???ViewItemPage???

???ViewItemOverview_lblAbstractSection???

Understanding how the built environment influences travel is key to low-carbon urban planning. However, previous cross-sectional studies lack a realistic operationalization of residential self-selection that accounts for its non-linear nature, limiting its applicability to urban planning. We propose a double machine learning (DML) approach that accounts for nonlinearities in residential self-selection and captures non-linear moderating effects. Using travel diaries of 32,201 Berlin residents, we estimate the built environment’s impact on per capita travel-related CO2 emissions. Our results indicate that neglecting nonlinearities overestimates this impact by 13%–18%, inflating the built environment proportion by 13%pt. Age, income, and car ownership also nonlinearly moderate the built environment’s effect, with the effect being largest for middle-aged, high-income, car-owning households, a novel finding. Applying the method to urban planning reveals a 43%pt emissions reduction potential for 64,000 planned Berlin housing units, highlighting the need for evidence-based urban planning to effectively mitigate CO2 emissions in cities.