Recurrent visitations expose the paradox of human mobility in the 15-Minute City vision

TL;DR

The paper tackles whether proximity-based urban planning ideas like the 15-Minute City translate into actual local use. It introduces the K-Visitation framework to compare recurrent mobility ($K_{ ext{freq}}$) with proximity-based exposure ($K_{ ext{dist}}$) using a greedy minimisation and the Jaccard similarity $q_K$, informed by 18 months of Finnish mobility data. It reveals a paradox: central, amenity-rich cores exhibit misalignment with local access as residents overshoot proximate options, while peripheral areas show more locally constrained routines; longer trips frequently target specialised amenities, and the social impact of localism is spatially contingent. A permutation null model, machine learning analyses, and elasticity of segregation elucidate the behavioural, infrastructural, and social nuances, underscoring that proximity alone is insufficient for proximity living and that policy must be place-sensitive and balance local provision with access to specialised amenities to avoid equity trade-offs.

Abstract

In the transition towards sustainability and equity, proximity-centred planning has been adopted in cities worldwide. Exemplified by the 15-Minute City (15mC), this emerging planning paradigm assumes that proximate amenity translates into localised utilisation, yet evidence on actual mobility behaviour remains limited. We advance a behaviourally grounded assessment by introducing the \textit{K-Visitation} framework, which identifies the minimal set of distinct visitations needed to cover essential amenities under two orderings: one based on observed visitation frequency ($K_{\text{freq}}$), and the other based on proximity to home ($K_{\text{dist}}$). Applying it to an 18-month, anonymised mobility data from Finland containing 720 thousand users, we directly compared local mobility potentials with recurrent destination choices, revealing a paradox of human mobility within the 15mC framework. A clear misalignment is observed between proximity and recurrent behaviour, most pronounced in urban cores--areas boast with amenities and traditionally viewed as ideal settings for local living--where residents voluntarily overshoot nearest options, while peripheral routines remain more locally constrained. The paradox further revealed asymmetric functions influences, as compared with everyday amenities, individual travels significantly further for to encounter specialised functions. Furthermore, the social consequences of localism are spatially contingent: increased reliance on local options reduces experienced segregation in central districts but can exacerbate it elsewhere. Our findings stress that proximity is necessary but insufficient for achieving the proximity living ideal; implementation of the 15mC should be behaviourally informed and place-sensitive, coupling abundant local provision of routine needs with access enhancement to specialised amenities to avoid unintended equity trade-offs.

Figures (22)

• Figure 1: a. Schematic illustration of the K-visitation approach. For a given individual, $K_{\text{freq}}$ (in red) is represented by a selection of frequent visitations, with $K_{\text{dist}}$ representing proximity to home (in blue). K-visitations represents the minimal number of visitations, $K$, required to cover all amenity categories (whether frequently visited, or close to home). The intersected places are the overlap of recurrent and proximate mobility. b. Probability density distribution of the mobility alignment coefficient ($q_K$) in three scenarios for all Finnish cities. Removing work-time anchors increases local alignment, while the randomised model validates the tendency to concentrate over distance-decay function. c. City level $q_K$ distribution for Finland's major urban areas, where each curve represents a city. Despite differences in population size and amenity supply, all cities exhibit a pronounced mode between 0.6 and 0.7.
• Figure 2: a. National population distribution on 1 km grid system; black boxes indicate the extent of panels b-e. b-e. Local $q_k$ values mapped at 1 km resolution for b. the Helsinki metropolitan area (consisting of the city of Helsinki, Espoo, Kauniainen and Vantaa), c. Turku, d. Tampere and e. Oulu. Cell colours interpolate from green (low alignment) to red (high alignment). Cell size represents POI density in each grid. Central districts consistently exhibit weaker behavioural-proximity concordance than their surrounding peripheries, despite rich in local POI provision.
• Figure 3: a-b. Cumulative share of the population able to reach all K-visitations (excluding work/school) within given travel time by public transport (PT): a.$K_{\rm freq}$ and b.$K_{\rm dist}$. The vertical dashed line marks the 15-minute threshold. c. SHAP values for the top predictors in the tuned XGBoost classifier of frequent but non-proximate visitations, coloured by feature value (low to high). d-e. SHAP dependence plots for travel time: d. driving and e. public transport. Points are coloured by the experienced segregation level of destinations, with marginal histograms showing the distribution of travel times. Greater travel times strongly increase the model’s prediction of non-proximate habitual trips.
• Figure 4: a. Standardised share of amenity categories across visitation types, showing the relative prevalence of each category in proximate, recurrent, home-area, work-area, and other visitations (the bar represent 90% confidence intervals). Home-area and work-area is defined as in the same H3 hexagon as inferred home/work location. b. Relationship between the closest available distance from home (x-axis) and the excess distance travelled when selected in $K_{\text{freq}}$ compared with $K_{\text{dist}}$ (y-axis), plotted for individual amenity classes. Points are coloured by functional category and scaled by the number of POIs.
• Figure 5: Bivariate map of $\epsilon_s$ with respect to increased $q_K$ across 1 km grid in the Helsinki metropolitan area. Each cell is coloured by the combination of resident income quartile (Q1--Q4) and elasticity quartile (E1--E4), as shown in the diamond legend (Q1--E1: low income & more mixing with higher $q_K$; Q4--E4: high income & stronger segregation with higher $q_K$).