Predicting the spatial distribution of population based on impervious surface maps and modeled land use change

Land use and changes in the spatial distribution of population are spatially and temporally linked and have an obvious impact on the urban environment. For instance, they influence the mobility and accessibility and play an important role in waste water management. This forecasting of the spatial distribution of population is thus a critical issue in planning. In order to allow this forecasting we have adjusted a multiple regression model to estimate the population distribution in function of land-use. The originality in our modeling strategy is the use of sealed surface proportion maps as weighting factor assuming that sealed surface proportion is a proxy of population density. The data exploited to adjust the parameters of the model are three time-series of landuse maps from the EU-MOLAND, census data and medium and high resolution remotely sensed images. We made use of these images in a spectral unmixing procedure that provides the sealed surface proportion maps. In the model, the population was normalized in order to get a model that is independent of time and space. This is required for prediction and spatial extrapolation which assumes a temporally and spatially stable relationship between land use, imperviousness and population density. We validated the model by means of a population disaggregation/re-aggregation procedures and tested its robustness regarding the resolution because predicted sealed surface proportion and predicted landuse maps using the calibrated EU-MOLAND model are generated at lower resolution (200 m) than the resolution used in the model adjustment. The results described in this paper regard the urban zone of Dublin.