The fewer predefined parameters, the better? A comparative analysis of urban delineation methods using building footprint data in the Netherlands.

Authors: Van Migerode, C., Samardzhiev, K. and Caset, F.
Journal: International Journal of Geographical Information Science
DOI: 10.1080/13658816.2025.2469240

abstract

Bottom-up methods drawing on building footprint data (BFD) offer promising ways forward for delineating urban boundaries. These methods typically rely on predefined parameters, such as a minimum distance between buildings, which can be difficult to justify. While recent delineation efforts aim to reduce reliance on predefined parameters, analyses verifying if such methods produce ‘better’ or more ‘suitable’ results are limited. Moreover, empirically supported discussions around what dimensions constitute a method’s suitability remain thin on the ground. To this end, we introduce an evaluation framework structured around a specific set of empirical and practical dimensions that collectively represent the suitability of an approach. We employ the framework to compare four delineation methods using BFD that progressively trade parameter choices for data-driven optimisation procedures: (1) a grid-based method, (2) DBSCAN, (3) HDBSCAN – a locally optimised version of DBSCAN, and (4) a method based on the head/tail division rule. Applying these methods in the Netherlands, we find that DBSCAN is most effective in capturing the Dutch urbanisation pattern and show that methods with fewer predefined parameters, like HDBSCAN and the head/tail method, do not necessarily lead to more suitable delineations. We also discuss trade-offs in delineation methodologies more broadly.