A parallel adaptive metropolis algorithm for uncertainty assessment of Xinanjiang model parameters.

Markov Chain Monte Carlo (MCMC) methods, which are popular for estimating parameter uncertainty of hydrological models, generally converge slowly, and are easy to get stuck in a local optimized region in the parametric space. In this paper we present a new MCMC sampler entitled the Parallel Adaptive Metropolis (PAM) algorithm, which is well suited to estimating the parameter uncertainty of hydrological models. The PAM algorithm provides an adaptive MCMC sampler to estimate the posterior probability distribution of parameters under a Bayesian framework. The performance of the PAM algorithm is greatly improved through parallel computing. The PAM algorithm is applied to assess the parameter uncertainty of the Xinanjiang model using hydrological data from Shuangpai Reservoir, China. The case study demonstrates that there is considerable uncertainty about the Xinanjiang model parameters.