Stormwater biofilters: A new validation modelling tool

Abstract

Stormwater biofilters must be validated before they can be a trusted component of the treatment train used for water supply augmentation. Currently, only in situ challenge testing is accepted for treatment validation, yet this is impractical for stormwater biofilters because of their size and operational conditions; e.g. stormwater harvesting biofilters are often large systems that receive significant volumes of urban stormwater during short periods of time. This study proposes an alternative validation tool for stormwater biofilters that uses a process-based model calibrated against in situ tracer and laboratory based data. The method is developed and tested using fluorescein as the reference micropollutant at two different biofilters: (i) a well-designed system that uses sand as filter media and has a submerged zone (S-SZ), and (ii) a system with loamy sand (with content of silt and clay well above best practice), which does not have a submerged zone (LS-noSZ). Firstly, a model that can simulate hydrodynamic and pollutant transport of micropollutants in stormwater biofilters was selected. In situ tracer tests and laboratory batch studies were then performed to derive the model parameters using soil samples collected from the two biofilters. Without further calibration, the model was applied to simulate a number of in situ fluorescein challenge tests performed on the biofilters. The modelled outflow concentrations were compared with the in situ measurements, showing that the proposed alternative validation method could provide reliable predictions of fluorescein removal in the S-SZ, with predicted outflow concentrations agreeable to the measured data (Nash Sutcliffe coefficient, E = 0.67). The peak outflow concentrations that are important for validation study were particularly well modelled; the differences between the modelled and measured peak values were -3.9% to +7.4% for spiking tests and -4.4% to 28% for flushing/rinsing tests. However, for LS-noSZ, the proposed tool did not work well (E = 1.7), which was attributed to the fact that flow through this system could not be reliably modelled due to high silt and clay content in the soil. The differences of peak concentrations of LS-noSZ were between 3.6% (under-predicted) and +76% (over-predicted).