Spatial immunization to abate disease spreading in transportation hubs

The study addresses how infectious diseases propagate via proximity contacts in crowded transportation hubs, which are inherently designed for efficiency and experience continuous inflow and outflow of people. Airborne pathogens pose particular challenges because transmission can occur through droplets and aerosols, with persistence on surfaces. Airports are sensitive nodes where infected individuals can seed outbreaks both at destinations and within terminals, potentially accelerating geographic spread. Given that travel restrictions alone tend to delay rather than prevent spread, the research question is whether targeted spatial immunization—reducing contagion probabilities in specific high-risk areas—can effectively diminish local transmission in an airport and, by extension, reduce regional and global dissemination. The work focuses on London Heathrow Airport, leveraging high-resolution mobility data to build a temporal copresence network and simulate disease spread under different compartmental models to identify contagion hotspots and assess mitigation by spatial immunization.

Prior research has examined transmission during travel on buses, cruise ships, trains, and airplanes, with risk influenced by pathogen infectivity, ventilation, trip duration, and occupancy. Screening strategies such as fever detection at airports have been explored, but less attention has been paid to transmission within transportation hubs themselves. The rapid global spread of viruses is facilitated by air travel networks, and hubs can influence both the velocity and geographic diversity of epidemics; partial travel bans often yield limited benefits beyond delays due to alternative routes. High-resolution studies of human contact networks using wearable RFID sensors (SocioPatterns) have characterized face-to-face interactions in contexts like conferences, hospitals, offices, museums, and schools, addressing data reconstruction, sampling biases, and implications for epidemic modeling. These works underscore the importance of accurate contact representations and the challenges of measuring temporal contact networks for modeling transmission risk.

Data and setting: GPS mobility data from anonymized, opted-in smartphone applications (Cuebiq) were collected within Heathrow Airport from February to August 2017 under a GDPR-compliant framework. To preserve privacy, analysis was confined to the airport perimeter. Users with fewer than 10 location points were excluded. Because a single day’s observed users were insufficient to reflect realistic foot traffic, trajectories from the entire 6-month period were supersampled to construct a synthetic “standard day” including 206,043 individuals (143,588 passengers and 62,639 workers), approximating a typical day at Heathrow. The 24-hour standard day preserves continuity for night-time trajectories. This standard day is replicated to simulate multi-day periods, with workers persisting across days and passengers renewed daily.

Copresence dynamic network: Space was discretized into 10 × 10 m² cells and time into 15-minute slots. Cells with fewer than a threshold number of visits (main results use threshold 30; robustness checked for 10–50) were treated as non-accessible and excluded from reconstruction. To compensate for discontinuous GPS sampling, trajectories were reconstructed by inserting intermediate cells along shortest paths consistent with terminal geometry between consecutive observations separated by ≤50 m, yielding roughly 4 million interpolated points and ~10 million total points. Users were classified as workers (present for ≥3 consecutive days with visits >4.5 hours and/or entering staff-only areas) or passengers. Passenger itineraries were classified semantically as departing, arriving, or connecting based on origin and destination within the airport; terminal-level flight destinations mapped passengers to London-arrivals, UK, EU, or intercontinental destinations. A temporal copresence network was built per time slot, linking individuals co-located in the same cell.

Epidemic models: Compartmental models were simulated on the temporal copresence network with 1000 stochastic realizations per scenario. The seed infected individual arrives randomly (arrival gate and origin) at 13:30 on day 1, and simulations run for 7 consecutive days (cloned contact sequences of the standard day). Three disease models were used: SIR (SARS-CoV-1-like), SEIR (H1N1 influenza), and SEIIR (COVID-19), with parameters chosen from the literature and adapted to the contact rate context. In SIR, infection occurs with per-contact probability p_s and recovery follows a Gamma-distributed infectious period with mean μ⁻¹. In SEIR, exposed individuals progress to infectious at rate γ; in SEIIR for COVID-19, the infectious compartment is split (prodromic and symptomatic/asymptomatic states) with relevant average durations.

Key parameters (Table 2):

• SIR (SARS-CoV-1-like): p_s = 0.92×10⁻³; mean infectious period μ⁻¹ = 10.6 days; calibrated to R0 ≈ 2.7 in a well-mixed approximation given observed contact rates.
• SEIR (H1N1 Influenza): p_s = 3.06×10⁻³; mean latency ε⁻¹ = 1.1 days; mean infectious period μ⁻¹ = 2.5 days; asymptomatic probability 0.33 with 0.5 relative infectivity; R0 ≈ 1.75.
• SEIIR (COVID-19): p_s = 4.31×10⁻³; mean latency ε⁻¹ = 3.7 days; prodromic period γ⁻¹ = 1.5 days; mean infectious period μ⁻¹ = 2.3 days; R0 ≈ 2.35. Severe disease/hospitalization states were aggregated into R as incompatible with being present in the airport.

Spatial immunization: For each model, infection events were logged at the cell level to identify hotspots. Cells were ranked by number of contagions occurring on day 1 across 1000 realizations. Spatial immunization consists of reducing per-contact infection probability p by 95% in selected cells (representing methods like far-UV-C, enhanced cleaning, filtration, ozone), and testing coverage of 50, 100, 200, 400, and 800 cells, corresponding to approximately 0.1%, 0.3%, 0.6%, 1.1%, and 2.3% of the 34,792 modeled cells. Immunized cells were chosen by hotspot ranking; for robustness and efficiency, SIR-based rankings were compared to disease-specific SEIR/SEIIR rankings. Additional robustness checks included redefining contacts with 30-minute time slots, which increased contacts and infections but preserved hotspot correlations and temporal patterns. Outcomes measured included the fraction of realizations with any secondary infections in the airport (PR), number and timing of new infections over the week, and infections disaggregated by category (workers; passengers arriving to London; UK, EU, and intercontinental destinations).

• The aggregated copresence network shows workers form a highly connected backbone due to recurrent presence and longer dwell times, while passenger-only copresence is fragmented; workers sustain connectivity and play a key role in transmission.
• Contagion hotspots are not strictly the most crowded areas; risk reflects a balance between local density and dwell time, highlighting security controls, services/retail, cafeterias/bars, some gates, and bottleneck areas inside terminals.
• SIR model (baseline, 15-minute slots): Ranking cells by day-1 contagions and immunizing top hotspots reduces outbreak likelihood and intensity. Immunizing 400 cells (~1.1% of space) reduces the number of realizations with secondary infections by about 20% relative to no immunization and markedly decreases day-1 infections. Temporal infection curves show daytime peaks, with the largest peak on days 3–4 (driven by infected workers). Spatial immunization both lowers and delays peaks; with 400 cells, early-day contagions are mostly prevented, the peak shifts toward the end of the week, and peak height drops to less than half of baseline. Increasing coverage further suppresses infections.
• Disaggregated impacts (SIR): All passenger categories (London arrivals, UK, EU, intercontinental) and workers experience reductions in infections over the 7-day period; reductions are less pronounced for workers and, to a lesser extent, intercontinental passengers. UK-destination passengers are least affected by outbreaks even without policies, reflecting their smaller relative contribution.
• Robustness to contact-time definition: Using 30-minute slots increases absolute contacts and infections but preserves hotspot locations and overall temporal dynamics, confirming stability of hotspot targeting.
• SEIR (H1N1 Influenza): Day-1 hotspot rankings from SIR and SEIR are nearly equivalent, making SIR an efficient proxy for hotspot identification. With 400 immunized cells (~1.1%), realizations with secondary infections decrease by about 30%. Over a week, infections concentrate during daytime and are sustained beyond day 1 mainly by workers; spatial immunization reduces infections across all categories, with smaller relative effect on workers.
• SEIIR (COVID-19): Using SIR-based hotspot ranking yields similar performance to SEIIR-based ranking. With 400 immunized cells (~1.1%), realizations with secondary infections drop by about 35%. Over a week, spatial immunization reduces per-slot contagions by roughly a factor of 3 with 400 cells, with further gains at higher coverage. Workers remain the most infecting category and the least protected by spatial immunization. Effectiveness persists under alternative R0 values for SARS-CoV-2 variants.
• Overall, targeting approximately 1% of airport floor space identified as hotspots yields substantial reductions in local transmission likelihood and intensity and delays epidemic peaks, potentially aiding downstream containment (e.g., contact tracing) and reducing seeding to external destinations.

The findings demonstrate that encoding complex individual movements into a temporal copresence network enables identification of high-yield transmission hotspots in an airport environment. Because workers constitute a recurrent, highly connected backbone, they sustain ongoing transmission beyond the first day, whereas passengers mainly contribute during short dwell times, particularly connecting travelers. Targeted spatial immunization of hotspot cells substantially reduces both the probability of any local outbreak and its intensity and timing, even when limited to around 1% of the total space. Importantly, day-1 hotspot rankings derived from a simple SIR model closely match those from disease-specific SEIR/SEIIR models, allowing a single hotspot configuration to be effective across distinct airborne diseases with different generation times. By reducing and delaying in-airport contagions, the strategy lowers the risk of onward transmission to London, UK, EU, and intercontinental destinations, thus mitigating regional and global spread seeded from the hub. However, because workers have prolonged exposure and recurrent presence, complementary measures (e.g., targeted vaccination, worker-specific protocols) may be needed to maximize protection for this group. The general approach is transferable to other transportation hubs and crowded facilities sharing similar flow and contact heterogeneities.

This study presents a generalizable framework to translate high-resolution mobility data into a temporal copresence network, simulate airborne disease spread, and identify contagion hotspots for targeted spatial immunization. Applying the method to London Heathrow shows that immunizing approximately 1% of floor space in top-ranked hotspots can cut outbreak likelihood by 20–35% (depending on the disease), reduce and delay infection peaks, and lower infections across traveler categories, with workers remaining the least protected. A simple SIR model suffices to rank hotspots effectively for multiple diseases, enabling a unified, efficient policy design. Future work should integrate complementary interventions aimed at workers (e.g., vaccination, improved PPE, scheduling to limit exposure), refine hotspot detection with real-time data streams, assess cost-effectiveness and operational constraints of different immunization technologies (far-UV-C, filtration, enhanced cleaning), and extend evaluations to other hubs (train/metro/bus stations) and large public venues.

• Disease progression and work behavior: For SARS-like diseases with severe symptoms, the assumption that infected workers continue working may be unrealistic, potentially overestimating sustained in-airport transmission by workers.
• Data and reconstruction: GPS data are discrete and required trajectory reconstruction (shortest-path interpolation up to 50 m), which may introduce errors in copresence estimation. The analysis uses 2017 data and supersamples six months into a synthetic day, which may not reflect current operations or behavioral changes.
• Contact proxy: Copresence in 10×10 m² cells over 15- or 30-minute slots is a proxy for transmission-capable contact and may over/underestimate true close contacts compared to face-to-face data.
• Parameter uncertainty: Infection probabilities and R0 values are drawn from literature and adapted to observed contact rates; actual transmissibility for emerging pathogens can differ. COVID-19 model aggregates severe disease states into R and lacks age structure.
• Intervention modeling: Spatial immunization efficacy is modeled as a 95% reduction in per-contact transmission in treated cells, a simplifying assumption that may not capture real-world performance variability of technologies (e.g., far-UV-C, cleaning, filtration). No other interventions (testing, masking, isolation) are modeled to isolate spatial immunization effects.
• Open system dynamics: The model focuses on early outbreak stages in an open hub with daily renewal of passengers; larger outbreaks might trigger operational changes (e.g., closures) not captured in simulations.