Neutral models for polygonal landscapes with linear networks

Dynamical mosaic models are very useful in ecology to study the influence of patterns on some ecological processes. Examples of such mosaics are patches of agricultural or forested landscapes. They are often built on the basis of either explicit processes or on neutral approaches using random or almost random element generation. Yet, most of the latter, called neutral landscape models, use a grid-based scheme and simulate pixel mosaics that are autonomous and/or independent. This scheme is not perfectly adapted to anthropogenic patchy landscapes made of contiguous and uniform polygons. Landscape ecology for example explains how a landscape may be seen as a complex mosaic of patches and corridors, with dynamical compositions (the patch attributes) and configurations (their shapes and neighbourhoods). This work presents three neutral landscape models dedicated to manipulations of polygonal patchy mosaic configurations as compared to an observed one. These models have respective advantages that can be summarised in an increasing ability to simulate realistic anthropogenic landscape mosaics. The tessellation approach is simple and rapid, but very much constrained by the patch shapes and positions. The Gibbs algorithm method adapted to landscapes is powerful to manipulate patch positions, with still a weak control on patch shapes. The last method based on a Delaunay triangulation technique offers the opportunity to modulate patch shapes and can still be combined with the Gibbs method to optimise the patch positions. With such neutral landscape models, it is possible to explore ecological hypotheses using a wide range of controlled patchy mosaics.