Abstract
In history, planning has been led by several philosophies, aiming on one hand at the production of self-sufficient small-scale urban nuclei and on the other hand at the functioning of the city as an entity, favoring networks planning and hierarchical development. The challenge of sustainable development obliges these philosophies to integrate, requiring precise tools to identify functional urban hubs. This paper applies Spatial Scan Statistics (SSS) to detect clusters of services in the Lausanne metropolitan area, demonstrating how a hybrid statistical baseline improves the detection of functional urban hubs.
The Challenge of Urban Clustering
Urban planners face a constant tug-of-war between creating self-sufficient local neighborhoods and managing the city as a massive, interconnected network. To do this effectively, they need to identify where services (like hotels and restaurants) are clustering relative to the underlying population.
The challenge is that "clustering" is relative. A high density of restaurants in a city center is expected, but the same density in a quiet residential suburb is a significant anomaly. Traditional methods often fail to account for the spatial distribution of these clusters or rely on stationary assumptions that don't hold up in a dynamic city. This study utilizes SSS to provide a statistically rigorous way to detect these "over-densities" and "under-densities" without needing to pre-define the size or shape of the clusters.
Methodology: Spatial Scan Statistics
The researchers utilize Spatial Scan Statistics (SSS), specifically the Poisson model. The process involves a circular scanning window that moves across the study area. At every point, the window expands in radius, creating thousands of overlapping "zones."
For each zone, the algorithm compares the number of observed events (e.g., hotels) inside the circle versus outside, conditioned on the underlying population. To ensure these clusters aren't just random noise, the authors use Monte Carlo simulations. They run 1,000 simulations of the data under a null hypothesis of total randomness; if the real-world cluster is stronger than 95% of the simulations, it is considered statistically significant.
The "Hybrid" Innovation
Initially, the study used resident population as the baseline. However, this created a "Resident Bias," incorrectly identifying areas like the Institute of Technology (EPFL) as massive service clusters simply because few people live on campus, making any single restaurant look statistically impossible.
To solve this, the authors introduced a hybrid baseline: Resident Population + Tertiary Sector Employment. By adding employment data, the model accounts for "daytime" populations. This weights the likelihood function more realistically, causing purely institutional clusters to disappear and actual service hubs to stand out.
Results in the Lausanne Agglomeration
The study focused on the Lausanne agglomeration in Western Switzerland, covering 68 municipalities and roughly 300,000 inhabitants. Data was aggregated into a high-resolution grid of 200-meter cells.
- Confirmed Hubs: Clusters 1 and 3 consistently highlighted the established city centers of Lausanne and Morges.
- Industrial Anomalies: Cluster 2 (Crissier) was identified as an industrial region where commercial centers provide services far in excess of the local population (1,200 times greater than expected).
- Model Sensitivity: When switching to the hybrid population model, Cluster 6 (the University/EPFL) completely disappeared, proving it was a statistical artifact. Meanwhile, Cluster 7 (Lakeside tourist areas) emerged, providing a clear signature of the tourism economy where service density is driven by visitors rather than local residents.
Conclusion
Spatial Scan Statistics offer a powerful tool for intra-urban analysis. While circular windows and static timeframes have limitations, the hybrid baseline approach significantly improves the accuracy of socio-economic mapping. Future work may expand this to track temporal dynamics or utilize non-circular windows to better capture linear features like transit corridors.
Summary updated