Towards an uncertainty propagation framework in urban drainage system modelling

Auteurs

J. A. Torres-Matallana, U. Leopold, K. Klepiszewski, and G. B. M. Heuvelink

Référence

in 10th International Urban Drainage Modelling Conference, Monte-Saint-Anne, Canada (UDM 2015), September 20-23, 2015

Description

This paper introduces an uncertainty analysis framework in urban drainage modelling and focuses on the application of the EmiStat model which simulates the volume and substance flows in urban drainage systems (UDS). EmiStat aids the planning and design of UDS without the requirement of extensive simulation tools. An implementation of EmiStat as an R-version, the EmiStat-R model, was realised. EmiStat-R can make use of R functionalities, such as time series analysis, modelling and visualisation. Uncertainty is often ignored in urban drainage modelling. Commercial software used in engineering practice typically ignores uncertainties and uncertainty propagation, among others because of lack of user-friendly implementations. This can have large impacts, such as the wrong dimensioning of UDS and the inaccurate estimation of pollution in the environment. The paper presents the EmiStat-R model and illustrates its use with a case study from the Haute-Sûre catchment in Luxembourg for 10 rainfall events. An accuracy assessment of the model predictions with independent observations was performed. The case study results indicate that model predictions and independent observations of volume in the combined sewer overflow tank (CSOT) for rain events without and with combined sewer overflow (CSO) agree overall on the temporal pattern. However, the inflow to the CSOT and accordingly the activated storage volume is overestimated by the model in events without CSO. The results of the simulations of rain events with CSO showed that the volume in the CSOT curve is not well simulated, having more volume than the observed curve. The causes are model input, model parameter and model structure uncertainty, while uncertainties in observations and the conversion of validation data explain part of the deviation between simulated and observed time series. An important aspect is, to compare simulations and observations at the same temporal support. In order to analyse the contributions of the various sources of uncertainty, we propose a formalised uncertainty framework for UDS modelling.