A Bayesian network-based framework to uncover social inequities in flood risk: an application to East Baton Rouge Parish, Louisiana

Abstract Regional flood risk assessment requires understanding the complex interplay of sociodemographic, socioeconomic, hydrogeologic, built environment, and climatic variables that impact flood vulnerability. Traditional studies often focus on large-scale units, overlooking localized variations and disproportionate exposure within populations. To address these gaps, a methodological framework of flood risk assessment using Bayesian networks was proposed and applied to East Baton Rouge Parish, Louisiana, for the August 2016 flood event to identify relationships between flood risk variables at the housing unit level. Flood exposure representativeness (FER) was determined using the Bayesian network’s inference capabilities to identify over-represented or under-represented groups. This study considered sociodemographic and socioeconomic (race, ownership, income group), hydrogeologic (land cover, waterbody proximity, ground elevation, water depth), built environment (building foundation height), and climatic (cumulative rainfall) variables for flood risk assessment. A probabilistic population allocation model was used to assign socioeconomic and sociodemographic data at the housing unit level, while geospatial methods were used to obtain other variables. Datasets were divided into training and testing sets, and the maximum likelihood algorithm was applied to train the Bayesian network. The testing dataset’s accuracy ranged from 0.76 to 0.79. Results highlight disproportionate flood risks, particularly for high-income Black or African American populations. This work advances flood risk assessment by integrating high-resolution and multivariable data, offering critical insights into the impacts on vulnerable populations.