Ensemble Markov Chain Monte Carlo Method for Assessing Uncertainties of Aerosol Properties from Multi-wavelength Lidar Measurements

An ensemble Markov chain Monte Carlo method of assessing uncertainty of aerosol properties from lidar measurements of extinction and backscatter is presented. The method applies the Metropolis-Hastings algorithm to an ensemble of Markov chains. Candidates are drawn from a hybrid random walk/independence sampler random generator. The independence sampler is formed by analyzing the ensemble partway along the chain to find regions of interests in which to generate random candidates. Convergence to the target Bayesian posterior probability density function (PDF) was found to be greatly expedited by also including sampling from the prior PDF. The Kolmogoroff-Smirnov test was applied to the evolving ensemble to verify convergence.