Author(s): Siming Gong; James Ball; Holger Paxi Alvarez

Linked Author(s): James Ball

Keywords: Hydroinformatics AI DNNs SWMM Urban

Abstract: Urban areas are susceptible to flood events. The modelling of urban catchments presents challenges due to the heterogeneous and dynamic nature of many catchment characteristics inclusive of land use and land cover (LULC) and the drainage infrastructure. Estimation of parameters for urban catchment models involves extensive data interpretation and processing, which can overwhelm conventional methodologies and yield inconsistent results. Recently, Deep Neural Networks (DNNs) have been used in the fields of image classification and segmentation. Motivated by these advancements, the primary aim of this research was to develop an approach for building an urban catchment model that harnesses DNNs to interpret spatial data for the purposes of parameter estimation. Proposed in this study is an integrated approach that combines DNNs, clustering algorithms, and Geographic Information System (GIS) techniques to facilitate data assimilation and interpretation for urban catchment modelling. A replicable framework is devised to estimate parameters by grouping the extracted LULC features and utilizing them to initialize parameters at the subcatchment scale. An EPASWMM model for a catchment in Sydney, Australia was constructed using the proposed strategy. Use of this uncalibrated model demonstrated the reliability and consistency of the output based on the AI estimated parameter values.

DOI:

Year: 2025