Modelling Land-Use Change with Dependence among Labels

Abstract

Current literature on land use highlights the considerable methodological challenges in predicting how land will be used in the future. This paper addresses one of these challenges, namely the restrictive nature of the mono-class assignment, in which a spatial unit has only one elementary label at a time. We apply the multi-label concept in which a unit may have several elementary labels. For instance, a spatial unit may belong to residential and commercial classes at the same time. Classes in land use may be correlated, and taking into account their correlation may improve the land use changes prediction. For instance, a spatial unit has more chance to be, or to evolve to a residential unit if it is already commercial. The applied model achieves very promising results, indicated by values of 0.923 and 0.910 for precision and recall, respectively. The application described in this paper demonstrates the advantages of modelling the dependence among the labels for predicting the land use change.

Keywords: Bayes rule, dependence between labels, k-NN, land use, multi-label