Validation and uncertainty analysis of a stormwater biofilter treatment model for faecal microorganisms

Stormwater biofilters, also known as rain gardens or bioretention systems, are effective stormwater treatment systems. This paper presents the validation, sensitivity and uncertainty analyses of a model for microbial removal in stormwater biofilters. The model, previously developed based on a rather limited laboratory study, was fully validated using the data collected in extensive laboratory experiments and field tests. The lab-scale and field-scale systems used for validation were of various designs (e.g., system size, plant type, media type), and have been operated under a wide range of operational conditions (e.g., length of antecedent dry period, and the inflow volume and concentration). For each tested biofilter design, the predicted E. coli concentrations in biofilters' outflow showed relatively good agreement with the measured ones: e.g., Nash-Sutcliffe Efficiency (E_c) ranged from 0.50 to 0.60 for the laboratory tests, and E_c = 0.55 for the field system. The results from sensitivity analysis confirmed the significance of adsorption and desorption processes, and also revealed the impact of temperature on microbial die-off (which was not fully represented in the model development stage). Finally, parameter transferability from one system to another with similar design was examined, achieving generally promising E_c values (0.04–0.56 with the best-fit parameter set for the other system; maximum value: 0.46–0.63) and acceptable uncertainties (intersection between prediction uncertainty band and observation: 50%–97%). Most importantly, the prediction of E. coli outflow concentrations from the field system was reasonably good when laboratory-determined parameter values were adopted: with the best-fit parameter set for the lab-scale system, E_c = 0.39; maximum E_c = 0.55; intersection between prediction and observation = 83%. These results suggested that the very rare biofilter model for microbial removal could provide reliable prediction for large scale field systems, by simply calibrating parameters with limited laboratory-scale experiments.