Assessment of sampling strategies for estimation of site mean concentrations of stormwater pollutants
Introduction
Since the 1970s, many studies have demonstrated the environmental impacts caused by polluted stormwater runoff, such as urban water quality degradation and large lake or costal eutrophication (e.g. Jeng et al., 2005, Lewitus et al., 2008). The estimation of pollutant concentrations/loads derived from urban stormwater systems is an important requirement for sound urban water management. For example, it is required for adequate design of urban stormwater treatment systems, assessment of pollution impact on receiving water bodies, or stormwater model development and assessment (Overton and Meadows, 2013, Daly et al., 2014). Understanding of stormwater runoff pollutant levels is indeed very valuable for setting sound targets for environmental management in urban areas.
The Event Mean Concentration (EMC) and Site Mean Concentration (SMC) are two common approaches to reporting the concentration of pollutants in urban stormwater runoff. EMC is the total mass of the pollutant in a particular runoff event divided by the total volume of runoff during that event. The SMC is the total load of a pollutant divided by the total runoff volume for a given period of time (e.g. one or more years) and is calculated using EMCs and event volumes recorded over a number of runoff events (WEF and ASCE, 1998). If the SMC is known for a pollutant, the long term discharge load of the pollutant is simply estimated by multiplying it by the total runoff volume (that can be often modelled or measured). This approach is widely used in practice (Gromaire et al., 2002), although it is still not clear how to accurately determine the SMC (Mourad et al., 2005), as discussed below.
To assess the EMC, two approaches are often used: (i) random grab sampling during events (often limited to one or only a few per event) and (2) flow-weighted or time-weighted sampling resulting in creating a composite sample from sub-samples (Schiff, 1996, Law et al., 2008). Random grab sampling is the easiest method (USEPA, 2002), yet it can introduce high variability and bias; e.g. samples taken at the start of an event or during very intense periods of the event are usually more polluted than samples taken towards the end of, or during low intensity periods of, the same event. Grab sampling requires a crew to be sent out for collecting samples, and is feasible only if the crew is located near the monitoring area. Continuous auto-sampling and sampling using flow-weightings is often considered the most robust (Lee et al., 2007), but comes with additional costs and can sometimes be problematic due to the complex installations that are inevitably associated with automatic sampling equipment. Automatic sampling is also considered unsuitable for some pollutants, such as pH, oil and grease, due to the need to analyse samples within a short timeframe after collection (Khan et al., 2006, Gulliver and Anderson, 2007). In general, automatic sampling is not widely practiced in industry and municipalities, although it is very likely the best method available and the most frequently used by researchers. Nonetheless, the extensive stormwater runoff monitoring campaigns carried out since the 1980s across the world (USEPA, 1983, Duncan, 1999, Zgheib et al., 2012) have all used different methods and protocols, offering very little consistency and comparability (Schiff, 1996, Leecaster et al., 2002). This is mainly due to the lack of agreed protocols on how to monitor stormwater.
There are a limited number of studies on stormwater monitoring strategies (Kronvang and Bruhn, 1996, Leecaster et al., 2002, Lee et al., 2007, Harmel et al., 2010, Madarang and Kang, 2013). Leecaster et al. (2002) recommended that sampling seven storms annually was the most efficient method for obtaining small confidence intervals (amount of datable trend is 10%) for annual average Total Suspended Solids (TSS) concentrations. The same study also found that taking 12 flow-weighted samples can most effectively characterize single storms while the simple random sampling of all storms or medium and large stormwater events resulted in the least bias in estimating annual TSS mass emissions and concentrations (Leecaster et al., 2002). Madarang and Kang (2013) found that 12 storms were required to estimate annual TSS loads using random methods, and monitoring with equal numbers of storms from the wet and dry seasons best estimated annual loads, with median relative error being 0.22–1.42%. Both studies, however, used data sets collected at one or two catchments over relatively short periods (<2 years) and for TSS only. Lee et al. (2007) evaluated several stormwater monitoring programs to identify high discharges and total maximum daily loads (TMDLs) of pollutants using monitoring data sets of six different catchments, for some of which over ∼10 years period existed. Results show that the required number of samples is related to the coefficient of variation of sampled pollutant concentrations. This study, however did not assess specific sampling strategies, such as the number of storms to sample, or the number of samples to take per event. Also, very few studies have ever evaluated the influence of pollutant type on sampling strategies; the question should be asked whether recommendations made for TSS also hold for nutrients and microorganisms. The lack of relevant studies may be due to the fact that large data-sets of good quality are required for a number of different pollutants and catchment types to draw reliable conclusions, since stormwater quality is highly variable (both temporary and spatially).
This study aims to fill this gap and develop sound sampling strategies for estimation of Site Mean Concentrations a range of common and important stormwater pollutants which represent three different pollutant behaviour-types; TSS, which measures the presence of particles (and often is used as a surrogate for attached pollutants such as some heavy metals (Herngren et al., 2005)), total nitrogen (TN), which is highly soluble (and behaves differently to TSS (Taylor et al., 2005)) and Escherichia coli (E. coli) which is a microorganism and inherently behaves differently than the more conservative traditional pollutants. The study uses large data sets collected at seven different catchments in Melbourne over six years. The specific objectives were to:
- •
Assess a total of 17 different strategies, including random sampling and fixed sampling strategies for assessing SMCs of TSS, TN and E. coli;
- •
Estimate the number of samples and the number of events needed to reasonably represent SMC for different pollutants and catchments; and,
- •
Understand the correlation between the variability of the pollutants concentrations and the number of events required to adequately estimate pollutant concentrations.
Section snippets
Sampling sites
Field measurements were carried out at seven catchments of different sizes and land-uses in Melbourne: Eley Road (ER), Gilby Road (GR), Monash Roof (MR), Narre Warren (NW), Richmond (RI), Ruffey's Lake Park (RU), and Shepherd's Road (SR). The catchment characteristics, including total catchment area, impervious area, time of concentration, runoff coefficient and land-use type, varied for the catchments according to Table 1. The monitoring methods and their results are fully published in Francey
Random vs. fixed sampling strategies
Among all the 17 sampling strategies, the random sampling strategies were most effective (Fig. 1). The ratio between estimated and measured/true SMCs had the smallest estimated ranges (ratio = 0.78–1.10) for these strategies, meaning that no matter which samples were selected randomly, results were similar. Further for TSS and TN, these median ratios were close to 1, meaning that they were close to the measured SMCs, while somewhat under-predicting the measured values of E. coli. The reason of
Conclusions
Effective and cost-efficient sampling methods are important for determining stormwater pollutant site mean concentrations (SMCs), which in turn is a requirement for effective stormwater management. This study evaluated various random and fixed sampling strategies for the SMCs of TSS, TN and E. coli for seven different catchments. The main findings were:
- •
Random sampling strategies could reproduce SMCs, especially for TSS and TN, with the average of the estimated/measured SMCs ratio of the studied
Acknowledgement
Melbourne Water and EPA Victoria are gratefully acknowledged for funding this project. Special thanks goes to Graham Rooney, Peter Poelsma, Justin Lewis, Gislain Lipeme Kouyi and Marjolaine Metadier for their contributions to the project. The authors would also like to thank Monica Sanders for her proof reading.
References (29)
- et al.
Redefining the stormwater first flush phenomenon
Water Res.
(2010) - et al.
Stormwater pollutant runoff: a stochastic approach
Adv. Water Resour.
(2014) Urban Stormwater Quality : a Statistical Overview
(1999)- et al.
Statistical evaluation and optimisation of stormwater quality monitoring programmes
Water Sci. Technol.
(2007) - et al.
New insights into the quality of urban storm water in south eastern Australia
J. Environ. Eng.
(2010) - et al.
Storm water runoff concentration matrix for urban areas
J. Contam. Hydrol.
(2007) - et al.
Operational Use of Urban Drainage Pollutant Flux Models - Results from a French Survey
(2002) - et al.
Assessment of Stormwater Best Management Practices
(2007) - et al.
Impact of sampling techniques on measured stormwater quality data for small streams
J. Environ. Qual.
(2010) - et al.
Understanding heavy metal and suspended solids relationships in urban stormwater using simulated rainfall
J. Environ. Manag.
(2005)
Impact of urban stormwater runoff on estuarine environmental quality
Estuar. Coast. Shelf Sci.
Oil and grease measurement in highway runoff—sampling time and event mean concentrations
J. Environ. Eng.
Choice of sampling strategy and estimation method for calculating nitrogen and phosphorus transport in small lowland streams
Hydrol. Process.
Monitoring to demonstrate environmental results: guidance to develop local stormwater monitoring studies using six example study designs
Cited by (29)
Effects of sampling strategy in rivers on load estimation for Nitrate-Nitrogen and total Phosphorus in a lowland agricultural area
2022, Water ResearchCitation Excerpt :The results show that periodic sampling of discharge causes bigger error than the nutrient concentrations. In addition, the accurate periodic grabbed samples of discharge would require the working staff to measure discharge values during the rain events or shortly after them, which is not always feasible in practice (McCarthy et al., 2018). Therefore, a continuous discharge is the best solution for measuring the discharge and to estimate the load.
A decade of monitoring micropollutants in urban wet-weather flows: What did we learn?
2022, Water ResearchCitation Excerpt :For each micropollutant and site, the ROS summary statistics mean was used as Site Mean Concentration SMCsite. The calculated SMCsite is thus the mean of the measured concentrations and it is not weighted with the volume of the event, as often done in other studies (e.g. McCarthy et al., 2018). The comparison of mean to median per micropollutant and site resulted in an 80%-interquantile range of 0.96 to 2.3, confirming that for most micropollutants the data are skewed and a lognormal distribution is more suitable than a normal distribution (SI Fig. A 5).
An IUWM incorporated model to improve water supply reliability in intermittent and no service areas
2022, Resources, Conservation and RecyclingCryptic night-time trace metal and metalloid contamination in an intensively cultivated coastal catchment
2021, Environmental PollutionCitation Excerpt :Effects of excess dissolved metals and metalloids include eutrophication (Khan and Mohammad, 2014), acute mortality events (e.g. fish kills) (Dawson, 2002), non-lethal effects (e.g. diminished reproductive success) (Sultana et al., 2017), bioaccumulation in the food chain (Abdel-Baki et al., 2011), loss of aesthetic/recreational use (Walker et al., 2001), and potential health risks to humans (Duruibe et al., 2007). Dissolved contaminant concentrations can vary by up to ∼1000% over both seasonal (Nimick et al., 2005) and diel timescales (Gammons et al., 2015), although many monitoring strategies often assume minor diel variability and implement seasonal or event-flow sampling schemes (McCarthy et al., 2018). Physiochemical factors such as pH, dissolved oxygen (DO), dissolved organic material, redox potential, electrical conductivity, temperature, precipitation, and catchment erosion/runoff influence contaminant concentrations (Loux, 2000; Nimick et al., 2011; Roussiez et al., 2013).