Inland communities face escalating flood risks as residential development continues within regulatory floodplains, Douglas, and Sarpy Counties in eastern Nebraska, an area heavily impacted by the 2019 Midwest floods. We develop a geospatial artificial intelligence (Geo-AI) framework integrating bi-temporal National Agricultural Imagery Program (NAIP) imagery, parcel-level assessment data, building footprints, and flood hazard maps, and apply a Siamese U-Net deep learning model to detect new, demolished, and existing residential structures. Results show a substantial increase in floodplain exposure, with over 2,250 new residential buildings constructed more than 21 times the number of demolitions primarily in high-assessed-value areas, while removals are concentrated in disadvantaged neighborhoods. The model achieved 98% overall accuracy, demonstrating strong capability for detecting building-scale change. These findings highlight growing inequities in flood-risk exposure and demonstrate the value of Geo-AI for spatially explicit flood-risk monitoring, risk-informed planning, and equitable land-use policy development.

Abstract Walking is a fundamental mode of urban mobility that promotes physical and mental well-being, mitigates environmental issues, and enhances community livability. As research increasingly emphasizes not only the physical but also the perceptual dimensions of walkability, understanding how pedestrians cognitively evaluate their surroundings has become a critical challenge. This study proposes a GeoAI-based framework for modeling and explaining perceived walkability in urban environments using street view imagery (SVI) and explainable AI techniques. A perceptual dataset derived from 171,118 pairwise comparisons was used to train the RSS-Swin model, enabling continuous estimation of walkability scores that reflect nuanced human judgments. Urban visual features were extracted from SVIs through semantic segmentation with SegFormer-B5, incorporating both object-level components and image brightness (HSV_V). To enhance interpretability, AutoML-based optimization and SHAP analysis were employed to identify and explain the key urban form features influencing perceived walkability. The proposed framework not only achieves high predictive accuracy but also provides transparent insights into the visual and environmental factors shaping pedestrian perception, offering practical implications for data-informed urban design and policy.