Supplementary material to "Optimising CH₄ simulations from the LPJ-GUESS model v4.1 using an adaptive MCMC algorithm"

Abstract

The Adaptive Metropolis algorithm used here contains three key concepts explained in the following sections. The adaptive random walk, allowing the algorithm to learn features of the target distribution; transformed proposals, providing a natural way of including limits on the parameters; and tempering of the target distribution, to reduce the effects of local maxima and allowing better exploration of the target.