GC43C-1206
Bayesian calibration of the Community Land Model using a multi-chain Markov chain Monte Carlo method
Thursday, 17 December 2015
Poster Hall (Moscone South)
Jaideep Ray1, Laura Swiler1, Maoyi Huang2 and Zhangshuan Hou3, (1)Sandia National Laboratories, Albuquerque, NM, United States, (2)Pacific Northwest National Laboratory, Atmospheric Sciences and Global Change Division, Richland, WA, United States, (3)Pacific Northwest National Laboratory, Richland, WA, United States
Abstract:
The Community Land Model, version 4 (CLM4) contains about 100 parameters, which are tuned to reproduce observations at the global scale. When CLM4 is used at a regional or site-scale, these parameters should ideally be re-calibrated. If the calibration is performed in a Bayesian manner, where the parameters are estimated as probability density functions, one practical approach is to first construct statistical emulators of CLM4 such that fearsome cost of Markov chain Monte Carlo (MCMC) solutions could be accommodated. On the other hand, it is often impossible to construct an accurate emulator when the CLM4 output of interest shows a complex behavior in the parameter space. In this work, we demonstrate the use of a multi-chain, parallel MCMC method to calibrate CLM4 without the use of emulators. We estimate CLM's hydrological parameters using measurements of latent heat flux at the US-ARM Southern Great Plains site, and compare these estimates against those obtained using a CLM4 emulator.
