Applicability of surrogate-based MCMC-Bayesian inversion of CLM at flux tower sites with various climate and soil conditions

Community Land Model (CLM) parameters affect model simulations in different ways. Previous work has been done to evaluate the significance of the effects of hydrological parameters, and their identifiability from heat flux and/or streamflow observations at different flux tower sites with various climate and soil conditions. Such exploratory parameter screening and sensitivity analysis provide guidance on model calibration design; meanwhile, the exploratory ensemble simulations, associated with effective sampling of the input parameter space, can be used to developed reliable surrogate models linking input parameters and output responses such as latent heat fluxes (LH). In this study, we test the applicability of a surrogate-based Markov chain Monte Carlo (MCMC) inversion method, at different flux tower sites, to identify favorable climate and soil conditions for such inversion efforts. The unknown parameters are decided given our previous parameter screening work, samples are then generated using the quasi-Monte Carlo approach based on the prior distributions of the unknown parameters determined given minimum-relative-entropy theory. The sample sets are ingested into CLM to generate simulations that can be used for fitting polynomial surrogates up to the third order. The surrogates are finalized by compromising the training and testing errors, and integrated in the MCMC-Bayesian framework at each flux tower site. The applicability of the calibration approach for different site conditions is discussed.