Derivation of urban structural area units for the analysis of population distribution

Abstract

Current research on the small scale population distribution faces the limitations of the coarseness of county level administrative units. But changes to the population distribution are expected to happen between finer grained spatial units, than those that are currently used. In this work the option to derive a constant delineation of categories such as suburban, urban sprawl or hinterland based on the morphology of the built up spaces in Germany is explored. To achieve that an algorithm for such a delineation of city borders from building density values is developed further, so that it suits a set of smaller cities than the original method was designed for. To enhance the results and adapt the algorithm to the set of smaller cities, several new implementations are introduced to the method, that allow for adjusting several parameters and dealing with certain challenges. The outcome of the algorithm is a multilayer raster, each containing the weighted value of an urbanity proxy for one of four different sets of cities, created according to their size. A random forest classifier is then applied to derive the targeted classes from the final raster. To find the most suited configuration for the newly added options, the accuracy values of different configurations of the algorithm are then compared. At the end the final classified raster with the configuration achieving the highest accuracy is presented and the steps leading to its creation are discussed.