The COVID-19 pandemic caused devastating losses globally, highlighting the need for insights into managing future large-scale infectious diseases. Urban areas, with their high population density, are particularly vulnerable. Existing research mainly focuses on the impact of government-mandated stay-at-home orders on mobility. However, understanding spontaneous changes in citizen mobility behavior in the absence of such restrictions is crucial for effective policy design. This study investigates these spontaneous changes in Shenzhen, China, a megacity with over 17 million people, following the abrupt end of the "Zero-COVID" policy in December 2022. This policy, while successful in containing earlier outbreaks, led to a situation where a large portion of the population had not been exposed to the virus and thus had limited immunity. The sudden removal of restrictions, coupled with the highly transmissible Omicron variant, provided a unique environment to study the unconstrained reaction of citizens to a widespread outbreak. The researchers aim to understand the temporal and spatial evolution of urban mobility in response to the pandemic's lifecycle, analyze the underlying mobility behavior across different regions and transportation modes, and develop a dynamic model capable of predicting city-wide mobility changes based on observed behavioral patterns.

The introduction cites several studies highlighting the economic and human cost of the COVID-19 pandemic and the importance of research into pandemic response. It also mentions existing research focusing on the effects of government-imposed mobility restrictions, contrasting this with the limited understanding of spontaneous mobility changes. Studies on the impact of the pandemic on urban mobility, economic dynamics, and mental health are referenced. The authors emphasize the unique opportunity presented by China's sudden shift from its "Zero-COVID" policy to study these spontaneous changes without government intervention.

The study uses origin-destination (OD) mobility data from Shenzhen's public bus, subway, and taxi systems, encompassing approximately 148 million trips. The OD data provides information on the number of individuals traveling between specific locations within the city's public transport network. Data from before and after the "Zero-COVID" policy change are analyzed to understand spontaneous mobility shifts. The data are pre-processed using Kalman filtering to smooth out fluctuations. The data are then normalized to compare mobility changes relative to pre-pandemic levels. A K-means++ clustering algorithm is applied to identify distinct mobility patterns based on the temporal characteristics of trip changes. These patterns are then linked to different travel purposes and transportation modes by integrating the data with urban land use (ULU) maps. A dynamic model, based on a Susceptible-Infectious-Recovered (SIR) model, is developed to simulate the impact of the pandemic on mobility behavior. This model considers both the physiological impact of infection and the emotional willingness to travel. The model incorporates willingness factors that vary depending on the type of travel purpose, as inferred from the ULU data. The model's parameters are estimated using maximum likelihood estimation (MLE) and validated against the observed data.

The analysis reveals significant spatial discrepancies in mobility changes, with central business districts (CBDs) experiencing a more substantial impact than peripheral areas. Four distinct mobility patterns are identified through clustering: * **Cluster 1:** Shows the most significant decline in mobility, with a slow recovery, primarily associated with trips to schools and colleges. This is attributed to school closures and increased risk aversion. * **Cluster 2:** Exhibits a U-shaped pattern, with a decline followed by a gradual recovery, reflecting the general epidemiological pattern of infection and recovery. This cluster represents the majority of mobility behavior. * **Cluster 3:** Shows a U-shaped pattern with a faster recovery rate than Cluster 2, suggesting a quick return to pre-pandemic mobility levels for certain travel purposes. * **Cluster 4:** Shows minimal impact, primarily representing travel to essential locations like transport hubs, hospitals and workplaces whose mobility patterns remain relatively stable. The study also finds that different transportation modes are affected differently: subways experience the most significant decline in ridership, followed by buses and then taxis. The dynamic model accurately captures the observed mobility changes across different clusters and transportation modes. The model parameters, particularly the willingness factors, highlight the heterogeneity in responses to the pandemic across different travel purposes. The model shows that the inclusion of willingness factors (psychological impact) is essential to accurately capture the observed behavior.

The findings demonstrate the complex interplay between individual behavior, public health, and urban mobility during a pandemic. The spatial variations in mobility highlight the importance of considering local contexts when designing public health strategies. The four distinct mobility patterns provide insights into different populations' responses and recovery trajectories. The significant impact on subways and the relatively less severe impact on taxis underscore the role of perceived risk in mode choice. The success of the dynamic model in capturing fine-grained mobility changes using a relatively simple model structure demonstrates its potential for forecasting the impact of future outbreaks. The model's ability to incorporate both physiological (infection) and psychological (willingness) factors demonstrates the importance of understanding both the physical and behavioral aspects of responses to a pandemic.

This study provides a novel framework and dynamic model for understanding spontaneous changes in urban mobility during a large-scale infectious disease outbreak. The findings emphasize the heterogeneity of mobility patterns and the importance of considering both physiological and psychological factors in forecasting pandemic impacts. Future research could focus on extending the model to incorporate additional factors, such as age-specific risk and other forms of transportation. Integrating this type of model with emerging urban sensing technologies and large language models offers great potential for enhancing the accuracy and applicability of urban mobility simulations for pandemic response planning.

The study is limited to Shenzhen, China, and the specific data sources used (public transport data). The model may not perfectly capture the complexities of individual decision-making or the full spectrum of urban mobility, as private vehicle use is not accounted for. The urban land use data used were from 2018, while the travel data used spanned 2022-2023. Therefore, the mapping between the two may not be perfectly accurate and may introduce inaccuracies into the interpretation of the findings. The willingness factors are a simplification of complex human behavior, and the model could be improved by incorporating more detailed behavioral data.