Effects of model complexity on rainfall-runoff computer modelling for a catchment in Singapore

Trang Vu, Tommy S.W. Wong, Soon-Keat Tan

Thursday 2 july 2015

17:33 - 17:36h at Central America (level 0)

Themes: (T) Water resources and hydro informatics (WRHI), (ST) Catchment hydrology, Poster pitches

Parallel session: Poster pitches: 13H. WRHI - Catchment

Numerical simulations have been widely used to facilitate urban stormwater management in catchments. To develop these computer models, there are no definitive guidelines on the complexity of the models. As such, how the model complexity affects the performance of a model has been a subject of discussion in numerous papers. Despite of all the earlier studies, it is still not clear how the spatial distribution of the model layout affects on the model performance. In this paper, by means of the Upper Bukit Timah catchment in Singapore, the sensitivity of rainfall-runoff modeling in response to model complexity is examined. Three models were developed: a complex model consists of 28 subcatchments, a simple model consists of ten subcatchments and a simplest model consists of two subcatchments. The software package Stormwater Management Model (SWMM) was used to generate the simulated hydrographs. The performance of the models were evaluated using the Nash-Sutcliffe coefficient (NSC), peak error (Ep) and root-mean-squared errors (RMSE). According to NSC and RMSE, the performance of the complex model is best and the simplest model is worst. On the other hand, based on the Ep, the performance of the simplest model is best and the complex model is worst. As such, this result suggests that the choice of the optimum spatial distribution to model a catchment may depend on the intended objective of the model. Further, in view of the much greater effort required to develop the complex model as compared to the simple model, while there is not much difference in the final simulated hydrographs, it may be more cost effective to develop the simple model rather than the complex model.