Resilience and recovery of public transport use during COVID-19

The study examines how public transport use in Kunming, China, varied and recovered during the first COVID-19 wave and subsequent reopening, addressing a gap in understanding the recovery dynamics of human mobility. While prior work linked mobility to transmission and assessed interventions, few studies tracked long-term, group-specific recovery in public transport usage. The authors define travel behavior resilience as the co-evolution of transport supply and demand back to a (temporary) equilibrium and hypothesize that willingness to travel recovers first, spatial mobility indicators recover more slowly than temporal ones, revisitation increases over exploration, and commuters display greater resilience than the elderly. Understanding these dynamics is important for transport management, urban restoration, and sustainability amid pandemic-related disruptions.

Prior studies have used geolocation and mobility data to relate human movement to COVID-19 spread, evaluate control measures, and predict waves across spatial scales. Transport research during COVID-19 has often relied on aggregated data, surveys of behavior and attitudes, or has documented modal shifts away from public transport toward private modes, raising sustainability concerns. However, few have measured how urban mobility varies and recovers at the group level over time. Concepts of system resilience (engineering) have been applied to transport supply, e.g., reopening and service frequency restoration, while social-ecological and psychological notions of resilience inform demand-side behavioral adaptation. The paper builds on these literatures by operationalizing a longitudinal measure of travel behavior resilience for public transport users and explicitly testing group differences in recovery.

Data: Anonymized subway trip records (smartcard and mobile payments) from Kunming, China, covering November 1–28, 2019 (pre-pandemic baseline) and January–May, July, and September 2020 (data for June and August 2020 unavailable). The city experienced confirmed cases from January 22 to February 19, 2020, and lockdown from January 28 to February 29, 2020. Sample: Frequent travelers are defined as individuals traveling at least 3 days per week for four consecutive weeks in November 2019. Subjects are those who remained frequent in September 2020 using the same identifier. The study tracks 16,403 subjects producing over 3.5 million trips across ten months. Subjects are categorized as commuters (36%), elderly (5.4%, age ≥60), students (4.2%), and others (54.4%). Commuters’ workplace-nearest station is inferred from November 2019 spatiotemporal commuting regularity. Indicators and computation: Weekly, for each subject, compute: (1) whether the subject traveled (binary); (2) traveled days; (3) number of trips; (4) total trip distance; (5) activity space (minimum convex polygon of unique boarding/alighting stations per week); (6) number of distinct stations visited. Pre-pandemic weekly averages are computed from November 2019. Group-level rates of change are calculated as ratios of weekly indicators to pre-pandemic baselines using formulas provided (N, D, Q, T, A, S). Week-to-week two-sample t-tests assess sequential changes for each group and indicator. Resilience framework: Travel behavior resilience is conceptualized via a resilience triangle using mobility reduction magnitude, reduction duration, and recovery duration relative to pre-pandemic levels. Recovery point is when an indicator returns to and stabilizes at pre-pandemic level (or co-evolves to a new equilibrium, though analysis focuses on recoveries to pre-pandemic levels). Revisitation analysis: For each subject and month (November 2019, March 2020, September 2020), rank stations by trip frequency and compute trip proportions by top-ranked stations to assess revisitation vs exploration tendencies. Recovery duration regressions: For subjects whose weekly trips, total distance, or activity space dropped to zero during the first wave, compute recovery duration (weeks from zero to restoring pre-pandemic level) and regress on pre-pandemic levels (trips/week, total distance/week, activity space/week) by group (commuters, elderly, others).

• Three phases in public transport use: drastic reduction (~3 weeks starting January 22, 2020), rapid growth, then stabilization. By March 21, 2020 (Week 10), most indicators exceeded 50% of pre-pandemic levels (students excluded due to school closures).
• Willingness to travel recovered first: The proportion of subjects who traveled fell fastest to a nadir of 3.34% during the sharp reduction, then returned to pre-pandemic level by Week 23 (from September 1, 2020), followed by traveled days (Week 25), and trips and total distance (Week 26), supporting Hypothesis 1.
• Spatial indicators lagged: Activity space and stations visited did not fully recover by September 28, 2020, reaching 94.72% and 97.43% of pre-pandemic levels, respectively. Activity space changes were less statistically significant across weeks than other indicators, supporting Hypothesis 2.
• Strengthened revisitation: For frequent travelers, ~95% of trips were concentrated at the top 4 stations per person pre-pandemic, with 71.5% at the top 2. During recovery, concentration at the top 2 stations increased to 78.3%. Despite trips/week reaching 103.65% of pre-pandemic levels by September 2020, activity space remained at 94.72%, indicating reduced destination diversity and increased revisitation (Hypothesis 3).
• Group differences in resilience (Hypothesis 4):
  • Commuters’ indicators increased earlier (from Week 4), with ~6 weeks of continuous rapid growth. Their traveled days stabilized near five days/week; they exhibited shorter reduction and recovery periods and smaller resilience triangles than the elderly. The proportion traveling decreased to 4.40% at the trough for commuters vs 2.60% for the elderly.
  • The proportion of commuting trips increased significantly for 11 consecutive weeks until end of April 2020; commuting share declined from 80.60% pre-pandemic to 71.08% during stabilization, consistent with ongoing work-from-home/flexible schedules.
  • Elderly recovery lagged by ~2 weeks and remained lower; their recovery rates were ~80% slower than other groups. Elderly decreased frequency but tended to preserve activity space more than their trip frequency.
  • Students’ mobility tracked school closures and reopenings, serving as a control for discontinuities.
• Regression of recovery duration vs pre-pandemic levels:
  • Trips: commuters y = 0.49x + 8.13 (R² = 0.71); elderly y = 0.42x + 10.55 (R² = 0.82); others y = 0.22x + 10.74 (R² = 0.48).
  • Total distance: commuters y = 0.01x + 11.61 (R² = 0.44); elderly y = 0.02x + 13.06 (R² = 0.17); others y = 0.01x + 12.41 (R² = 0.39).
  • Activity space: commuters y = 0.05x + 12.80 (R² = 0.71); elderly y = 0.04x + 9.29 (R² = 0.12); others y = 0.03x + 12.88 (R² = 0.43). These indicate higher pre-pandemic mobility often entails longer recovery; commuters restored frequency and distance faster than spatial extent, while elderly restored spatial extent relatively faster than frequency.
• System-wide observations: In September 2020, infrequent travelers reached ~40% of pre-pandemic mobility, whereas frequent travelers nearly recovered. Only 21.87% of pre-pandemic frequent travelers remained frequent by September 2020 under the same identifier.

Findings show that even in a single-wave city, public transport use required roughly half a year to recover, underscoring the need for sustained interventions and adaptive transport management. Recovery sequences reveal that willingness to travel and temporal intensity (frequency, distance) recover sooner than spatial diversity, with revisitation intensifying and exploration weakening, leading to smaller activity spaces. Group heterogeneity is pronounced: commuters exhibit greater resilience, rapidly re-establishing commuting patterns, whereas elderly users recover more slowly, reduce frequency, and maintain narrower but persistent spatial footprints, reflecting risk aversion and differing trip purposes. These insights imply policies should be tailored to mobility groups, considering age-related vulnerabilities and the persistent impact of work-from-home and flexible schedules on commuting shares. Stimulating safe intra-urban mobility and addressing reduced trip purpose diversity can support broader urban restoration, particularly in contexts adhering to stringent control measures.

The study introduces a generalizable framework to measure travel behavior resilience using longitudinal smartcard data and group-level mobility curves, applying it to Kunming’s subway users across COVID-19’s first wave and recovery. It demonstrates slow, phased recovery; faster restoration of travel propensity and temporal indicators than spatial ones; intensified revisitation; and marked group differences in resilience, with commuters more resilient than elderly users. Policy recommendations include accounting for group heterogeneity, sustaining targeted interventions during extended recoveries, and considering strategies to safely stimulate intra-urban activity. Future research should quantify re-equilibrium when demand does not return to pre-pandemic levels, and assess implications for traffic congestion and carbon emissions, ideally leveraging datasets longer than a year to mitigate seasonal effects.

• Single-city case (Kunming), which may limit generalizability.
• Data span of 10 months with missing weeks (no data for June and August 2020), potentially affecting continuity and seasonality control.
• Analysis focuses on frequent travelers who recovered to pre-pandemic levels, not on persistent new equilibria where demand remains below baseline.
• Does not directly evaluate impacts on traffic congestion or carbon emissions.
• Elderly categorization relies on fare concession data; equal-contribution note (superscript 6) is not an affiliation but may affect interpretation of author roles, not results.