Finite-state Markov chains are a useful tool for modelling communication channels with correlated fading and have recently also been applied with success to terrestrial free-space optical communication channels. However, the issue of how such Markov models should be optimised in order to accurately approximate the original continuous fading channel has not been addressed in a systematic manner. In this study, the authors improve on previous proposals by optimising the state space partitioning of the considered models. In particular, they investigate the properties and approximation accuracy of Markov models which are optimised according to information-theoretic considerations. They validate and evaluate their approach using a set of experimental measurements over a 12 km link distance. The obtained results confirm that optimised Markov models can provide better accuracy at lower state complexity, yet there remain shortcomings in capturing the autocovariance of the fading process.