A Bayesian MCMC approach to reconstruct spatial vegetation dynamics from sparse vegetation maps

Abstract

In studies of vegetation dynamics, data points describing the changes are often sparse, because changes were not recognized in early stages or investigations were part of different projects. The snapshots at hand often leave the nature of the dynamics unrevealed and only give a rough estimation of the directions of changes. Extrapolation of the dynamics with traditional cellular automaton modeling is also complicated in such cases, because rules often cannot be deduced from field data for each interaction. We developed a Bayesian MCMC method, using a discrete time stochastic cellular automaton model to reconstruct vegetation dynamics between vegetation maps available and provide estimation of vegetation pattern in years not surveyed. Spread capability of each vegetation type was characterized by a lateral spread parameter and another for establishment from species pool. The method was applied to a series of three vegetation maps depicting vegetation change at a grassland site following abandonment of grazing in north-eastern Hungary. The Markov chain explored the missing data space (missing maps) as well as the parameter space. Transitions by lateral expansion had a greater importance than the appearance of new vegetation types without spatial constraints at our site. We estimated the trajectory of change for each vegetation type, which bore a considerable non-linear element in most cases. To our best knowledge, this is the first work that tries to estimate vegetation transition parameters in a stochastic cellular automaton based on field measurements and provides a tool to reconstruct past dynamics from observed pattern.