Flood modelling in sewer networks using dependence measures and learning classifier systems

By changing the hydrological cycle, urbanisation has led to frequent flooding worldwide. These phenomena, combined with Climate Change, threaten the capacity of sewer networks for safe conveyance of runoff. In this context, there is a need for efficient methods of modelling sewer networks, which are the main drainage systems used to deal with runoff accumulation. Hence, this research emerged to provide an efficient alternative to specialised stormwater software in terms of time and input requirements to model urban flooding. This was achieved through a methodology consisting of the combination of dependence measures in the form of factor and correlation analyses with machine learning classifier systems. The use of dependence measures enabled minimising the number of variables required by learning classifiers to perform as predictors in estimating node flooding in sewer networks. The proposed approach was tested in an urban catchment in Espoo (Finland), whose hydrological response had been previously calibrated and validated with the Storm Water Management Model (SWMM). The comparison of the node flooding distribution across the catchment was carried out under different rainfall events associated with Climate Change. As a result, the methodology was demonstrated to be capable of reproducing the flooding results obtained both with SWMM and Multiple Regression Analysis (MRA) approaches with high accuracy.