Extracting socioeconomic information with the help of urban structure types indata-scarce environments

Linking socio-economic and spatial structural data is a key research gap in the assessment of urban climate risk. Urban populations are increasingly exposed to adverse effects of climate change, particularly in coastal urban areas in the Global South, where information is scarce. Researchers often rely on survey results, which deliver highly detailed micro-level point data but those are costly and lack spatial coverage. Conversely, Earth observation
(EO) data offers greater spatial coverage but provides limited insights into socio-economic factors. Hence, there is a gap in integrating urban morphology data derived from EO and micro-scale survey data. We aim to address this gap by linking socio-economic and morphological data using a novel machine learning-based approach and testing whether and to what extent urban structure types (UST) can function as spatial predictors for
socio-economic profiles.
To do so, we implement a two-stage process based on k-means and random forest (RF) algorithms in two case studies. First, we employ a k-means clustering algorithm to delineate socio-economic profiles based on household survey variables relating to education, household size and composition, and asset ownership. We then used an RF classification algorithm that used USTs as predictors to extend the survey socio-economic profiles into
morphologically similar areas in two case study cities: Mumbai (India) and Ho Chi Minh City (Vietnam). In Mumbai, current socio-economic data is severely outdated (i.e., the last census was in 2011), while in Ho Chi Minh City it is only available in coarse spatial units (i.e., at the commune level). These conditions hinder research on assessing flooding vulnerability, as population growth and the mismatch between administrative units and flood hotspots introduce severe bias and uncertainty in the available data. To overcome this situation, we implemented household surveys (n=1240 and 751, respectively) in flood hotspots that cover a variety of urban structure types (e.g., compact low-rise, open mid-rise, or lightweight low-rise).
In our study, morphological information functions as a proxy for socio-economic profiles, thus providing a cost-effective and spatially explicit novel approach to assessing social vulnerability in data-scarce conditions. By starting from the household-level survey data, we avoid the most critical problems from reductionist approaches (e.g., social determinism) and recognize the limitations of data triangulation. To this end, we measure the uncertainty of the association in each step and validate the assumptions and results with local stakeholders. Our novel approach aims to use the available EO data to improve the identification of high-vulnerability areas. Albeit experimental, the spatially explicit identification provided in
this study provides crucial insights for targeted climate adaptation policy and research. By studying two rapidly evolving coastal cities, we provide a comprehensive and reproducible method to assess the challenges urban populations face under climate change.