Modelling spatial association in pattern based land use simulation models.

Pattern based land use models are widely used to forecast land use change. These models predict land use change using driving variables observed on the studied landscape. Many of these models have a limited capacity to account for interactions between neighbouring land parcels. Some modellers have used common spatial statistical measures to incorporate neighbour effects. However, these approaches were developed for continuous variables, while land use classifications are categorical. Neighbour interactions are also endogenous, changing as the land use patterns change. In this study we describe a single variable measure that captures aspects of neighbour interactions as reflected in the land use pattern. We use a stepwise updating process to demonstrate how dynamic updating of our measure impacts on model forecasts. We illustrate these results using the CLUE-S (Conversion of Land Use and its Effects at Small regional extent) system to forecast land use change for the Deep Creek watershed in the northern Okanagan Valley of British Columbia, Canada. Results establish that our measure improves model calibration and that ignoring changing spatial influences biases land use change forecasts.