An ensemble data-driven approach for incorporating uncertainty in the forecasting of stormwater sewer surcharge

ABSTRACT

Stormwater surcharge events are unavoidable above a certain rainfall intensity. Thus, for protection and damage mitigation, forecast systems are of outstanding importance. This study develops an ensemble forecast system (EFS) to predict the beginning and end of sewer surcharge events. It applies a nonlinear autoregressive with exogenous inputs (NARX) network to each member of the ensemble, making it suitable for real-time predictions. The fundamental idea is the forecast of water depth time series within manholes based on the given rainfall. The novelty lies in the consideration of uncertainty through the ensemble structure for which the numbers of neurons in the hidden layer, the weights, and biases are considered to be uncertain. The results are evaluated based on observed values captured within the uncertainty band $(P_{factor}$), and the width of the band $(R_{factor}$). The $P_{factor}$ varied between 74% and 94% and the $R_{factor}$between 1.36 and 10.68.

KEYWORDS:

• Urban drainage systems
• surcharge
• ensemble flood forecasting
• Nonlinear Autoregressive Network with exogenous inputs (NARX)
• neural networks

Acknowledgements

The research presented in this paper has been carried out as part of the HiOS project (Hinweiskarte Oberflächenabfluss und Sturzflut) funded by the Bayerisches Staatsministerium für Umwelt und Verbraucherschutz (StMUV) [Bavarian State Ministry of the Environment and Consumer Protection] (69-0270-92086/2017) and supervised by the Bayerisches Landesamt für Umwelt (LfU) [Bavarian Environment Agency].

Disclosure statement

No potential conflict of interest was reported by the author(s).