Changes in visitor behaviour across COVID-19 pandemic: Unveiling urban visitation dynamics and non-linear relationships with the built environment using mobile big data

COVID-19 has significantly affected the behavioural patterns of urban visitors. However, the non-linear relationships between visitor behaviour and built environments, particularly how these relationships have evolved during the pandemic, have not yet been extensively studied. Using over 10 million mobile big data records collected over three years in Fukuoka, Japan, incorporating the XGBoost machine learning model and SHAP-PDP interpretation method, we identified non-linear relationships between visitor behaviours and built environments. Our findings uncovered significant non-linear impacts on visitor behaviour of several urban characteristics, such as floor area ratio, building coverage ratio, road density, and POI richness. Notably, the floor area ratio exhibits a negative correlation with visit frequency below 1 but a positive correlation above 2. Building coverage ratio positively impacts visit frequency up to 1000 m 2 per 10,000 m 2 , after which it turns negative; this correlation shifted towards a consistent negative trend during the pandemic. Road density, which usually correlates negatively with visit duration, became positively correlated above 700 m 2 per 10,000 m 2 during the pandemic. Similarly, the influence of floor area ratio on visit duration reversed from negative to positive under pandemic conditions. Revealing the dynamic and non-monotonic nature of how urban visitors respond to the built environment under the influence of the pandemic, our results provide valuable insights for urban planning strategies in response to crisis resiliently. • Using mobile big data to identify urban visitor destinations and finely quantify visitor behavior through the case of Fukuoka City, Japan. • During the COVID-19 pandemic, the number of urban visitors and their visit range significantly decreased, but the average visit duration increased. • Machine learning model revealed a strong nonlinear relationship, which changed significantly during the COVID-19 pandemic, between urban land use, buildings, roads, with visitor behaviour. • Formulating urban planning and land use policies targeting urban visitors is crucial for enhancing city vitality and resilience.