Uncertainty Analysis in Data-Scarce Urban Catchments

Abstract

The evaluation of the uncertainties in model predictions is key for advancing urban drainage modelling practice. This paper investigates, for the first time in Mexico, the effect of parameter sensitivity and predictive uncertainty in an application of a well-known urban stormwater model. Two of the most common methods used for assessing hydrological model parameter uncertainties are used: the Generalised Likelihood Uncertainty Estimation (GLUE) and a multialgorithm, genetically adaptive multi-objective method (AMALGAM). The uncertainty is estimated from eight selected hydrologic parameters used in the setup of the rainfall-runoff model. To ensure the reliability of the model, four rainfall events varying from 20 mm to 120 mm from minor to major count classes were selected. The results show that, for the selected storms, both techniques generate results with similar effectiveness, as measured using well-known error metrics; GLUE was found to have a slightly better performance compared to AMALGAM. In particular, it was demonstrated that it is possible to obtain reliable models with an index of agreement (IAd) greater than 60 and average Absolute Percentage Error (EAP) less than 30 percent derived from the uncertainty analysis. Thus, the quantification of uncertainty enables the generation of more reliable flow predictions. Moreover, these methods show the impact of aggregation of errors arising from different sources, minimising the amount of subjectivity associated with the model’s predictions.