Abstract Much contemporary ecology emphasises the importance of taking a spatial perspective in linking ecological patterns and processes. However, collecting and analysing spatial data is expensive. Here we compare spatially implicit and spatially explicit versions of a model of successional dynamics in a forest-shrubland mosaic in a mountain-top reserve in New Caledonia. The models are used to (i) understand the circumstances driving change in abundance of forest and shrubland, and (ii) compare the outcomes of spatially explicit and spatially implicit models of the same system. The spatially explicit model is grid-based and uses a spatially implemented ‘state-and-transition’ approach, with fire spread and seed dispersal the main spatial processes considered. The spatially implicit model is based on a transition matrix approach. Two alternative transition matrices were constructed, one based on field measurements and the other parameterised using output from the spatially explicit model. Although the averaged dynamics of the two models appear similar, the models make very different qualitative predictions about the landscape. Under the same initial parameter conditions two alternative landscape states emerge from the spatially explicit model; this is not the case for the spatially implicit model. Further, the spatial model produces outcomes much closer to those documented historically and inferred from the palæoecological record. The differences between the non-spatial and spatial models arise because, in this system, fine-scale interactions between landscape pattern and process are drivers of coarser scale dynamics, and such interactions are not included in the spatially implicit model. More generally, in order to understand coarse-scale spatial dynamics it may be important to consider local spatial patterns; spatially explicit models are those most likely to incorporate these.