The COVID-19 pandemic initially showed positive short-term impacts on urban sustainability, but medium-term effects are less clear. Lockdowns caused mobility shifts, particularly away from public transport, potentially negating environmental benefits. Controlling human mobility is a key non-pharmaceutical intervention. While the Japanese government implemented emergency declarations and mobility reduction requests, the effectiveness varied across time and mobility types. This study addresses the question of which specific mobility types should be controlled to minimize COVID-19 cases, focusing on human mobility as a governable factor that significantly influences transmission, especially considering challenges in vaccine accessibility and rapid vaccine development for novel diseases. Previous interventions like school and business closures have been effective; however, this study examines places not previously considered, aiming to balance infection control with maintaining socio-economic activity. The study analyzes six mobility categories (retail/recreation, groceries/pharmacies, parks, transit stations, workplaces, residential areas) from Google Community Mobility Reports data in Osaka, Kyoto, and Hyogo prefectures from March 1, 2020, to September 30, 2021, using random forest analysis to determine the relationship between mobility changes and subsequent COVID-19 cases.

Existing research highlights the correlation between human mobility and COVID-19 transmission. Studies using mobile phone data have identified high-risk locations such as restaurants and shops. The importance of transit station mobility in policymaking has also been noted. Previous analyses using Google Community Mobility Reports data found that retail/recreation, groceries/pharmacies, and transit stations were important mobility areas in early pandemic periods. Other studies in different regions highlighted the impact of groceries/pharmacies mobility and the decrease in workplace and retail recreation mobility. However, this study addresses a gap by focusing on the medium-term pandemic, acknowledging the potential for non-linear relationships and the complexity of factors influencing transmission during this period. The use of random forest analysis is chosen to mitigate overfitting, common in other machine learning methods when dealing with multiple waves of infection increase and decrease.

This study utilized Google Community Mobility Reports data, publicly available data that tracks movement trends across six categories: retail/recreation, groceries/pharmacies, parks, transit stations, workplaces, and residential areas. The data shows changes in visits to these places relative to a baseline (median from January 3 to February 6, 2020). Data from Osaka, Kyoto, and Hyogo prefectures were analyzed. Daily COVID-19 case data from the Japanese Ministry of Health, Labour and Welfare were also used. The Research Ethics Committee of the Graduate School of Life Science, Osaka City University approved the ethical protocol, adhering to guidelines for using de-identified location data. The study employed random forest analysis to investigate the relationship between daily mobility data (predictor variables) and the total number of COVID-19 cases after two weeks (response variable). A two-week lag was used due to the incubation period of SARS-CoV-2. The random forest model, with 10,000 trees, was used to assess the relationship because it prevents overfitting in comparison with other machine learning models, such as logistic growth model, partial differential equations, and neural networks. The analysis was conducted at the prefecture level, considering the Google user population as a fraction of each prefecture's population (79.2% usage in Japan). The variable importance was assessed using dependent resampled inputs.

Analysis of human mobility patterns (Figure 2) showed decreases in all mobility types except residential areas after March 2020, reflecting increased stay-at-home behavior. The number of COVID-19 cases (Figure 3) exhibited five waves of increase and decrease. Random forest analysis revealed significant relationships between mobility changes and COVID-19 cases (R² > 0.7 for all prefectures). In Osaka Prefecture, the most significant impacts were observed for groceries/pharmacies (total effect = 0.437), parks (total effect = 0.368), workplaces (total effect = 0.253), and retail/recreation (total effect = 0.211). Importantly, transit stations did not show a significant impact on the number of infections. Results (Table 1 and Figure 4) indicate that controlling human mobility in groceries/pharmacies to within -5% and 5% and in parks to more than -20% is essential to reduce COVID-19 cases. The findings regarding transit stations highlight a potential opportunity for urban sustainability by encouraging the return to public transport with implemented hygiene measures.

The findings address the research question by identifying specific mobility types that significantly influence COVID-19 transmission during the medium-term pandemic. The results underscore the importance of targeted interventions. The lack of significant impact from transit station mobility is a notable finding, suggesting that promoting the use of public transportation, alongside hygiene protocols, can contribute positively to urban sustainability by reducing reliance on private vehicles and their associated carbon emissions. The study's findings offer valuable insights for policymakers, allowing for more effective allocation of resources and more targeted interventions to minimize infection spread while maintaining socio-economic functions. This study highlights the need for considering prefecture-specific characteristics when designing and implementing public health policies.

This study demonstrated the effectiveness of targeted control of human mobility in reducing COVID-19 cases. Specifically, controlling mobility related to groceries/pharmacies and parks proved crucial. The insignificant role of transit stations suggests opportunities for promoting public transport while maintaining infection control. Future research could investigate the correlation between human mobility data and more granular, matched data on infection sources. This would provide deeper insights into the dynamics of COVID-19 transmission and inform more targeted and efficient public health policies.

This study uses Google Community Mobility Reports data, which may not represent the entire population due to variations in smartphone ownership and Google app usage. The analysis focuses on the Osaka metropolitan area, and the findings might not be directly generalizable to other regions with different population densities, cultural norms, and policy responses. Further research is needed to explore the influence of other factors, such as weather, variant prevalence, and population density.