Search for a command to run...
The resilience of urban drainage systems is increasingly threatened by rapid urbanization and shifting precipitation patterns, making accurate real-time monitoring a foundational requirement for intelligent control and pollution mitigation. However, operational sewer monitoring is affaected by harsh environments that cause intermittent sensor failures, data dropouts, and significant measurement uncertainties. While physics-based hydraulic models are established tools for planning, their high computational cost and sensitivity to uncertain parameters, specifically hydraulic roughness, hinder their utility for real-time state estimation.This study presents a robust, physics-informed Data Assimilation (DA) framework designed to generate continuous, reliable discharge estimates by fusing dual-sensor measurements (level and velocity) with a hydrologic model, that integrates real-time rainfall forcing via a nonlinear routing model. This approach allows the Particle Filter (PF) to anticipate flow responses to precipitation events while utilizing observational data to correct model trajectories.The methodology employs a Particle Filter to perform joint state-parameter estimation, simultaneously retrieving the discharge (Q) and the effective Manning’s roughness coefficient (n). By treating roughness as a time-varying parameter, the framework automatically compensates for slow-varying systematic changes such as biofilm growth or sedimentation. To ensure hydraulic consistency, the observation operator is strictly constrained by the Manning equation, enforcing a physical relationship between the estimated flow, depth, and velocity.A critical innovation of this work is the implementation of a robust sensor fault model. We utilize a mixture likelihood function combining heavy-tailed Student’s t distributions with uniform outlier distributions to dynamically detect and down-weight faulty sensor readings. This is augmented by adaptive noise covariance tuning and gating logic that identifies physically impossible transitions, such as sudden velocity dropouts or depth saturation.The framework is demonstrated on a sanitary sewer reach in Beaverton, Oregon. Results indicate that coupling rainfall-driven routing with dual-sensor assimilation yields flow estimates that are robust to individual sensor failures. The system successfully maintains valid flow hydrographs even during velocity sensor dropouts by leveraging the rainfall-stage relationship and the learned hydraulic roughness. This approach represents a significant advancement in generating high-quality, physics-consistent data for digital twins and real-time control of urban water infrastructure.