Identifying multi-dimensional vulnerability profiles in flood-prone urban environments

Assessing climate and, more specifically, flood vulnerability in rapidly urbanizing regions remains a challenge due to the complexity of diverse socio-economic, demographic, and spatial factors. This case study of Mumbai integrates household-level survey data (n = 1106) with morphological information to capture the multi-dimensional nature of vulnerability at the intra-urban scale. Focusing on flood-prone neighborhoods in Mumbai, we analyze household survey data (e.g., education, employment, income security, household assets) to identify distinct ‘archetypes’ of vulnerability.  We implement an advanced, unsupervised machine learning approach to generate distinct and heterogenous socio-economic profiles by grouping households across multiple variables (e.g., education, employment status, household assets) rather than relying on static thresholds. We further incorporate statistical association measures to robustly examine relationships between clusters and key vulnerability outcomes and indicators (e.g., perceived flood severity, loss of workdays, and health impacts). To examine the influence of urban development on flood-related hazards, we complement the socio-economic clustering with a geospatial analysis that connects local urbanization conditions to the identified vulnerability profiles. First, we analyze household-reported impacts from flooding and perceived causes (e.g., blocked drainage channels, lack of maintenance) for each cluster to understand specific pathways by which urbanization exacerbates or alleviates flood risk. Second, we integrate these survey-based findings with geospatial data of topography (e.g., household location in the watershed) and urban form (e.g., open, or compact types) to assess the extent to which household location and built form shape or modify local flood vulnerability.Our findings provide a data-driven baseline for capturing vulnerability that goes beyond singular proxies such as income. However, low data availability and quality—particularly in Global South contexts—can limit the replicability of this approach, and the high socio-spatial diversity within cities like Mumbai may not always be captured by coarser spatial data. Moreover, it remains unclear how well these findings hold over time, as vulnerability patterns may shift rapidly in evolving urban areas. Despite these caveats, by simultaneously assessing a range of household-level and urban form variables, this approach produces vulnerability profiles that can inform spatial prediction models and serve as inputs for spatial simulations of urbanization. The resulting flood vulnerability maps help to identify areas in need of interventions and offer a reproducible template for other flood-prone settings in the Global South.