Changing travel patterns in China during the early stages of the COVID-19 pandemic

The study addresses how human mobility in China changed during the early COVID-19 outbreak (January–March 2020) and the implications of those changes for disease spread and public health response. The context is the overlap of the initial Wuhan outbreak with Chunyun (the 40-day Lunar New Year travel period), the largest annual human migration, potentially accelerating transmission domestically and internationally. Wuhan instituted a cordon sanitaire on 23 January 2020, one day before the Lunar New Year holidays, drastically restricting movement. While prior work assessed the effectiveness of such restrictions, key uncertainties remained regarding whether observed travel patterns reflected normal seasonal movements or responses to the epidemic/interventions. This study uses Baidu Huiyan mobility data to characterize spatio-temporal travel patterns from Wuhan and across China, assess network-level changes, and evaluate how holiday-related mobility shifted COVID-19-related healthcare pressure relative to regional healthcare capacity.

The paper situates its work within studies evaluating travel restrictions and mobility during COVID-19 and prior pandemics. Previous analyses used Baidu mobility data in transmission models and examined changes around Wuhan. Studies have assessed non-pharmaceutical interventions and mobility’s effect on COVID-19 spread in China (e.g., Kraemer et al., Tian et al., Lai et al.), and travel restriction impacts (e.g., Chinazzi et al.). Literature on Chunyun documents billions of journeys annually and migration dynamics along the urban hierarchy. Additional references discuss effective distance on networks, holiday impacts on respiratory epidemics, and mobility tracking for disaster/outbreak response. The authors note a gap in distinguishing regular holiday travel from outbreak/response-driven movement and in understanding how holiday mobility shifted healthcare demand across regions.

Design: Observational analysis of human mobility patterns across Chinese prefectures from 1 Jan to 1 Mar 2020, integrating movement, epidemiological, demographic, and health system data. Data sources: - Mobility: Baidu Huiyan (Location-Based Service) daily prefecture-level migration indices and top-100 OD proportions, aggregated from 8-hour windows. Constructed symmetric 366×366 daily connectivity matrices for 61 days (1 Jan–1 Mar 2020). Travel volume from i to j on day t: T_ij,t = F_i,outbound,t × P_ij,outbound,t; validated where possible by inbound equivalence. - Epidemiology: Daily confirmed COVID-19 incidence and first case dates compiled via DXY.cn and related tools. - Demographics: 2018 prefecture populations (China Statistical Yearbook) and geographic boundaries (Chinese Academy of Sciences). - Healthcare capacity: Number of Grade II and III hospitals per prefecture (National Health Commission), georeferenced via Amap API. Analyses: 1) Wuhan outflow clustering: Selected destination time series with average flow index >0.005 between 1–23 Jan 2020. Normalized outflow time series by total route flow in the pre-LNY window and clustered using k-means (k=4 determined by silhouette analysis). Alternative clustering methods and cluster numbers assessed in sensitivity analyses. 2) Case timing association: Tested association between destination cluster membership and timing of first detected case; adjusted for potential surveillance bias using linear regression with 2018 population size as proxy covariate (Detection date ~ β0 + β1·pop + β2·cluster). 3) Surge evaluation: Compared outbound flows in the 6-day window preceding LNY (2020: 18–23 Jan; 2019: 29 Jan–3 Feb) using two metrics: V1,i = mean(F_i,2020)/mean(F_i,2019) − 1 and V2,i = [mean(F_i,2020) − mean(F_i,2019)]/sd(F_i,2019); ranked Wuhan relative to other prefectures. 4) Hierarchical movement: Classified prefectures into population quartiles (Low, Medium-low, Medium-high, High). Quantified proportions of inbound and outbound flows between quartiles over time. 5) Healthcare pressure: Defined healthcare capacity HC_i as hospitals per 100,000 residents (primary) and as hospital counts (sensitivity). Computed weekly healthcare pressure HP_i,w = confirmed cases_i,w / HC_i. Compared distributions between low vs high capacity prefectures using one-tailed Mann–Whitney U tests weekly (weeks 3–9; n_low=157, n_high=153), with robustness checks using two-tailed and opposite-direction tests. 6) Network community structure: Built daily directed, weighted mobility networks and detected communities using the Leiden algorithm maximizing modularity Q on time-sliced networks (inter-slice weight 1e−5). Tracked total and sub-community modularity over time; highlighted communities centered on Wuhan, Beijing, Shanghai, Guangzhou & Shenzhen. Produced snapshots pre- and post-cordon. 7) Distance kernels: For major cities, computed daily frequency of journeys exceeding distance thresholds up to 4185 km for inflow and outflow (Supplementary). Limitations addressed: OD proportions only available for top-100 connections; no pre-Jan 1 pairwise baseline; data aggregated to day-level; Baidu user representativeness; LNY confounding; lack of traveler demographics.

- Wuhan outbound mobility patterns: Two peaks in early January 2020—an early-January peak (not seen in 2019) and a higher pre-LNY peak compared to 2019. Outgoing traffic from Wuhan decreased by 89% within two days of the cordon sanitaire (23 Jan 2020). - Clustering of Wuhan outflows identified four temporal patterns (Clusters A–D). Clusters A and B (geographically closer, smaller, lower density destinations) showed pre-LNY increases; Cluster C had two peaks (~7 and ~22 Jan); Cluster D had an early-January peak without a pre-LNY peak. Cluster membership was associated with timing of first case detection (p=0.0004), persisting after adjusting for population as surveillance proxy (p=0.00002). Cluster D included major cities (Beijing, Shanghai, Guangzhou, Shenzhen) that detected cases earlier and may have seeded further spread. - Epidemic timing: First detection outside Wuhan on 17 Jan 2020; most prefectures detected first cases between 23–26 Jan; by late March, >90% of prefectures/province-level cities had at least one case. - Surge analysis: While Wuhan showed increased outbound travel in the six days before LNY, similar increases occurred widely across China. Wuhan ranked 46th (top 13%) and 88th (top 24%) of 305 prefectures by the two surge metrics relative to 2019, suggesting the increase was consistent with nationwide pre-LNY travel rather than a unique exodus due to the impending cordon. - Hierarchical mobility: Across population quartiles, inbound and outbound volumes rose before LNY and dropped sharply afterward, with increased within-quartile flow after 23 Jan. Before LNY, inbound travel to all quartiles increasingly originated from high-population prefectures, especially into lower-population areas. Outbound flows from large prefectures concentrated toward medium-sized destinations, while medium/small prefectures exchanged more with similar-sized areas, consistent with migration step effects along the urban hierarchy. - Healthcare capacity and pressure: Pre-LNY movements shifted population away from high-capacity healthcare settings (more outbound than inbound in high-capacity areas). From the week before LNY through two weeks after, low-capacity prefectures experienced significantly higher COVID-19-related healthcare pressure (cases per hospital capacity) than high-capacity ones (one-tailed Mann–Whitney U tests; weekly comparisons with significant differences in most weeks; similar results using hospital counts without population adjustment). By 1 Mar 2020, the geographic distribution of residents had not fully returned to pre-LNY conditions. - Network structure: Modularity patterns indicated a stable community structure pre-LNY with substantial inter-community travel (low Q preceding LNY). Immediately after the cordon sanitaire, Wuhan’s community modularity contribution increased briefly (greater relative integration despite lower absolute flows), potentially reflecting resource inflows. Countrywide connectivity declined after 23 Jan, with no evidence of major rerouting or emergence of alternative hubs; the transport network’s community structure returned quickly to its prior configuration at lower flow levels, indicating no structural reorganization during the study period.

The findings disentangle holiday-driven mobility from outbreak-response movements during early COVID-19 in China. Pre-LNY surges were widespread across prefectures, suggesting Wuhan’s observed outflow increases were primarily seasonal rather than triggered by the cordon sanitaire announcement. Larger, distant cities detected cases earlier than some closer, smaller prefectures, indicating that functional connectivity (mobility flows and urban hierarchy) can be more predictive of epidemic spread than geographic proximity. The hierarchical divergence in flows implies medium-sized prefectures act as critical conduits toward smaller regions; targeting non-pharmaceutical interventions and surveillance in these intermediaries could help prevent spread into rural or less-resourced areas. Holiday mobility shifted population away from high-capacity healthcare settings, increasing COVID-19-related healthcare pressure in low-capacity prefectures during peak transmission—highlighting the need for temporary resource mobilization (personnel, equipment) to destinations experiencing atypical influxes from areas with higher prevalence. Despite drastic restrictions, the overall transportation network did not exhibit compensatory rerouting or lasting structural change; community structure remained stable at reduced volumes, informing expectations about network responses to short-term large-scale movement restrictions. These insights support planning of travel restrictions by considering mobility-defined connectivity rather than purely geographic proximity and by anticipating healthcare demand redistribution during mass travel periods.

This study provides a comprehensive assessment of human mobility in China during early COVID-19, distinguishing seasonal holiday travel from responses to Wuhan’s cordon sanitaire, and linking mobility patterns to healthcare pressure and national transport network structure. Key contributions include: identifying temporal and hierarchical mobility patterns from Wuhan and across population quartiles, demonstrating limited association between proximity and epidemic spread relative to mobility connectivity, evidencing increased healthcare pressure in low-capacity regions due to holiday movements, and showing the absence of structural reorganization in the transport network under short-term restrictions. Future work should incorporate finer-grained mobility with traveler demographics (age, occupation) to better estimate transmission risk and healthcare demand, and extend analyses to longer timeframes to assess long-term impacts of restrictions on mobility networks and to contexts without overlapping mass travel periods to isolate causal effects of policy announcements.

- Mobility data represent Baidu users, a potentially non-representative subset of the population. - OD proportions are available only for the top-100 connections per prefecture, limiting validation and coverage of lower-ranked flows. - Movement data aggregated to daily level and prefecture level; trips spanning more than one day cannot be analyzed, and pre–1 January 2020 pairwise baselines are unavailable. - LNY coincided with the outbreak’s initial phase, confounding separation of holiday travel from epidemic- or policy-driven movements. - Use of 2018 residential populations to normalize healthcare capacity may not reflect true transient populations during LNY. - Lack of traveler characteristics (e.g., age, occupation) limits understanding of transmission risk and healthcare demand implications. - Surveillance bias adjustment relied on population size as a proxy due to limited direct indicators of testing capacity.