The University of Sheffield
F84.pdf (412.23 kB)

CCWI2017: F84 'Prediction of CSO Chamber Water Levels Using Rainfall Forecasts'

Download (412.23 kB)
journal contribution
posted on 2017-09-01, 15:01 authored by T. Rosin, M. Romano, K. Woodward, E. Keedwell, Z. Kapelan
CSOs are a major source of pollution, spilling untreated wastewater directly into water bodies and/or the environment. This paper describes an investigation of the use of rainfall forecasts when predicting water level in a Combined Sewer Overflow (CSO) chamber. A prediction model based on a multi-layer feed-forward Artificial Neural Network (ANN) with a sliding time-window of inputs was developed to model the relationship between contributing rainfall and water level within a CSO chamber and predict the occurrence of spills. Three versions of the aforementioned prediction model were developed and compared: 1) without using forecast rainfall data, 2) with perfect rainfall forecasts (i.e. forecasts assuming perfect knowledge of historical rainfall into the future) and 3) with real rainfall forecasts. These models were tested on a CSO located in the Wirral area of the United Utilities network and were all able to successfully predict CSO spills 2 hours ahead. Use of forecast radar rainfall data, generated by the Met Office, was shown to demonstrably improve the accuracy of the ANN predictions and increase the prediction range of the model. Unsurprisingly, this accuracy was further improved when using perfect rainfall forecast data, due to errors inherent in real forecasts. It is envisioned that ANN models could be beneficially used by water utilities for near real-time modelling of water level in CSO chambers and generate alerts for upcoming spills events. The use of forecast data increases the detection time of CSO discharges, potentially enabling water utilities to take appropriate and timely action to address potential issues before the customers and the environment are adversely effected.


Usage metrics

    Department of Civil and Structural Engineering



    Ref. manager