The recovery process of a Dutch heathland after fire is investigated. The study area, 12 m x 20 m, has been surveyed yearly between 1963 and 1993. Previous work has shown that a stationary Markov chain models the observed recovery process well. However, the Markov model fails to capture an important observation, the existence of a phase structure. The process begins deterministically, but small random (non-Markov) effects accumulate through time and at some point the process suddenly becomes noisy. Here we make use of the spatial information contained in vegetation maps to examine dynamics at a fine spatial scale. We find that the phases observed at a large spatial scale separate themselves out distinctly at finer spatial scales. This spatial information allows us to investigate hypotheses about the mechanisms governing deterministic versus noisy vegetation dynamics.