The Twin2Expand Project is tackling an ambitious challenge: modeling urbanised areas across the European Union (EU). Our goal is to better understand the walkability of street networks and how easily pedestrians can access amenities such as green spaces and grocery stores. Traditionally, creating these models required detailed manual work, including constructing street network datasets and integrating numerous data sources such as points of interest and population densities.
In our automated approach, we use network data from open sources such as OpenStreetMap (OSM) and Overture Maps. These datasets offer a broad view of urban networks but require cleaning and simplification. Unlike city-specific data sources, OSM and Overture Maps can be encumbered with extraneous details (for the purposes of network analysis), which distort the outcome of the measures.
Our cleaning process involves several steps. We start by extracting nodes from the raw data and connecting these with links (edges). Curved streets are represented with many small segments to capture their curvature; this introduces multiple nodes along street segments, which can distort the outcome of network centrality measures. To address this, we extract the geometric representation of roads from their topological structure. This means distinguishing between the actual network connections (intersections) and the way roads are described geometrically.
Yet, complexities remain, particularly with features such as complex footpaths in green spaces and intricate intersections at motorway flyovers and roundabouts. These elements can introduce redundancy and complicate network analysis, raising the question of whether we can further simplify the network while preserving its essence.
Over the past few years, our research community has made significant progress in developing routines capable of automating network cleaning, such as collapsing parallel routes and simplifying complex intersections. However, the challenge remains in formally chaining these methods together—determining which street network segments to extract from OSM (footpaths, motorways, etc.), which sequence of tools to apply, and which distances to use when collapsing parallel edges or simplifying complex intersections.
We’re currently approaching this by comparing different automated workflows to manually prepared networks. For example, we’re examining how network centrality measures from automated models in cities like Madrid compare with those from manually curated networks, which helps us learn what work best and how reliable the outcomes might be.
We look forward to sharing the outcomes of these methods with the wider community!
