An artificial neural network-based rainfall runoff model for improved drainage network modelling

This Presentation is brought to you for free and open access by the City College of New York at CUNY Academic Works. It has been accepted for inclusion in International Conference on Hydroinformatics by an authorized administrator of CUNY Academic Works. For more information, please contact AcademicWorks@cuny.edu.11 th International Conference on Hydroinformatics HIC 2014, New York City, USAModelling rainfall-runoff processes enables hydrologists to plan their response to flooding events. Urban drainage catchment modelling requires rainfall-runoff models as a prerequisite. In the UK, one of the main software tools used for drainage modelling is InfoWorks CS, based on relatively simple methods which are relatively robust in predicting runoff. This paper presents an alternative approach to modelling runoff that will allow for the complex inter-relation of runoff that occurs from impermeable areas, permeable areas, local surface storage and variation in rainfall induced infiltration. Apart from the uncertainties associated with the measurement of connected surfaces to the drainage system, the physical processes involved in runoff are nonlinear, making artificial neural networks (ANNs) an ideal candidate for modelling them. ANNs have been used for runoff prediction in natural catchments, and recently on a study for predicting the performance of urban drainage systems. This study seeks to determine an input set that predicts sewerage flow in urban catchments where the runoff is dominated by infiltration, a major issue for the water industry. A framework is proposed in which an ANN is trained by an evolutionary algorithm, which optimises ANN weights; results are assessed using the Nash-Sutcliffe Efficiency Coefficient. The model is demonstrated on a real-world case study site for which rainfall, flow, air temperature and groundwater levels in three boreholes have been measured. Various combinations of these data are used as model inputs, examining a mixture of daily and sub-daily timesteps. The best predictions are generated from daily linearly combined antecedent rainfall and air temperature, although sub-daily information improves the worst-case performance of the model. Although infiltration is affected by groundwater levels, incorporating groundwater into the model does not improve predictions. The proposed ANN model is capable of producing acceptable predictions, thus avoiding many of the uncertainties involved in traditional infiltration modelling