Conceptual stormwater quality models by alternative linear and non-linear formulations : an event-based approach

Different linear transfer functions (TFs) models are compared to the traditional non-linear rating curve (RC) model in a novel event-based exploratory frame. This approach aims to scrutinize for alternative representations of the stormwater TSS load dynamics at the output of a 185-ha urban catchment, when not considering explicitly a virtual state of available pollutant mass accumulated during the dry weather and subsequently washed by rainfall (as a simplification of the widespread accumulation/wash-off model). The advantages of using TFs remain unproved compared to the RC model, which shows more performant results in terms of parsimony and simulation uncertainties with a dataset of 255 calibration and 110 verification online-monitored rainfall events. This can be explained by the non-linear properties of RC. Modeling performance when rainfall is used as the input remains lower than for the case of flow rate (from mean NS of 0.6 to 0.4 in verification events). On the other hand, the RC by itself can be still considered an unsatisfactory model (e.g., NS about 0.6 in verification with the flow rate as input), suggesting the lack of an essential missing process in this formulation. However, this missing process (if there is one) might not be necessarily linked to a time-decreasing state of available pollutant mass.