Assimilating flow and level data into an urban drainage surrogate model for forecasting flows and overflows

Area of Science:

• Urban hydrology and water resource management
• Computational hydraulics and modeling
• Real-time control systems for infrastructure

Background:

• Effective real-time control and warning systems for urban drainage require accurate flow and overflow forecasting.
• Detailed 1D hydrodynamic models are often computationally infeasible for real-time applications, necessitating the use of surrogate models.
• Historical time series data, common in rural hydrology, are typically unavailable for dynamic urban drainage systems.

Purpose of the Study:

• To develop and validate a fast, reservoir-based surrogate forecast model for urban drainage systems.
• To investigate the effectiveness of updating the surrogate model using observational data for improved forecast accuracy.
• To assess the impact of different data assimilation strategies (water level vs. flow) on forecast performance.

Main Methods:

• Construction of a reservoir-based surrogate forecast model from a 1D hydrodynamic urban drainage model.
• Implementation of the Ensemble Kalman filter for updating the surrogate model's internal states with observational data.
• Assimilation of water level or flow observations, directly or indirectly via rating curves, into the model.

Main Results:

• Model updating significantly improved flow and overflow forecasts up to 2 hours in advance.
• Assimilation of water level observations yielded superior flow forecasts compared to using flow data.
• Water level-based updating demonstrated robustness, being insensitive to noise formulation, making it suitable for operational use.

Conclusions:

• A reservoir-based surrogate model, updated via the Ensemble Kalman filter, provides a viable alternative to traditional hydrodynamic models for urban drainage forecasting.
• Water level data assimilation is a more effective strategy for improving flow forecasts in urban drainage systems than flow data assimilation.
• The proposed model updating method is robust and suitable for operational real-time applications in urban drainage management.