Evaluation and sizing of proprietary sedimentation devices for decentralised stormwater treatment

Organisations

DIBt

German Institute of Construction Engineering

DIN

German Institute for Standardisation

DWA

German Association for Water, Wastewater and Waste

NJCAT

New Jersey Corporation for Advanced Technology

NJDEP

New Jersey Department of Environmental Protection

NRW

Federal State of North Rhine-Westphalia

Statistics

COV

Coefficient of variation

m

Number of observations

NSE

Nash-sutcliffe coefficient

p

Number of parameters in the fitted function

R²

Coefficient of determination

RMSE

Root mean square error (%)

STDV

Standard deviation (%)

η_fit

Fitted value (%)

η_mean

Mean measured removal efficiency (%)

Data analysis

A

Area (m²)

d

Device diameter (m)

d₅₀

Median particle diameter (m)

g

Acceleration due to gravity (m s⁻¹)

PSD

Particle size distribution

Q

Flow rate (m³ s⁻¹)

R

Particle radius (m)

SLR

Surface loading rate (m h⁻¹)

TSS

Total suspended solids

Vs

Settling velocity (m s⁻¹)

η

Observed removal efficiency (%)

μ

Dynamic viscosity (kg m⁻¹ s⁻¹)

ρ_p

Particle density (kg m⁻³)

ρ_f

Fluid density (kg m⁻³)

Dimensional analysis

CSTR

Continuously stirred tank reactor model

Fr

Froude number

Fr_j

Froude number of the influent jet velocity

Ha

Hazen number

Pe

Péclet number

PFR

Plug flow reactor model

SMG

Similarity modified gamma model

SOR

Surface overflow rate model

The mitigation of diffuse pollution loads remobilised from urban surfaces during rainfall events is one of the primary aims of stormwater treatment. Heavy metals, phosphorous, nitrogen, polycyclic aromatic hydrocarbons and mineral oil hydrocarbons present in stormwater threaten the quality of receiving waters (Göbel et al. 2007; Revitt et al. 2014). Total suspended solids (TSS), present at concentrations between 43 mg L⁻¹ for roof runoff and 150–163 mg L⁻¹ for carparks and main roads (Göbel et al. 2007), can act as the primary carrier for pollutants (Kayhanian et al. 2012), with 63–81% of heavy metal loads (Zn, Cu and Ni) being particulate bound (Hilliges et al. 2017). This has led to TSS removal efficiency becoming one of the primary mitigation approaches for stormwater treatment systems (DWA 2020). To achieve cost-effective TSS removal, practitioners are moving towards decentralised stormwater treatment through the implementation of proprietary or manufactured stormwater treatment devices (Dierkes et al. 2015). Survey data from UK stormwater practitioners indicates that 70% consider proprietary devices as integral in reaching removal targets (Engineering Nature's Way 2016), with 7–8% of sites employing such systems (Melville-Shreeve et al. 2018). Proprietary sedimentation systems include oversized pipes (pipe sedimentation), lamella settlers, hydrodynamic separators and combinations with filtration–adsorption media. Hydrodynamic separators are becoming popular due to their practical benefits of being easily retrofitted into existing chambers (Lee et al. 2014) and high treatment efficiencies through the use of fixed internal structures to induce both gravitational settling and advection (Fenner & Tyack 1997; Wilson et al. 2009). However, stormwater professionals are often reliant on manufacturer reported removal efficiencies when assessing device performance, as there is currently no European or international standard for proprietary devices (Dierkes et al. 2015) and field studies note variable device performance (Faram et al. 2007; Lee et al. 2014; Fettig et al. 2017). Manufacturers rely on a patchwork of national and local testing protocols to provide standardised third-party confirmation of performance (Dierkes et al. 2015). In Europe the key test protocols are administered by the German Institute of Construction Engineering (DIBt) for sedimentation–filtration–adsorption devices (DIBt 2017), the Federal state of North Rhine-Westphalia (NRW) and the German Association for Water, Wastewater and Waste (DWA) for sedimentation devices (DWA 2020), and in the US the New Jersey Department of Environmental Protection (NJDEP) through the New Jersey Corporation for Advanced Technology (NJCAT) (NJDEP 2013).

During device selection, exact stormwater characteristics are typically unknown; instead, a practitioner will use classification systems that stipulate treatment levels and technology types based on land use within the connected treatment area (Dierkes et al. 2015). To ensure appropriate treatment levels, proprietary treatment device options are typically limited to a list of approved devices and technologies, with associated treatment flow rates (NJCAT) and defined scaling rules (identical SLR, NJCAT), or connectable treatment areas (DIBt) for each tested device. From a practitioner's perspective, a generalised method for the translation of devices between certification protocols would serve to extend the practical usefulness of certification data. In addition, when assessing the likely performance of an uncertified device, it is important to have a firm grasp of general performance capabilities of a given technology type.

Within the scientific literature, the performance of proprietary sedimentation devices is commonly assessed through dimensional analysis. Such analysis assesses the relative impact of gravity, advection, turbulent diffusion and inertial forces (Fenner & Tyack 1997; Wilson et al. 2009; Boogaard et al. 2017; Tang et al. 2018), and incorporates correction or fitting parameters to account for hydraulic under performance (Wilson et al. 2009; Boogaard et al. 2017). As such, Li & Sansalone (2022b), applied the Hazen number (Ha) as the dimensionless settling velocity (gravity removal), estimated the performance of 22 devices submitted for NJCAT testing, combined data from cylindrical and rectangular designs, and demonstrated no performance benefit relative to cylindrical catch pit geometries developed by Howard et al. (2012). However, Howard et al. (2012) applied the Péclet number (Pe), the ratio of settling due to gravity and advection, as the dimensionless settling velocity, and noted that device aspect ratio impacts performance through changing the characteristic length scale i.e. transition from tank diameter to water depth as the shortest dimension (Wilson et al. 2009). Similar studies, applying Ha or Pe as the dimensionless settling velocity within hydrodynamic separators (Wilson et al. 2009; Tang et al. 2018) note the need for fitting functions unique to each device, indicating that device design has an impact on performance. In addition, the similarity modified gamma (SMG) model, applied by Li & Sansalone (2022b) to the NJCAT data set, was unable to estimate TSS removal efficiency. This was attributed to geometric differences between devices preventing the assumption of dynamic similitude. Moreover, Li & Sansalone (2022b) applied the nominal NJCAT particle size distribution (PSD), where the median particle size (d₅₀) is 75 μm; however, the average PSD applied in practice is finer (d₅₀ = 58 μm). As such, any technology-specific performance increase may be hidden by the assumption of a coarser sediment. As a consequence, this study seeks to distinguish hydrodynamic (cylindrical, water depth as shortest dimension) and non-hydrodynamic (rectangular, tank diameter as the shortest dimension) within the NJCAT data set, and to apply the average PSD, when assessing the impact of technology type on performance.

The scientific literature addressing the link between technology type, performance estimation (removal efficiency at design flow rate) and experimental measurement is inconsistent, with practitioners doubting manufacturer performance claims (Pretorius 2012; Minton 2013). Studies have found little difference in performance between technology types (Boogaard et al. 2017; Li & Sansalone 2022b), little difference between designs of a given technology type (Wilson et al. 2009), and variability between technology types (Fettig et al. 2017). Studies at low (1–16 m h⁻¹) SLR have indicated the dominance of gravity separation for lamella, sedimentation pipes and hydrodynamic separators, and therefore little difference in performance (Boogaard et al. 2017). However, design flow rates for hydrodynamic separators may extend up to 140 m h⁻¹ (Field et al. 1997), with Wilson et al. (2009) reporting that hydrodynamic separators at design flow rates are well described when including the impact of advection and turbulent diffusion. This is further compounded by studies indicating the need for predictive equations to transition from gravity-dominated descriptions at low SLR to descriptions including the impact of advection, turbulent diffusion and inertial forces at higher SLR (Fenner & Tyack 1997; Tang et al. 2018), indicating that operational conditions may impact observed performance. Consequently, to determine the influence of technology type on performance, this study seeks to assess performance data across the entire operational range (<1–200 m h⁻¹) of proprietary sedimentation devices.

This study therefore applies dimensional analysis to official test results from third-party certification, covering a large operational range (<1–200 m h⁻¹) and multiple technology types, supplemented with experimental testing of uncertified devices, to elucidate the general benefits of commercial hydrodynamic separators relative to other technologies and the translatability of performance data between certification data sets. Specific objectives are to: (i) compare the performance of devices tested at low SLR to determine the influence of gravity; (ii) compare the performance of devices tested at high SLR to determine the influence of advection; (iii) determine the suitability of a generalised single parameter estimation as a preliminary performance indicator and (iv) determine the application benefits of hydrodynamic separators.

A powder-metering pump (K-MV-KT20, Coperion K-Tron (Schweiz) GmbH), located 1 m upstream of the device, injected sediment through an open t-piece into the influent flow to create an influent TSS concentration of 200 mg L⁻¹, in accordance with NJDEP stipulations. A transparent section (1 m) directly upstream of the powder-metering pump was included to monitor for any unintended sediment deposition in the connecting pipework at low flows.

Quality assurance for the experimental set-up and sample collection was addressed by employing methodology described in the NJDEP protocol (NJDEP 2013). Accordingly, the feed sediment's concentration coefficient of variance (COV) did not exceed 0.1, whilst the COV for the target flow rate did not exceed 0.03. Once a constant flowrate and sediment feed was established, 15 effluent samples were taken via grab sampling at 15 min intervals. TSS analysis was achieved using the German DIN 38409-2 method, as stipulated in the DIBt protocol (DIBt 2017), where the TSS mass in 1L effluent grab samples taken from the sample point (Figure 1) was measured via filtration, drying and weighing. The samples were first passed through a 0.45 μm filter connected to a vacuum pump (LABOPORT N 816, KNF Neuberger GmbH) prior to drying at 105 °C in an oven (UF30, Memmert GmbH + Co. KG), desiccation and weighing (ABS 80-4, KERN & Sohn GmbH).

Removal efficiency data for sedimentation devices were gathered from publicly available reports. A report database was selected if it contained device performance data gathered using a standardised test procedure, a device-specific test report, third-party witnessing of testing, and acceptance of results by practitioners. To ensure technology type remained the independent variable, the standardised test procedure was required to employ a quartz sand sediment with a well-defined PSD and specific density of 2,650 kg m⁻³. In addition, to measure the mass of non-captured sediment at variable SLR, the standardised test procedure was required to stipulate that sediment be introduced continually into the influent upstream of the tested device and grab sampling employed at the outlet. As such, data sets published by NRW (based on the DIBt protocol) and NJCAT through the NJDEP protocol (Table 1) were chosen. A similar data set published by the Canadian Environmental Technology Verification protocol was not chosen, as it measured the mass of TSS retained in a device, rather than employing effluent grab sampling, and PSD information was only provided graphically. The two selected protocols employed different sediment types (Figure 2). However, the PSD of each test sediment was similar (Figure 2) and representative of stormwater in their respective countries (Dierschke et al. 2010; Boogaard et al. 2014, 2017). Detailed information on test procedures can be found in the published protocols (NJDEP 2013; DIBt 2017) and within individual test reports cited in Table 1. An individual report was selected for analysis if the tested device employed sedimentation as its removal mechanism (no filtration element), if it reported the available surface area for sedimentation (for the calculation of an SLR), and if it demonstrated a change in treatment efficiency with flow rate. Therefore, two reports from the NRW data set were excluded, which did not define surface area. In addition, two device reports were excluded from the NJCAT data set; one where the device contained a mesh filter, and the second where removal efficiency did not change with flow rate.

In hydrodynamic separators, tank geometry and internal structures are designed to induce additional settling though bulk fluid transport (advection) (Wilson et al. 2009; Howard et al. 2012). Typically this is in the form of rotational flow induced by tangential flow entry in a cylindrical tank (Wilson et al. 2009; Shrestha & Brodie 2011). For the purposes of this study, only devices that demonstrate rotational flow within a cylindrical tank are classified as hydrodynamic. Efficiency calculations for hydrodynamic separators include the impact of advection through the Péclet number (Pe = V_sd²/Q), where d is device diameter (m) and d²/Q analogous to SLR (Fenner & Tyack 1997; Wilson et al. 2009; Pretorius 2012), and the Froude number (Fr = Q²/(d⁵g)) to include inertial forces (Fenner & Tyack 1997; Egarr et al. 2009; Pretorius 2012; Tang et al. 2018). Howard et al. (2012) studied standard sumps and reported a good statistical fit through the dimensionless parameter Pe/Fr_j², where Fr_j is the Froude number of the influent jet velocity. However, Fenner & Tyack (1997) noted that Froudian scaling overestimates removal at low SLR, corresponding to removal efficiencies >50%. As the NJDEP test protocol considers weighted annual removal efficiencies of ≥50% as pass criteria for practical application, and due to the variability in inlet pipe diameters (25%–50% of device diameter), it was decided to exclude the contribution of inertial forces (Fr) from the estimation. This is compounded by the trend for device manufacturers to include oversized inlet and outlet connections relative to the designed maximum treatment flow rate to accommodate integrated bypass structures.

To assess the goodness of fit of Ha and Pe to the data, statistical measures were employed. The two key measures being the Nash–Sutcliffe coefficient (NSE) as a measure of variance relative to the mean of the data set, and the root mean square error (RMSE) as the average deviation of the fitted removal efficiency function from the observed data.

For hydrodynamic separators this covers a tight range of device diameters (0.76 m to 1.22 m), where differences in performance can be more easily ascribed to design features. Removal efficiency under the NRW procedure (Figure 3) displays a non-linear decrease in η with increased SLR, where η ranges from 50% at 20 m h⁻¹ to 98% at 0.13 m h⁻¹. A similar non-linear decrease is evident in Figure 4 for the NJCAT data, where η ranges from 34% at 198 m h⁻¹ to 67% at 20 m h⁻¹. This is in line with previous literature reporting on η as a function of SLR and PSD rather than a discreet (d₅₀) particle diameter (Lee et al. 2014). However, NJCAT η appears higher at identical SLR, for example 67% at 20 m h⁻¹ for NJCAT but 50% at the same SLR for NRW. This is partially explained by the impact of temperature differences (Table 1) on fluid viscosity and the variation is PSD (Figure 2) on settling velocities. Whilst the d₅₀ of both sediments is around 60 μm, the NJCAT sediment contains a readily settable >300 μm fraction not present in the NRW sediment, this additional fraction accounts for ∼7% of the total mass (Figure 2) and can be expected to increase observed η. To contextualise the impact of PSD and temperature on observed η, an estimation of η based on Ha is plotted in Figures 3 and 4. As a generalised trend, observed η for each technology type and individual device follows the Ha estimation for the NRW procedure (Figure 3). However, the generalised trend is less clear for the NJDEP procedure (Figure 4) with some hydrodynamic devices appearing to over perform, and most non-hydrodynamic devices appearing to underperform. For example, H5 outperforms the estimation by 16% at an SLR of 20 m h⁻¹ reducing to a 4% outperformance at 198 m h⁻¹. In comparison, L4 underperforms by 9% at an SLR of 4.3 m h⁻¹ increasing to 24% underperformance at 20 m h⁻¹.

Removal efficiency under the NRW and NJDEP procedure (Figure 5) displays a linear decrease in η with increased Ha, when plotted semi-logarithmically. Statistical analysis of the NRW data set (Table 2) yielded an NSE value of 0.81, an R² of 0.89 and an RMSE of 5.1%. The analysis indicates that the single parameter predictor (Ha) yields a good statistical fit in terms of reproducing the data set mean (NSE), of data variance (R²), and the average deviation of η_fit from observed data (RMSE). In contrast, statistical analysis of the NJCAT hydrodynamic data set (Table 2) yielded a lower NSE value of 0.06, an R² value of 0.59 and an RMSE of 8%. The low NSE value indicates that Ha offers an estimation marginally better than the average of the observed data, and the RMSE value of 8% is comparable to the STDV (8.7%), indicating that Ha as a single parameter predictor offers an unsatisfactory statistical fit. When plotting η as a function of Pe for the NJCAT hydrodynamic data set (Figure 5, inset), there is a visual tightening of the data grouping. Statistical analysis confirms this observation (Table 2) with the NSE increasing to 0.5 and the RMSE dropping to 5.9%. This indicates that Pe yields an improved but imperfect statistical fit. Statistical analysis (Table 2) of the non-hydrodynamic NJCAT data set (Figure 4) indicates an unsatisfactory statistical fit, where the negative NSE value (−0.61) indicates that the data set average offers a better estimation than the Ha estimate. However, the R² (0.62) indicates that Ha is reasonably reproducing the data variance.

Following literature recommendations, a predictive model is described as unsatisfactory when the NSE value lies below 0.5, acceptable when the NSE value lies between 0.5 to 0.75 and as good if the value lies above 0.75 (Moriasi et al. 2007; Ritter & Muñoz-Carpena 2013). When interpreting NSE, it is important to consider context, as NSE (Equation 1) represents the ratio between the mean square error of observed η versus predicted η, and variance of observed values (Ritter & Muñoz-Carpena 2013). The selection of a 0.5 NSE cutoff point was deemed appropriate for the NJCAT data set, due to the impact of a truncated data range (Figure 4, η = 59% to η = 33%) relative to sedimentation limits (η = 100% to η = 0%), and the non-liner relationship between SLR and η, on measured STDV (Table 2).

When selecting proprietary devices, a stormwater practitioner requires a way of assessing manufacturer efficiency claims in a timely manner, and estimation accuracy should be weighed against method complexity. Analysis by Li & Sansalone (2022b), of single parameter estimation methods, have shown a good statistical fit when applying a trained SMG model to cylindrical catch pits, with R² values 0.95 to 0.98, higher than the R² values reported in this study (Table 2). However, Li & Sansalone (2022b) were unable to reproduce observed η within the NJCAT data set due to variable tank geometries. Analysis of multi-parameter estimation methods undertaken by Howard et al. (2012) and Wilson et al. (2009) studying standard sumps and hydrodynamic separators, respectively, have shown good statistical fits with NSE values ranging from 0.87 to 0.99, higher than the NSE values reported in this study (Table 2). Interestingly, the RMSE values reported by Wilson et al. (2009) range between 2.1% and 6.9% with a mean of 4%, and 8.1% by Howard et al. (2012), which is comparable to the RMSE values reported in this study (Table 2). Further multi-parameter estimation studies by Tang et al. (2018) and Ansari & Khan (2014) of sedimentation devices similarly report comparable RMSE values between 5.8 and 44%. This indicates that the single parameter Ha or Pe predictor yields a comparable estimation as more complex models, in terms of deviation of the estimated efficiency from the observed efficiency. Typically, a model becomes more accurate through the inclusion of more parameters, with an associated increase in computational steps. To reach a high NSE, Wilson et al. (2009) employed fitting parameters unique to each device, and Howard et al. (2012) included the Fr_j which limits the application to devices with known fitting parameters and necessitates a separate estimation for each inlet pipe diameter offered by manufacturers. In contrast, the single Ha and Pe predictors reported in this study are able to yield an NSE between 0.5 and 0.81, fulfilling the requirements of a satisfactory estimation with a simpler calculation. However, the NSE is lower within the hydrodynamic estimation (0.5). A weakness of the NSE as a measure of goodness of fit is its sensitivity to outlying data (Ritter & Muñoz-Carpena 2013). In the context of hydrodynamic separation, the 2013 NJDEP procedure did not set a limit on inlet pipe diameter, and states that sediment found in the inlet pipe should be added to that found in the device (NJDEP 2013). As a consequence, any pre-sedimentation in the inlet pipes would lead to an increase in observed η, relative to the estimated value, for the analysed NJCAT data set. Of the 13 hydrodynamic devices (Table 1) analysed, five (H3, H7, H9, H14 and H15) employ inlet pipes with a diameter 50% that of the treatment device, which results in an average 6% underestimation in separation efficiency (Figure 7). When these devices are excluded from the statistical analysis, the Pe NSE increases to 0.61, the RMSE reduces to 4.3%, the Ha NSE increases to 0.5 and the RMSE reduces to 4.9%. This indicates that the inclusion of an oversized inlet pipe affects the accuracy of both single parameter estimations; however, Pe continues to yield a better fit. The 2021 version of the NJDEP protocol prevents the use of inlet pipe diameters greater than 25% of the device diameter, implements a mass balance (weighing retained sediment in the device) measure of efficiency, and excludes sediment mass found in the influent pipe from efficiency calculations (NJDEP 2021). Future analysis of devices re-tested under the new test protocol would help clarify the possible impact of secondary sedimentation mechanisms, such as pre-sedimentation, on performance.

The accuracy of uncorrected single parameter estimations is limited when assessing non-hydrodynamic separators operating at high SLR (Figure 4), with statistical analysis indicating an unsatisfactory fit (Table 2). Interestingly, the R² value (0.62) indicates that the model is capturing some of the expected data variance, but the RMSE (9.3%) and NSE (−0.67) values indicate it fails to reproduce the data mean and η. This intimates that the estimations reflect the trend that higher SLR results in lower η, but over-estimates η at a given SLR, as can be seen in Figure 4. This can be ascribed to the impact of hydraulic shortcutting and maldistribution reducing effective surface area (Boogaard et al. 2017), and is noted in the literature as a common cause of underperformance in lamella separators (Fuchs et al. 2014; Boogaard et al. 2017; Fettig et al. 2017), with a lamella separator (L4) being the worst performing device in this study (Figure 4). Studying separation efficiency as a function of hydraulic conditions and settling velocity, Luyckx et al. (2005) noted that a reliable predictive model is dependent on an accurate description of hydraulic conditions, whilst settling velocity (PSD and density) influence field performance. Consequently, good hydraulic performance is a prerequisite when applying a single parameter estimation. Within the context of hydraulic performance, hydrodynamic separators have demonstrated the capability of maintaining near plug flow at high flow rates (Alkhaddar et al. 2001) and enabling single parameter predictions to maintain accuracy without the need for fitting parameters.

Single parameter estimations therefore offer a simple tool for estimating performance of hydrodynamic separators at higher SLR (Pe if SLR > 20 m h⁻¹), and for all devices operating at lower SLR (Ha if SLR <20 m h⁻¹), without the need for hydraulic fitting parameters and for translating results between certification procedures. However, hydraulic underperformance at higher SLR (>20 m h⁻¹) prevents the same approach from being applied to other technologies, lamellae in particular, without the application of a hydraulic fitting parameter.

Stormwater practitioners seek to translate performance data from certification procedures into appropriate device sizing. To achieve this, flow-weighted η based on average rainfall intensities and connected area are employed, and an annualised η calculated. The most cost-effective device size is then selected. Within a typical design scenario, 100% of a 1:1 rainfall event is treated, and the remainder by-passed, where common annualised η are 83% (average NRW data, Table 1) and 50% (NJDEP). Li & Sansalone (2022b) conducted physical model testing and demonstrated that an SMG model yields a better performance estimate than a surface overflow rate (SOR) model, a plug flow reactor (PFR) model and continuous stirred reactor (CSTR) model; however, no model is able to estimate performance within the NJCAT data set. Applying their chosen modelling approaches to clarifier design, Li & Sansalone (2022b) translated model inaccuracies into 100% oversizing (SOR), 304% oversizing (PFR) and 904% oversizing (CSTR). However, sizing was based on a fixed flow rate and a single particle diameter in contrast to the industry practice of flow-weighted averages and a PSD. Within this study, the Ha (non-hydrodynamic) and Pe (hydrodynamic) predictors were applied to the NJCAT average sediment and the recommended German annual rainfall intensities and weightings of 0.9 mm h⁻¹ (50%), 2.2 mm h⁻¹ (33%) and 9 mm h⁻¹ (17%) to estimate connectable treatment area. A sedimentation surface area of 1.22 m² was taken for the example and used to elucidate the impact of technology type on annualised η.

Within the context of stormwater treatment, the implication is that when applying sedimentation as the primary pollutant mitigation measure, then it is necessary to employ low SLR to remove <300 μm particulate-bound pollutants, and the technology type therefore has little impact. From a practitioners perspective, it becomes selection primarily based on cost, specific surface area and maintenance concerns. However, when sedimentation is included as a gross pollutant removal step and overall efficiency targets are reduced (50%), then hydrodynamic separators may offer a competitive advantage. An example application would be the protection of underground geocellular storage or litter removal from an area with a low pollutant load, such as a pedestrian zone.

This study demonstrates that no sedimentation technology type offers a performance advantage at low SLR (<20 m h⁻¹), with efficiency well explained by gravity separation. Therefore, when a practitioner targeting >80% η selects a proprietary treatment device, secondary considerations such as cost, maintenance periods or footprint are more relevant differentiators than technology type. However, hydrodynamic separators operated at higher SLR (>20 m h⁻¹) offer increased η relative to non-hydrodynamic separators, with η adequately explained by gravity and advection. As such, hydrodynamic separators offer an up to 33% increase in connectable area when targeting a 50% annualised η. In this scenario, technology type is an important differentiator. When seeking to translate the performance of hydrodynamic separators between high (>20 m h⁻¹) and low (<20 m h⁻¹) SLR, a single parameter predictor is comparable to more complex models and is shown suitable for use as a preliminary performance indicator. Such single parameter predictors (Ha or Pe) applied to third party certification data of proprietary sedimentation devices offer a statistically good estimation of performance (under certification conditions) for all technology types at low SLR (Ha if SLR <20 m h⁻¹), and are statistically acceptable for hydrodynamic separators operated at higher SLR (Pe if SLR >20 m h⁻¹). In addition, most proprietary hydrodynamic separators have similar performance characteristics under standardised conditions. Consequently, a practitioner seeking to estimate the performance of a US-certified (NJCAT) hydrodynamic separator at low SLR (NRW-DIBt protocol) to achieve >80% η expected for European operation can apply a single parameter predictor within a simple calculation capable of being completed within a spreadsheet. A final point worth further study is that devices, certified under the 2013 NJDEP protocol, may over-perform due to oversized inlet connections.

All relevant data are included in the paper or its Supplementary Information.

The authors are employed by ACO. No commercial interests were directly realised by this study, and the authors did not stand to gain financial reward.