This paper presents a method to downscale aggregated land use data based on statistical techniques. A purely spatial multinomial logistic regression (MNLR) model is proposed using observed fine resolution land use data. This model provides initial probability maps of land use presence, which are updated using aggregated land use data and an iterative procedure based on Bayes' theorem. The simplicity of the method as well as its low data requirements makes it easily reproducible. An example is shown using the CORINE land cover dataset (1990) to downscale future land use change scenarios (2020) for a small area in Belgium. The results from the MNLR as well as from the iterative procedure gave appropriate representation of land use patterns. The method was also useful in removing potential artificial border effects, which often arise when downscaling from adjacent spatial units. The resulting probability maps could be used for a variety of applications.
- Bayes' theorem
- Multinomial logistic regression
- Probability maps