Navigating Urban Exposome Futures: Advances in spatial agent-based scenario modeling for environmental health policy insights

Abstract

Cities pose serious health risks through high levels of air pollution, noise and limited green spaces, contributing to health inequalities across neighborhoods and social groups. However, cities also offer vast opportunities for sustainable, health-focused transformations. Understanding how urban environments can improve health requires looking at people's behaviors and their interaction with the surroundings in space and time. This thesis proposes a framework for using a computer simulation method called Spatial Agent-Based Modeling (ABM) to study how different hypothetical urban policies would impact people's health. ABM creates "virtual people" who follow their daily schedules, move around the city, interact with each other and their environment in a geographic context. Unlike traditional methods that estimate exposure at people’s home addresses, ABM can model personal exposure of people along activity and mobility patterns. Moreover, the method improves on previous scenario analysis methods by capturing both unintended and intended intervention effects across time, space and population groups. This research addresses important technical challenges of applying the method to this use case. To generate synthetic urban populations, we developed an adaptive and iterative method and software package that combines the most detailed aggregated datasets, capturing multivariable correlations (e.g., between age, sex, income) and fine-grained spatial distributions (e.g. street-block level). For behavior model validation, we created a method using natural language processing and deep learning to automatically extract evidence on behavioral determinants for different population groups. These insights are organized into a knowledge graph that informs and validates variable selection for predictive models. Applied to transport choice modeling, this approach improved performance by incorporating often-overlooked but impactful behavior determinants. To simulate air pollution, we built a novel hybrid model combining land-use regression and cellular automata, calibrated with genetic algorithms. This model, designed for integration with ABM and traffic scenarios, enables fast, spatially and temporally detailed air quality projections based on traffic changes. We validated it using 2020 Amsterdam Covid-19 lockdown traffic data, showing strong agreement between projected and measured pollution levels. The different methodological innovations were combined in a novel ABM framework, which simulates the urban population, their daily activities, travel behavior, resulting air pollution, exposure to air pollution, transport physical activity and health impacts. We implemented and validated this model for Amsterdam to test various transport-related interventions. Simulating the past parking price increase (2019) and comparing results to transport data, we were able to show that our model projects the population transport behavior changes correctly. Our scenario analysis shows that no-emission zones drastically improve health by cutting air pollution and boosting transport physical activity, but also lead to longer travel times, especially for lower-income residents living in the urban outskirts. Surprisingly, the "15-minute city" concept (where all essentials are within walking distance) might reduce pollution but also decrease physical activity in bike-friendly cities like Amsterdam. This research shows ABM can be empirically grounded, validated and predict both positive and negative effects of urban policies, offering insights that could help cities create healthier and fairer environments.