Spatiotemporal Graph Neural Networks and Digital Twins for Urban Drainage Networks: Predictive Inundation, Critical-Node Detection, And Infrastructure Prioritization

Abstract

This study addresses the growing challenge of urban drainage networks operating under intense rainfall, spatially connected hydraulic failures, aging infrastructure, sparse monitoring, and limited rehabilitation resources, while conventional approaches often remain fragmented and reactive. The purpose was to develop and quantitatively evaluate an integrated framework combining Spatiotemporal Graph Neural Network Predictive Inundation Intelligence (SPII), Digital-Twin Synchronization and Simulation Capability (DTSC), Critical-Node Detection Capability (CNDC), and Risk-Informed Infrastructure Prioritization (RIIP) to improve Urban Drainage Resilience and Decision-Support Effectiveness (UDRDE). A quantitative, cross-sectional, explanatory case-based design was applied within a bounded urban drainage network using purposive expert sampling, a five-point Likert questionnaire, and graph, hydraulic, and inundation-related computational data. Of 320 questionnaires distributed, 302 were returned and 290 valid responses were retained, yielding a 90.6% valid response rate. Descriptive statistics, reliability and validity testing, Pearson correlation, multiple regression, and STGNN performance assessment were employed. UDRDE recorded the highest mean (M = 4.20, SD = 0.51), followed by DTSC (M = 4.18, SD = 0.53), SPII (M = 4.16, SD = 0.55), CNDC (M = 4.13, SD = 0.57), and RIIP (M = 4.11, SD = 0.59). All predictors correlated positively with UDRDE, including DTSC (r = .75), SPII (r = .72), RIIP (r = .71), and CNDC (r = .69), all p < .001. The regression model explained 71.4% of UDRDE variance (R² = .714, adjusted R² = .710, F (4,285) = 177.87, p < .001), with DTSC the strongest predictor (β = .31), followed by SPII (β = .28), RIIP (β = .24), and CNDC (β = .21). The STGNN achieved R² = .91, RMSE = 0.118 m, MAE = 0.081 m, and F1 = .88. The findings imply that integrating graph-based prediction, synchronized digital twins, node criticality, and risk-informed prioritization can strengthen flood preparedness, targeted maintenance, infrastructure investment, and urban drainage resilience.