In praise of (spatial) bundles

Systematic characterisation of the way cities look and the composition of their fabric form much of the “bread and butter” of this journal. In this editorial, we focus attention on an approach to classifying the (urban) landscape that we call spatial bundles - a mixture of the combination and arrangement of a series of lower-level atomic units. From the geodemographic classification of populations to the definition of functional areas based on activity spaces, exploring how (new) data and methods can be combined to produce increasingly more detailed, timely, and accurate representations of cities is what EP-B is all about. These endeavours are also sorely needed. There is much we do not know about how cities combine their ingredients (built environment, people, opportunity) to result in more than the sum of their parts; and there is much of that “recipe” we need to change through design, policy, and development if we are to make cities work for the next century, which will be dominated by climate and social change. Much (though not all!) of this understanding starts from quantitative representations that capture the elements of interest. It is not always true that, paraphrasing the famous management motto, you can only improve what you measure, but it is true that granular, timely and accurate evidence helps, sometimes decisively. For this editorial, we suggest that there is value in a new breed of classifications that we term spatial bundles. These are not entirely new, and we are certainly not the first ones to suggest their existence. But the recent appearance of new sources of data is making them more feasible and useful. We are convinced there is enough untapped potential in this approach that warrants calling the attention of the EP-B community and embedding some of their features in common practice.

Spatial bundle classifications (or, for simplicity, bundles) are descriptions of the landscape based on the combination and arrangement of a series of lower-level, atomic units like buildings, streets, small administrative areas, or regular geographies such as squared or hexagonal grids. As such, bundles require an atomic level characterisation describing the core components of the landscape, but also a grammar to describe how these are spatially arranged. Bundling is thus an approach that emphasises mix and configuration as much as components. We see bundles as complements to atomic classifications that provide additional insight into the way that atomic units combine into stylised patterns of development. By definition too, a bundle is expressed at a geographical scale that is an aggregate of that at which the atomic units are defined. Therefore, an important part of bundling well relies on choosing the appropriate geographic scale at which bundles should operate.

Bundles and bundling can be applied in a wide range of contexts, depending on the nature of the atomic units and their attributes. The important feature of atomic units is that they describe the landscape in a granular and discrete way. These are the building blocks, the units and their description based on which the landscape can be structured and understood. Of course, depending on which aspect of the landscape one is interested in, these will vary. But atomic units provide a thorough way of covering the geographical extent we are interested in classifying. Examples of typologies that are limited to atomic units abound, and include traditional land cover and land use classifications, but one could also see a typology of buildings or street segments as a fitting illustration.

If atomic units provide everything that makes up the landscape, what is the value of bundles? The answer, as the Geographer would say, resides in scale and relation. Atomic units – such as building footprints - sometimes provide too much resolution and spatial detail. Which is to say, there are cases where the processes we are interested in operate at geographically coarser scales than is provided by atomic units. Policies are rarely enacted at the pixel, plot, or building level. Neighbourhood effects operate, well, at the neighbourhood scale. Understanding urban vibrancy depends, of course, on granular representation - for example at the building or street level - but ultimately, it relates to broader scales such as the entire city. This is not simply a matter of where the line is drawn to measure things, although that in itself is important enough. It is also about what falls within that boundary, and how atomic units are spatially arranged, internally. Increasingly, we are realising that composition is not enough to understand cities. Two areas may have the same amount of residential and commercial land (to extend the hint at land use classifications provided above), but if those are fully segregated in one and perfectly mixed in the other, the picture they paint and the outcomes they will have in a range of outcomes (e.g., walkability, liveability, emissions) will vastly differ. Thus, the value of bundles resides in capturing, on the one hand, what is the right scale at which atomic elements combine for the outcome of interest; and, on the other, characterising the layout of such elements within space.

Let us unpack what makes a classification a bundle to clarify the concept a bit. There are three components - an atomic unit with attributes describing itself, a characterisation of the spatial pattern, and a classification incorporating both. Atomic units can be any spatially delimited unit of analysis, from those directly reflecting components of landscape, like buildings, to abstract grids. Each can be described with a “type” or “score” using one or more attributes that capture the aspect of the landscape a researcher is interested in. What makes a unit atomic is the nature of these attributes, which are assumed to be derived from the unit itself as if it were isolated in space. In other words, these attributes are devoid of (spatial) context. Geography is directly included in the next step. The ways of doing so differ, as space can be embedded in many ways, ranging from explicit inclusion in the clustering approach (as in regionalisation; e.g., Duque, Ramos and Surinach, 2007) through to the capture of spatial patterns using scores of access (e.g., Tenkanen et al., 2023) or association (e.g., Anselin, 1995). The important element here is that the characterisation of the spatial pattern composed of atomic units becomes core input of these methods alongside the atom’s attributes: space as a first-class citizen. Bundles can be either organically delineated by joining up contiguous atomic units in the same class or aggregated at a pre-defined set of known higher-order boundaries. Given there is no restriction on the number of atomic attributes used to characterise the atomic units or their nature, bundles may reflect one perspective of looking at the landscape or combine different ones. What is shared by all bundles is the nature of classes that reflect the mix of information originally captured at the atomic level and its spatial configuration. In the end, it is the (spatial) pattern that makes up a (spatial) bundle.

Bundles are not entirely new. For example, Glaster (2001) already defines neighbourhoods as a “bundle of spatially based attributes”. We believe neighbourhoods are a good conceptual example of the more general spatial bundle approach we outline here. However, we feel the time is right for the concept to receive more attention. Their flexibility makes bundles a great way to combine and integrate disparate sources of data, and the granularity they afford can now be fully realised with new sources of data. This very journal has published several methods using the logic of spatial bundles. To name a few, the curious reader can look at the work of Tomal and Helbich (2022) employing regionalisation for housing submarket delineation; the method by Araldi and Fusco (2019) developing a classification of urban fabric from atomic units derived from proximity bands around street segments, and the application of network-constrained local indices of spatial association (ILINCS) within a clustering algorithm; or the work of one of us, Fleischmann et al. (2022), developing a classification of urban form through the clustering of spatially smoothed attributes.

Bundles are not here to replace but to complement. They do not substitute for land cover and land use classifications, which are still useful for many applications, and form the atomic basis of many bundles. Instead, bundles add value by providing information beyond what traditional classifications offer. In doing so, they take landscape classifications from their foundational nature, more obvious at the atomic level, into a more applied and use-tailored place. This implies less “for everything”, but (much) better “at some things”. For example, the now famous 15-minute cities (Moreno et al., 2021) are all about land use. But a traditional classification of land use only serves to provide the foundation for studying 15-minute cities. A bundle built around access to certain uses (e.g., amenities) would provide a much closer measure of the (policy) relevant phenomenon at hand. Bundles are also excellent bridges to alternative sources of data because they focus on the pattern a set of components make up rather than on the components themselves. A good illustration of this is satellite imagery. Even with contemporary advances in computer vision, it is proving challenging to develop algorithms that can extract building footprints or street networks correctly. These are the features that provide the basis for many classifications of cities. Yet, the pattern that buildings, streets, and other elements of the built environment combine to create is much easier to recognise all at once in an image captured from above, than it is to extract each one of these elements to then create the pattern subsequently. This is an under-recognized advantage we believe will become more prominent in the coming years.

It is no coincidence that we have decided to place attention on bundles at a time when new data sources of all types are becoming available. The literature has long pointed out that measuring a phenomenon requires identifying the right spatial scale at which it operates, and then actually measuring it at that geographical level. For a long time, this has been an interesting but mostly intellectual thought experiment: one could spend entire brain cycles on the first part (identifying the spatial scale), but would most likely be stuck with “given” geographies to measure phenomena of interest. Most of the data available would come from a handful of providers (e.g., offices of national statistics) and then be aggregated to a point where there would be little room for manoeuvre (e.g., census geographies). We now live in a different world. Complementing all traditional sources of data, current urban researchers have a wide host of additional ones with which to work on the second requirement to measure phenomena mentioned above, from smartphone pings to high-frequency, high-resolution satellite images, to name a few. In many cases, thinking about the right spatial scale is not a hypothetical any longer but an imperative if we want to push the envelope. Spatial bundles are an example of what is possible in this new world. And this editorial is thus a small plea for more researchers to engage with these ideas, jump full body into this brave new world and, perhaps, submit the results of such adventures to this journal.