Delineating the perceived functional regions of London from commuting flows

Urban growth leads to agglomerations which are not only geometrically dense but also spatially integrated with respect to the urban economy (Krugman, 1991). Well-connected urban areas defined by different types of urban flow, define the boundaries reflecting the relatedness of places in terms of their functionality (Batty, 2013; Batty and Cheshire, 2011). A functional urban region is thus normally characterised as a spatial system of places with a high frequency of intra-urban interactions (Karlsson and Olsson, 2006). Commuting patterns, for example, provide a much better account of the urban economic geography of a city than conventional economic statistics based solely on location (Ratti et al., 2010). However, attempts to define cities and their communities based on aggregated flow data normally neglect the inherent differences between different groups of people, as for example measured by social status, occupations or similar differentiations. To address this problem using disaggregate commuting data, we extract perceived functional regions for various occupations in the London Metropolitan Area at the Middle-layer Super Output Area level using community detection algorithms.

The method we use here is based on the multi-level modularity optimisation algorithm, also known as the Louvain method (devised by Blondel et al., 2008) which detects community structure within weighted flow graphs. The advantages of this method include its ability to grapple with large networks and the methodological robustness of its computation. The method produces an optimal partition of the network in a purely data-driven way by maximising its modularity, a measure proposed by Newman (2006) to quantify the degree to which the flows may be separated into different communities or spatial groups. Modularity is defined here as a function of the summation of the differences between all observed network flows and a random allocation of such flows across all communities, and it varies from −1 to 1 with larger values implying greater clustering within the communities derived.

The data used are the origin–destination flows differentiated by the eight occupational groups defined in accordance with the National Statistics Socio-economic Classification called here G1a, G1b, G2–G7. We first construct a weighted graph of the flows between places, and allocate these flows to different communities. We then compute the modularity, the difference between the flows and a random allocation, and then we can measure the difference in modularity produced by moving the nodes in each community out of that community to another, one at a time. We then select the allocation that increases the modularity the most. This continues until there is no further change. The resulting communities are then reclassified as single nodes with appropriate weighted network flows between each and the process begins again. The algorithm continues in this way moving to higher levels of aggregation until it converges giving the final allocation of base line nodes and flows to each community at the level reached which is assumed to be optimal.

Figure 1 presents the perceived functional regions for the eight occupations. The number of detected regions is smaller for the populations with more professional-managerial-oriented occupations than that for the workers in the less professional groups. It varies from 7 to 13. This is influenced by the difference between the population distributions and the associating commuting behaviours among these occupations: managerial occupations (G1a, G1b and G3) with more commuting affordability are able to travel longer distances than the other groups, thus making their perceived functional regions larger than those perceived by other occupations of workers who are more diffused across the London Metropolitan Area. The modularity scores, on the other hand grow steadily from 0.4 to 0.7, which shows that the goodness of the community partition is higher for the flow patterns of the lesser managerial occupations who have more segmented and local perceptions regarding their functional regions than the more managerial. The main reason for this could be the interplay between the distribution of different housing types and the differentiation of commuting habits.

OPEN IN VIEWER

Figure 1.

Perceived functional regions for occupations in the London Metropolitan Area.

When scrutinising the internal commuting flows, the main connections between the nodes are nicely captured by the communities. This reveals that the core routes of the transport network and various natural boundaries such as rivers, do play a part in the locational patterns of the derived communities for each occupation, but that the relative self-containment of functional regions is interpreted differently by different occupations. Reading, for example, seems to be perceived as an individual region by the supervisory and technical occupation (G5), but is part of a larger community for other occupations. Many comparisons of this kind can be made between the communities generated for the eight occupational classes in Figure 1. Computing and visualising the perceived functional regions in this way seems to capture the areas shared by different groups of commuters. This kind of representation is of great value in advancing our knowledge regarding how different places are perceived by different occupations with all the implications for future commuting that future planning for housing and employment will bring.