Modelling combined sewer overflow based on sewer network graph representation and rain radar data: application to the Seine river

Rivers are at the heart of human activity. They provide many ecosystem services: drinking water, agriculture, transport, hydropower, bathing, freshness, etc. They are also hotspots for biodiversity. However, the water quality of these rivers is deteriorated as a result of human activity. The current work focuses on fecal contamination, which is a discriminating criterion for bathing.

In urban watersheds, fecal bacteria contamination comes from point sources related to the operation of the drainage network. During rainy weather, the combined sewer network, mixing both wastewater and stormwater, can become saturated. As a consequence, part of the flow is discharged directly into the river via combined sewer overflows (CSOs). This is the case for the city of Paris. The possible CSO overflow can be modeled by a function linking its discharge to precipitation. This relationship is currently poorly understood, with little related work, and even less for the Seine river.

To build such linking function, we rely on a dataset that includes location and hourly discharged volume of the monitored CSOs in the Seine River within Paris. Urban watersheds have been delineated within the study site. Rainfall height over these watersheds have been obtained from weather radar. We broke down the data timeseries into events. An event begins with the cause, the rain, and ends with the consequence, the overflow. To link rainfall to CSOs a directional graph based on the drainage network map, was created. It represents the wastewater transport from one watershed to another. This highlights which rainfall variables to consider regarding the CSO location. Principal component analysis (PCA) is used to assess for rain characteristics selection. An unsupervised non-linear technique (Isomap) is then used to build linking function structure.

The overflow volume in time can be modeled by a triangular shape. This shape is described by the overflow initial time, its total and maximum volume and the time of the maximum. We expect to retrieve these overflow variables by reducing the number of rainfall event characteristics to single indicators using sequentially PCA and Isomap.

Modeling and forecasting source discharges would enable better management of bathing and water supply risks, and better evaluation of mitigation infrastructures.