Satellite-based condition assessment of urban green stormwater infrastructure using a hybrid machine learning framework

Abstract Effective management of urban stormwater infrastructure is essential for reducing runoff impacts, improving water quality, and enhancing flood resilience. However, conventional field inspections of distributed green stormwater infrastructure (GSI) are resource-intensive and difficult to scale across large urban areas. Although remote sensing offers opportunities for scalable infrastructure monitoring, existing studies have primarily focused on vegetation characteristics rather than comprehensive assessments of GSI functionality. This study uses a hybrid machine learning framework integrating high-resolution multispectral satellite imagery, unsupervised anomaly detection, and supervised classification to evaluate multiple GSI condition indicators, including overall performance, vegetation health, vegetation stability, erosion, and debris presence. The framework incorporates spectral bands, remote sensing indices, and environmental variables to classify compliant and non-compliant conditions. Results demonstrate strong performance for vegetation stability (accuracy = 0.93; F1-score = 0.93), followed by debris presence (accuracy = 0.78; F1-score = 0.77), vegetation health (accuracy = 0.72; F1-score = 0.77), and erosion (accuracy = 0.70; F1-score = 0.73). Overall GSI performance was the most challenging target (accuracy = 0.61; F1-score = 0.59), with lower performance for non-compliant conditions. The proposed approach provides a scalable tool for GSI screening and maintenance prioritization while highlighting the need for improved detection of degraded conditions.