gridded free-troposphere data through Bayesian inference and Markov Chain Monte Carlo (MCMC) methods

Wednesday, 4 August 2010: 4:30 PM
Red Cloud Peak (Keystone Resort)
Bin Deng, Indiana University Bloomington, Bloomington, IN; and H. P. Schmid and D. Dragoni

To represent the surface heterogeneity in a global or regional model grid, one common approach is to divide the grid into a collection of “homogeneous” subgrid patches (or tiles) (100~101 km). Under this framework, realistic subgrid atmospheric variables at near surface level (e.g. temperature and humidity at canopy level) are required in many applications of land surface parameterizations or ecological models over vegetation (e.g., to drive photosynthesis models or to estimate the emission of greenhouse gases or biogenic VOCs). Despite the importance of subgrid near-surface data, they are usually unavailable except for selected locations. On the other hand, corresponding data at larger-scale (101~102 km) are usually available at a higher level (e.g. in the free troposphere) from global or regional atmospheric models or reanalysis data.

Therefore, we propose a method to downscale grid values at a higher level to subgrid-scale resolution canopy-level estimates, using an inversion technique. The strength of this method lies in its coupling between the two levels by taking account of boundary layer processes in a computationally inexpensive way. Specifically, a forward model which takes the target near-surface values as parameters is built in the first step, based on the concept of blending height and a simple mixed layer model. Then inversion theory is applied to estimate the posteriori values of the target variables, in which Markov Chain Monte Carlo simulation is adopted with the Metropolis-Hasting sampling algorithm. In this study, the method is tested against a simple synthetic dataset and the results indicate that a converged solution zone can be identified after a burn-in period. Within the solution zone, potential solutions can be identified as the points corresponding to the highest likelihoods, or with the help of data mining techniques (e.g., clustering). The location and shape of the solution zone within parameter space is found to be sensitive to the subgrid field of sensible heat flux and the temperature lapse rate in the free atmosphere, while the subgrid field of friction velocity, the horizontal scale of grid heterogeneity, and a-priori information of mixed layer height influence the distribution of potential solution within the solution zone. 9.174.141 on 4-5-2010-->

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner