Comparative analysis of different methods and obtained results for delineation of functional urban areas

Abstract

European Spatial Planning Observation Network (ESPON) recognizes Potential Urban Strategic Horizons (PUSH) and Potential Polycentric Integration Areas (PIA) as territory of one or more neighboring Functional Urban Areas (FUA). Delineation of FUA territory can be done by using general ESPON methodology, based on a 45-minute car travel time from the center of respective FUAs. This approach is based on network proximity by using shortest path in road network between two nodes. Later, results are approximated on administrative or statistical territorial units, so that PUSH areas are determined. However, other methods for delineation of FUA territory can be used. This paper deals with other methods that can be used for delineation of FUA territory. Some of those methods are based on machine learning, a branch of artificial intelligence which develops algorithms that take as input empirical data, such as that from sensors or databases. Created algorithms identify complex relationships thought to be features of the underlying mechanism that generated the data, and engage these identified patterns to make predictions based on new data. Clustering and artificial neural networks are some of approaches that can be undoubtedly used for delineation of FUAs territory, based on unsupervised learning and statistical data analysis. This is statistical approach, which clusters administrative or statistical territorial units based on statistical data, and not by network proximity. Such methods involve usage of Self Organizing Maps (SOM) which implies usage of neighborhood function to preserve the topological properties, or using k-means clustering, which partition observations into clusters by dividing space into Voronoi cells. Results obtained from both approaches will be analyzed in order to define the most appropriate method for FUAs territory delineation in Serbia.