Refractivity from clutter (RFC) estimation using a hybrid genetic algorithm-Markov chain Monte Carlo method

Abstract

A hybrid genetic algorithm - Markov chain Monte Carlo sampler (GA-MCMC) is introduced for estimation of low altitude atmospheric radio refractivity. This is done by inverting for the environmental parameters using the returned radar clutter data. A classical Bayesian framework is used so that the solution can be described in terms of a posterior probability distribution (PPD). An electromagnetic split-step fast Fourier transform (FFT) parabolic equation is used as the forward propagation model. The problem is solved with five different optimizers/samplers including the exhaustive search, genetic algorithms, Metropolis-Hastings and Gibbs samplers, some of which were used in previous literature, as well as the new GA-MCMC hybrid based on the nearest neighborhood algorithm (NN). The results show that the new hybrid method improves the speed of a conventional MCMC sampler by a factor of 10 or more while conserving the accuracy in estimating the probability distributions of the inverted parameters