Adaptive continuity-preserving simplification of street networks

TL;DR

The paper tackles the challenge of turning transportation-focused, highly detailed street networks into a morphology-friendly, continuity-preserving representation suitable for urban analysis. It introduces neatnet, an automated, open-source algorithm that combines topology verification, adaptive face-artifact detection, contiguity-based artifact classification (CES), and a two-pass geometry-replacement scheme to produce simplified networks with preserved flow. Across seven functional urban areas, neatnet generally outperforms state-of-the-art open tools (OSMnx, cityseer, parenx) in mirroring manually simplified ground truth while maintaining computational practicality. The work advances accessible, reproducible preprocessing for morphological studies and related applications, with potential extensions to broader data sources and downstream urban analyses.

Abstract

Street network data is widely used to study human-based activities and urban structure. Often, these data are geared towards transportation applications, which require highly granular, directed graphs that capture the complex relationships of potential traffic patterns. While this level of network detail is critical for certain fine-grained mobility models, it represents a hindrance for studies concerned with the morphology of the street network. For the latter case, street network simplification - the process of converting a highly granular input network into its most simple morphological form - is a necessary, but highly tedious preprocessing step, especially when conducted manually. In this manuscript, we develop and present a novel adaptive algorithm for simplifying street networks that is both fully automated and able to mimic results obtained through a manual simplification routine. The algorithm - available in the neatnet Python package - outperforms current state-of-the-art procedures when comparing those methods to manually, human-simplified data, while preserving network continuity.

Figures (14)

• Figure 1: Illustration of the simplification process by the example of a street network fragment in Seraing, Liège (Belgium). (a) The left panel shows the input data, downloaded from OpenStreetMap openstreetmap_contributors_openstreetmap_2025, before simplification. Due to transportation-focused mapping, the input network contains several intersections, roundabouts, and dual carriageways that are mapped with a high level of granularity. (b) The same network after simplification. The level of granularity has been reduced where appropriate, so that every intersection is represented by one node and every street segment is represented by one edge.
• Figure 2: Conceptual diagram of the proposed simplification algorithm. Input street network (0) undergoes topological verification and adaptation (1) before face artifact detection (2). The face artifacts are then classified based on contiguity of polygons and continuity of edges forming them (3). The result goes through the geometry replacement pipeline that alters the existing geometry (4) based on the artifact classification to ensure minimal change and preservation of main network characteristics. Steps 2-4 are then repeated before returning the output consisting of a simplified street network.
• Figure 3: A subset of the CES typology derived from the Liège street network. A single CES type is composed of the number of nodes forming the artifact and the continuity types of strokes forming its boundary. For example, the most common type, 3CES, is composed of 3 nodes and 3 continuity strokes, one of each kind, while 5S is composed of 5 nodes and a single continuity stroke of type S (as it never leaves the artifact). The total number of CES types is theoretically infinite given no upper bound on number of nodes exists. The full overview of types present in Liège is available in Appendix \ref{['appendix:ces']}.
• Figure 4: Chatterjee's $\xi$ correlation coefficient between properties of manually simplified networks and networks based on each of the tested algorithms vs. the original network as a baseline. Higher is considered better -- approaching $1.0$ on the $y$-axis.
• Figure 5: Euclidean distance between properties of manually simplified networks and networks based on each of the tested algorithms vs. the original network as a baseline. Lower is considered better -- approaching $0.0$ on the $y$-axis.