Assessing the Accuracy of Changes in Spatial Explicit Land Use Change Models

Results of land use change models are often assessed by comparing simulation results with actual land use data. For this the Kappa coefficient of agreement is a common algorithm for map comparison; it expresses the agreement between two categorical datasets, corrected for the distribution of class sizes. However, most land use has a high inertia over the length of a typical simulation period, and therefore the changing land use comprises only a small part of the map. Since methods for accuracy assessment do not acknowledge this inertia, measured accuracies are generally very high, which suggests that models are very accurate. However the similarity between simulation results and actual land use data depends on the amount of land use change at least as much as the accuracy of simulated land use changes. To gain more insight in the performance of a land use change model, it should be acknowledged that most land use persists rather than changes. Therefore the expected agreement is much higher than what is computed from the distribution of class sizes only. This paper describes Ksimulation, a method that does consider the amount of change in the accuracy assessment of results of land use change models. This method is illustrated with a simple example and then applied to the results of a case study.