Modeling transit deserts in real-time

Transit desert studies are largely descriptive and static, identifying zones areas where transit demand exceeds supply at a single point in time, without exploring underlying causes. This study of Seoul, South Korea used advanced machine learning techniques including a random forest model to forecast the presence of transit deserts based on key factors including density, distance to transit, diversity in the built environment, actual population movements over time, and sociodemographic characteristics. Paired with an interactive dashboard, the model updates transportation equity and accessibility analysis results as users adjust the threshold values for their neighborhoods.

As user-friendly tools for interacting with complex models for classifying vulnerable urban zones become more widespread, they will transform how AI models are used in planning. In adaptation planning, this will mean models become equally important for informing dialogue and participatory processes in an ongoing way, rather than simply producing supporting experts to produce static analyses. Additionally, access to such tools for city residents can help enhance the understanding, acceptance, and impact of AI models.