Statistical modelling of hydrological performance in a suite of green infrastructure practices

Abstract

Statistical modelling procedures (feature selection in conjunction with multiple linear regressions) were applied to determine the performance of a suite of stormwater green infrastructures (GIs) installed at the Belknap Campus of the University of Louisville. Two separate multiple linear regression models (MLRMs) were developed and calibrated to estimate the reductions of the flow regime parameters (flow volume and peak flow rates) within the down-gradient combined sewer system (CSS). The developed MLRMs showed that wet-weather-related CSS flow was mitigated post implementation of the stormwater GIs. At the down-gradient combined sewer flow-monitoring site, the average reduction rates of flow volume and the peak flow were estimated to be 22 and 63% per rainfall event, respectively. Unlike the black-box nature of most machine-learning techniques, the MLRM has the advantage of showing the unique statistical relationship between the rainfall features and the investigated CSS flow parameters. The results from this study indicate that proper statistic modelling can be applied effectively to evaluate the hydrological performance of stormwater management practices when lacking instrumentation and having limited drainage or sewer information. HIGHLIGHTS Feature selection plus MLRM is an effective technique for evaluating the hydrological performance of stormwater GIs.; Following the implementation of the GIs, down-gradient wet-weather flow decreased.; The flow volume reduction rate had an average of 22% per rainfall event.; The peak flow attenuation rate had average of 63% per rainfall event.;

Keywords

• green infrastructures
• flow volume reduction
• multiple linear regression
• sewer flow
• stormwater management