Three Complementary Sampling Approaches Provide Comprehensive Characterization of Pesticide Contamination in Urban Stormwater

Abstract

Urban areas are expanding rapidly and experience diverse and complex contamination of their surface waters. Addressing these issues requires different tools to describe exposures and predict toxicological risk to exposed biota. We surveyed 21 stormwater management ponds in Brampton, Ontario using three types of sampling methods deployed concurrently: time-integrated water sampling, biofilms cultured on artificial substrates, and organic-diffusive gradients in thin films (o-DGT) passive samplers. Our objective was to compare pesticide occurrences and concentrations to inform monitoring in stormwater ponds, which reflect pesticide pollution in urban areas. We detected 82 pesticides across the three sampling matrices, with most detections occurring in o-DGT samplers. The in situ accumulation of pesticides in o-DGTs during deployment and the high analytical sensitivity achieved establishes o-DGTs as excellent tools for capturing the mixtures of pesticides present. Water and biofilm sampling demonstrated that pesticide concentrations available for uptake are relatively low, with most below toxicological thresholds. Yet our results demonstrate that urban areas are subject to a wide range of pesticides, including herbicides, insecticides, and fungicides, and underscores the urgency of research to quantify the risks of chronic exposure to this chemical mixture.

1. Introduction

Globally, urbanization is accelerating rapidly, with urban land cover expanding at a rate of 1.16 km² per hour since 1990 [1]. Urban pollution is intensified by the widespread coverage of land with impervious surfaces resulting from this rapid development and land clearing [2]. In areas with low rates of infiltration, contaminants are concentrated in runoff as water moves along surfaces and into constructed stormwater management systems (e.g., including stormwater management ponds (SWPs)) [3,4], or directly into surface waters. This phenomenon drives the high diversity of substances found in stormwater, from organic contaminants like pesticides [5,6], pharmaceuticals [4], and tire compounds [7] to metals [8], oils [9], and salt [10], which are subsequently released into receiving water bodies like creeks and streams. Such discharges are known contributors to downstream habitat degradation [2] and the impairment of benthic communities [11,12].

Urban-use pesticides are prevalent in stormwaters [6,13,14], including herbicides [15], insecticides [16,17], and fungicides [18], several of which are known toxins to aquatic organisms (e.g., refs. [19,20]). The large assortment of potential pesticide contaminants presents a challenge for characterizing contamination, because even within this singular contaminant type, there are a variety of physico-chemical properties represented, which affect the environmental fate of each compound. For example, some herbicides, such as 2,4-D and mecoprop, are hydrophilic, while fungicides, such as azoxystrobin, are relatively hydrophobic. The propensity to dissolve in water, among other physico-chemical properties, can determine which environmental compartments a pesticide will partition into once released [21]. Relying solely on sampling a single environmental compartment overlooks the other ways in which biota may be exposed [22]. To sufficiently predict the potential ecotoxicological risk of pesticide contamination in urban aquatic environments, it is crucial to accurately monitor the complete exposure profile. Below, we describe our study of the breadth of pesticides occurring in stormwater ponds in one of Canada’s fastest growing municipalities and compare three methods of sampling, (1) time-integrated water sampling, (2) biofilms cultured on artificial substrates, and (3) passive sampling with organic-diffusive gradients in thin films (o-DGT) samplers, yielding recommendations for future pesticide monitoring and research.

1.1. Water, Biofilm, and Passive Sampling for Monitoring Urban Pesticide Contamination

The Canada-Ontario Agreement (COA) on Great Lakes Water Quality and Ecosystem Health represents the commitment of the federal and provincial governments working together to meet their obligations under the Canada-United States Great Lakes Water Quality Agreement, which was first signed in 1972 by both nations to restore, protect, and conserve the health of the Great Lakes [23]. Although much is understood about the contribution of agricultural lands to pesticide loading in the Great Lakes, research on the contribution of urban lands is relatively underrepresented. In 2021, a Wastewater and Stormwater Annex was added to COA under the priority area Protecting Waters, which reflects growing concern around the impacts urban developments can have on water quality in the Great Lakes and their watersheds [23].

Water grab sampling is the mainstay of monitoring pesticide contamination in surface waters. For example, the monitoring of agricultural pesticides in Ontario’s streams, led by the Ontario Ministry of Agriculture, Food, and Rural Affairs (OMAFRA) and the Ontario Ministry of Environment, Conservation, and Parks (MECP), consists of monthly grab sampling during ice-free months [24]. Instantaneous water sampling is also used by the U.S. Geological Survey for measuring organic contaminants in lentic waters in their National Water Quality Program [25]. In 2024, Health Canada’s Pest Management Regulatory Agency (PMRA) completed a nation-wide pesticide monitoring pilot program consisting of biweekly grab samples from over 100 sites across the country, as a model for a forthcoming Canadian Water Monitoring Program for Pesticides [26]. Exposure thresholds to avoid adverse effects on freshwater biota are available for many common pesticides (e.g., CCME’s Canadian Environmental Quality Guidelines; Health Canada Pest Management Regulatory Agency’s Aquatic Life Reference Values; the U.S. EPA Office of Pesticide Program’s Aquatic Life Benchmarks) and can be used for comparison with measured concentrations in water samples. However, water grab sampling represents a “snapshot” of contaminant concentrations and biotic exposure, which may fluctuate in urban stormwaters due to the flashy hydrology characteristic of urban runoff [2]. The lack of temporal integration when taking a single grab sample of water means that contaminant pulses can be missed [27]; thus, the predicted exposure profile and toxic risks can be inaccurate over a longer period than when the sample was taken [28].

Over the past two decades, the use of biofilm sampling has been developed as a means of monitoring pesticide concentrations that better represent average conditions through time than single grab samples of water [6,22,29,30,31,32,33]. Biofilms are microbial assemblages held together by a matrix of extracellular polymeric substances (EPSs). The surfaces of biofilms are heterogenous, and this complexity offers a diverse set of potential sites for contaminant interaction [34,35]. Biofilms are known to bioaccumulate both hydrophilic (e.g., MCPA [33]) and hydrophobic (e.g., azoxystrobin [6]) pesticides, although recent evidence suggests compounds with higher octanol–water partition coefficients are more likely to accumulate in biofilms [22]. Biofilm sampling was demonstrated to be more sensitive than traditional water grab sampling for detecting pesticides in agricultural streams [36,37]; however, this is not always the case (e.g., ref. [22]). As key components in aquatic food webs, biofilms are consumed by other biota, representing an alternative route of exposure to contamination [38]. Thus, in addition to being valuable biological indicators for environmental contamination, biofilms are also important tools for assessing toxicological risks (e.g., ref. [39]), although published ecological thresholds for dietary exposure to pesticides do not currently exist, and research into pesticide exposure via dietary pathways is not nearly as well studied as exposure via immersion or contact pathways.

Passive sampling devices for monitoring low level environmental contamination of organic compounds have been used since the 1990s as another alternative technology to traditional water sampling [40]. The diffusive gradients in thin films for an organic contaminant (o-DGT) passive sampler have been used since 2012 [41]. Differential chemical potential drives the sequestration of contaminants from the ambient water onto a binding gel via a diffusive hydrogel layer contained within the sampler. The concentration of analyte in the gel and the duration of deployment can be used to model time-weighted average (TWA) concentrations in the surrounding water [42]. The o-DGT sampler differs from other similar passive sampling devices because of the inclusion of a semi-permeable membrane and diffusive gel to minimize the influence of environmental conditions on sampling performance; this adaptation of the o-DGT sampler has increased the accuracy of this method for detecting polar organic contaminants compared to other passive sampling devices [43]. Not only can passive samplers integrate contamination through time, catching acute but periodic contaminant fluxes, but the accumulation of contaminants within the binding gel enhances the sensitivity to trace contaminants compared to traditional water grab sampling [44]. However, the use of TWA concentration data from o-DGTs poses challenges in assessing risk to exposed biota, as relatively higher pulses of contaminant concentrations that may exceed toxicity thresholds are averaged across the period of sampler deployment. Passive sampling for pesticides with o-DGTs was studied under laboratory conditions (e.g., refs. [45,46]) and used successfully for monitoring pesticides in Canada [43] and New Zealand [47].

1.2. Objectives and Hypotheses

To tackle the complex challenge of assessing urban pesticide contamination in the Great Lakes Basin, we require many different tools and approaches. Yet, understanding how these tools and approaches compare to each other in effectiveness and biological relevance is critical. The overall objective of our study was to compare three sampling methods for extensively assessing pesticide occurrences and concentrations in stormwater ponds, and to make recommendations for future pesticide monitoring programs in the Great Lakes watershed. These sampling methods were the following: a composite of weekly water grab samples, biofilm cultured on artificial substrates, and o-DGT samplers (summarized in

Table 1. Summary of three sampling methods for assessing pesticide contamination in surface waters.

	Grab Water Sampling	Biofilm Sampling	Passive Sampling with o-DGTs
Description of method	Instantaneous sampling of a water body by filling a sampling bottle with water directly from the study site	Harvesting biofilm grown in situ at a study site from either artificial or natural substrates	Deployment of passive sampling device (e.g., o-DGT) in situ at a study site
Benefits of this method	Concentrations in water are comparable with other studies and with ecotoxicological benchmarks or thresholds	Accumulation of pesticides in biofilms from the surrounding water can help detect pesticides at relatively low levels; time-integration allows capture of fluctuations in concentrations; biologically relevant as many aquatic organisms consume biofilm directly	Accumulation of pesticides from the water can help detect pesticides at relatively low levels; time-integration allows capture of fluctuations in concentrations; well supported in the literature
Limitations of this method	Snapshot of concentrations in time may not represent all conditions or exposures; bias toward hydrophilic compounds	No dietary ecotoxicological thresholds or benchmarks exist for comparison	Artificial media is not as biologically relevant; assumptions in the calculation of TWA concentrations complicate informed risk assessments

). Considering the high cost of pesticide lab analysis, we recognize a trade-off in pesticide monitoring between a comprehensive assessment of multiple matrices, each of which has advantages and disadvantages (Table 1), and the economy of minimizing the number of sample types at the risk of failing to detect important pesticides present in the system but not well represented by the chosen sampling matrix. We hypothesized that (1) the three sampling methods would yield an integrated understanding of pesticide concentrations in the stormwater management ponds and consequently predicted that time-weighted average concentrations of pesticides estimated from o-DGTs would be well correlated with concentrations measured in weekly water sample composites. We also hypothesized that (2) compared to water and biofilm samples, the number and frequency of pesticide detections would be higher in o-DGT samplers due to the increased analytical sensitivity, but that including the water samples and biofilm samples would increase the range of pesticides types (with different physico-chemical properties) detected and provide important insight into potential exposure routes that could help contextualize exposure risk.

2. Methods

2.1. Study Context

We sampled surface water and biofilms in a heavily urbanized area within the Lake Ontario watershed to assess pesticide contamination in stormwater management ponds (see ref. [6]). Due to the high cost of pesticide analysis (the cost of analysis was about CAD 500 per sample per matrix submitted to the Agriculture and Food Lab in Guelph, Ontario), we were constrained to these two matrices. However, co-authors with the federal agency Agriculture and Agri-Food Canada (AAFC) enabled us to deploy o-DGT passive samplers, which AAFC’s lab analyzed in kind. Differences in the analytical methodologies used by these two laboratories constrained our ability to compare the performance of the three sampling methods yet provided an excellent opportunity to compare the value of monitoring the three different matrices with different advantages and disadvantages (Table 1). Throughout this research, we consider the merits of working collaboratively and with different sampling and analytical methods, and we leverage these data sources to make recommendations for pesticide monitoring in urban areas.

2.2. Study Area

Our sampling took place at 21 stormwater management ponds in the City of Brampton, located in Southwestern Ontario, Canada (Figure 1). As part of the Greater Toronto Area, Brampton is one of the most rapidly expanding municipalities in Canada [48] and its highly urbanized watershed represents an ideal location for monitoring urban pesticides in stormwater ponds. The selected ponds were wet ponds, designed to hold water year-round, and were a minimum of 10 years old with no previous dredging within the past 10 years. The surface area of the ponds ranged from 1051 to 3686 m² (mean = 2272 ± 851 m²). The sites represent a gradient of urbanization, with impervious cover within a buffer of 300 m surrounding each pond ranging from 10.9 to 55.2%.

2.3. Sampling

We used three environmental media types (water, biofilms, and o-DGTs) concurrently to assess pesticide contamination, described in the following sections. The survey period took place over the course of nine weeks, from 25 May to 25 July 2022.

2.3.1. Water

We followed the methods described in a previously published paper [6]. Briefly, we created a 1.35 L time-integrative composite sample composed of nine 150 mL weekly water grab samples taken at 20 cm depth using an HDPE polyethylene bottle, which was kept frozen at −20 °C until submission for pesticide analysis. Constrained by the high costs of pesticide analysis, we used a temporal composite rather than keeping the instantaneous grab samples of water separate and analyzing each independently. The effect of the holding period (i.e., time between sampling and extraction) on pesticide degradation in water samples is minimal when stored frozen, compared to storage at above-zero temperatures [49].

2.3.2. Biofilm

The sampling of biofilm also followed the methods previously described [6]. To culture biofilms in situ, we deployed floating biofilm sampler racks, holding ten acrylic plates (20.2 cm by 44.4 cm), in each pond, secured with anchors (Figure 2). After submersion for 54 days, we harvested the biofilm by scraping both sides of each plate into a composite container for each site, which was subsequently stored frozen at −20 °C. We freeze-dried 300–500 g (wet weight) of each composite biofilm sampling using a FreeZone I Labconco benchtop freeze dryer (Labconco Corporation, Kansas City, MO, USA) prior to pesticide analysis. Insufficient biofilm mass from two of the SWP sites meant these were only analyzed for phenoxy herbicides and polar pesticides, as these screens contained the highest number of analytes. All other stormwater pond sites (n = 19) had adequate biofilm mass to analyze for all the intended pesticide screens.

2.3.3. o-DGT Samplers

At each pond, we deployed two o-DGT (DGT Research^®; LSND-AT with PTFE membranes; exposed area = 3.14 cm²) duplicate samplers mounted on custom-built holders (Figure 3), which were attached to the floating biofilm sampler rack. Each set was replaced after 27 days of deployment, for a total of 54 days between the two sets, to match the duration of biofilm and water sampling. To detect any potential contamination from handling, we used a field travel blank for each day of deployment and collection (n = 7). Following collection, o-DGTs were removed from their holder, wrapped in clean aluminum foil, and stored frozen at −20 °C.

Prior to analysis, we disassembled each o-DGT casing and transferred the gel membranes into 25 mL Falcon tubes. We then performed sequential methanol extractions into 15 mL polypropylene centrifuge tubes (Corning reference 430052; see [50] for detailed method). The supernatant extracts were evaporated to dryness using a Buchi Multivapor™ P-6. The dried residue was reconstituted with 400 µL of 80:20 methanol/water, and transferred to an amber glass high-performance liquid chromatography (HPLC) vial with glass insert for analysis.

2.4. Pesticide Analysis

Water and biofilm samples were submitted to the Agriculture and Food Lab (AFL) of the Laboratory Service Division at the University of Guelph, Ontario (ISO/IEC 17025 accredited). Water and biofilm samples were analyzed for 542 current-use and legacy pesticides using high-performance liquid chromatography paired with electrospray ionization and tandem mass spectrometry (HPLC-ESI-MS/MS) and gas chromatography paired with tandem mass spectrometry (GC–MS/MS). Details of this analysis are described in the Supplementary Materials, and a list of analytes and associated method detection limits (MDLs) and method quantification limits (MQLs) can be found on the FigShare repository, linked in the Supplementary Materials.

Extracts from the o-DGT samplers were analyzed by the Agriculture and Agri-Food Canada (AAFC) chemical laboratory in London, ON, Canada. Samples were analyzed for 491 chemical contaminants by HPLC-ESI-MS/MS using a Vanquish-Duo HPLC coupled to a Thermo Altis triple quadrupole mass spectrometer. Details of this analysis are described in the Supplementary Materials. Target analytes, along with their limits of detection (LODs), limits of quantification (LOQs), multiple reason monitoring (MRM) transitions, and retention times are available in the FigShare repository, linked in the Supplementary Materials.

2.5. Calculation of Detection Frequencies on Common Set of Analytes

To appropriately compare pesticide occurrences among the three matrices analyzed by two different chemical laboratories, we created a common set of analytes composed only of compounds targeted by both laboratories. Among the 542 compounds screened for by AFL and 491 screened for by AAFC, 238 were screened for by both labs.

We calculated pesticide detection frequencies (%) as the number of detections divided by the number of stormwater pond sites (i.e., 21). For o-DGT samplers, we define a detection as the occurrence of a compound above the detection limit in at least one sampler out of the two duplicates deployed in each pond for each deployment period. As the detection frequency represents the entire sampling period (two deployment periods), if a compound is detected in one deployment period, but not the other, we still define the occurrence as one detection. If a compound is detected in more than one sampler in a pond or across both deployment periods, the overall occurrence of that compound is still defined as one detection.

2.6. Calculation of Time-Weighted Average Concentrations in o-DGT Samplers

2.6.1. Modeling of Diffusion Coefficients

We used the Hayduk–Laudie diffusion model (Equation (1)) to estimate the diffusion coefficients of pesticides [51].

$${\mathrm{D}}_{\mathrm{w}}={\displaystyle \frac{1.326\times {10}^{-4}}{{\mathrm{ƞ}}^{1.14}{\mathrm{V}}^{0.589}}}$$

(1)

where D_w is the diffusion coefficient (cm² s⁻¹); η is the viscosity of water (centipoise) at the average temperature of pond water (24.28 °C); and V is the molar volume (cm³ mol⁻¹) calculated from the structure of each compound.

2.6.2. Sampling Rates

Sampling rates for each pesticide analyte in the o-DGT samplers were calculated using Equation (2).

$${\mathrm{R}}_{\mathrm{s}}={\displaystyle \frac{{\mathrm{D}}_{\mathrm{w}}\mathrm{A}}{∆\mathrm{g}}}$$

(2)

where R_s is the sampling rate (cm³ s⁻¹); D_w is the diffusion coefficient (cm² s⁻¹) of the analyte; A is the exposed surface area of the o-DGT (3.14 cm²); and ∆g is the thickness of the o-DGT diffusive gel (0.094 cm).

2.6.3. Time-Weighted Average Concentrations

Any quantity of analyte detected in a field blank was first subtracted from its paired sample prior to calculating concentrations. Equation (3) was used to determine the time-weighted average (TWA) water concentrations (C_w).

$${\mathrm{C}}_{\mathrm{w}}={\displaystyle \frac{\mathrm{M}}{{\mathrm{R}}_{\mathrm{s}}\mathrm{t}}}$$

(3)

where C_w is the estimated concentration in water (ng L⁻¹); M is the mass of analyte on the o-DGT sampler (ng); R_s is the sampling rate (L d⁻¹); and t is the deployment time (27 d).

2.6.4. Method Detection Limits

We converted the limits of detection (LODs) to method detection limits (MDLs) for pesticide analytes in o-DGT samplers using Equation (3), with M representing the LOD (pg per sampler). We used these MDL values (in µg L⁻¹) for comparison with the MDL values for the water samples (Table S1).

2.7. Statistical Analysis

Statistical analyses were completed with R Statistical Software (version 2.2.2) in RStudio (version 2023.06.01 [52]). To compare the concentrations of pesticides detected in the water samples and o-DGTs, we used the lm function in base R to perform linear regressions between the concentrations in water and the TWA concentrations calculated from o-DGT samplers. We were not able to investigate relationships between concentrations in biofilms and o-DGTs due to a lack of data > MQL in the biofilm samples. We only investigated pesticides which were present above the MQL at least twice in the water samples; thus, we investigated 5 of the 11 compounds detected in both water samples and o-DGTs: flupyradifurone, imidacloprid, mecoprop, 2,4-D, and MCPA. We log transformed the water concentrations for flupyradifurone and imidacloprid because they appeared to have a saturation effect in the o-DGTs. To obtain a singular data point representative of the four o-DGTs deployed in each pond, we averaged the concentrations measured across both deployments in each pond to represent average conditions over the same period as the biofilm sampler deployment and the composite water sampling for each pond. We only included data points where the compound was detected in at least one of the two matrices being assessed; sites with no detections in both matrices were excluded. If a compound was detected at a site in one matrix but not in the other, we replaced the non-detect with ½ the MDL value to rule out type II errors. For detections below the MQL, we used the MDL value of that compound to avoid overestimating the concentration. We evaluated the model fits of these regressions by assessing the adjusted r² values and the statistical significance (p < 0.05). We created plots (Figure 4) where each concentration is represented as a point, with the linear regression represented by a solid (significant) or dashed (not significant) line (R packages ggplot2, patchwork).

3. Results and Discussion

3.1. Pesticide Detections

We detected 82 pesticides across the 3 sampling matrices (

Table 2. Summary of the number of pesticides out of the 82 we detected in total that were detected by each sampling matrix: either in water, biofilm, o-DGT samplers, or in all three sample types combined. The values on the diagonal represent the number of pesticides detected exclusively by that sampling matrix, whereas the values in the triangular matrix represent the number of pesticides detected in common by the pair of sampler types indicated by the column and row headings.

	Water	Biofilm	o-DGT	All 3 Matrices
Water	3			5
Biofilm	0	1
o-DGT	6	3	64

): 14 pesticides were detected in water samples, 9 in biofilm samples, and 78 in o-DGT samplers (see also Table S2). The total number of pesticides detected at each site ranged from 22 to 52 pesticides, with an average of 36 pesticides detected per site. There were 17 pesticides detected at all sites and 31 pesticides were detected in at least half of the sites (

Table 3. Detection frequencies (%) of pesticides in water, biofilm, and o-DGT samples. n.d. = not detected or concentration was below MDL. Detection frequency is number of detections divided by number of sites (n = 21) multiplied by 100%. Pesticides are listed in order from highest to lowest detection frequency. Note that detection limits vary by matrix and analytical lab.

	Detection Frequency (%)
Pesticide	Water	Biofilm	o-DGT
2,4-D	100	14	90
atrazine	n.d.	n.d.	100
azoxystrobin	n.d.	43	100
carbendazim	5	10	100
chlorantraniliprole	n.d.	n.d.	100
clomazone	n.d.	n.d.	100
fluopyram	n.d.	n.d.	100
MCPA	100	5	43
mecoprop	100	n.d.	95
metalaxyl	n.d.	n.d.	100
metolachlor	n.d.	n.d.	100
propazine	n.d.	n.d.	100
propiconazole	n.d.	n.d.	100
simazine	n.d.	n.d.	100
tebuconazole	n.d.	19	100
tebufenozide	n.d.	n.d.	100
thiabendazole	n.d.	5	100
diuron	5	19	95
imazethapyr	n.d.	n.d.	95
prometon	n.d.	n.d.	95
triclopyr	95	n.d.	n.d.
clothianidin	33	n.d.	90
flupyradifurone	38	n.d.	90
paclobutrazol	n.d.	n.d.	90
imidacloprid	62	n.d.	86
dimethenamid	n.d.	n.d.	81
ametryn	n.d.	n.d.	71
tebuthiuron	n.d.	n.d.	71
difenoconazole	n.d.	n.d.	67
carbaryl	n.d.	n.d.	52
epoxiconazole	n.d.	n.d.	52
hexazinone	n.d.	n.d.	48
myclobutanil	n.d.	n.d.	48
cyantraniliprole	n.d.	n.d.	38
sulfentrazone	n.d.	n.d.	38
acetamiprid	n.d.	n.d.	33
diazinon	n.d.	n.d.	33
ethiofencarb	n.d.	n.d.	33
benalaxyl	n.d.	n.d.	29
metribuzin	n.d.	n.d.	29
pyrimethanil	n.d.	n.d.	29
bentazon	24	n.d.	14
bromoxynil	n.d.	n.d.	24
indaziflam	n.d.	n.d.	24
dichlorprop	19	n.d.	5
dithiopyr	n.d.	n.d.	19
flonicamid	19	n.d.	n.d.
flufenoxuron	n.d.	n.d.	19
flutriafol	n.d.	n.d.	19
prometryn	n.d.	n.d.	19
pyridate	n.d.	n.d.	19
spiroxamine	n.d.	n.d.	19
dimethomorph	n.d.	n.d.	14
fenpropimorph	n.d.	n.d.	14
pyridaben	n.d.	n.d.	14
terbuthylazine	n.d.	n.d.	14
terbutryn	n.d.	n.d.	14
trifloxystrobin	n.d.	n.d.	14
picolinafen	n.d.	n.d.	10
pyraclostrobin	n.d.	n.d.	10
pyriproxyfen	n.d.	n.d.	10
spiromesifen	n.d.	n.d.	10
alanycarb	n.d.	n.d.	5
benfuracarb	n.d.	n.d.	5
benzoximate	n.d.	n.d.	5
bifenazate	n.d.	n.d.	5
bromacil	n.d.	n.d.	5
bromuconazole	n.d.	n.d.	5
bupirimate	n.d.	n.d.	5
butafenacil	n.d.	n.d.	5
chlorpropham	5	5	5
etoxazole	n.d.	n.d.	5
fenazaquin	n.d.	n.d.	5
fenobucarb	n.d.	n.d.	5
fluazifop-p-butyl	n.d.	n.d.	5
fludioxonil	n.d.	n.d.	5
hexaconazole	n.d.	n.d.	5
imazapyr	5	n.d.	n.d.
indoxacarb	n.d.	n.d.	5
methiocarb	n.d.	n.d.	5
ofurace	n.d.	n.d.	5
picoxystrobin	n.d.	n.d.	5
propoxur	n.d.	n.d.	5
quizalofop-ethyl	n.d.	n.d.	5
simetryn	n.d.	n.d.	5
tetraconazole	n.d.	n.d.	5
thiacloprid	n.d.	n.d.	5
thiophanate-methyl	n.d.	5	n.d.

).

Many of the most commonly detected pesticides in our stormwater ponds were also widely detected elsewhere in urban surface water studies. Seventeen pesticides were detected in all of our stormwater pond sites: 2,4-D, atrazine, azoxystrobin, carbendazim, chlorantraniliprole, clomazone, fluopyram, MCPA, mecoprop, metalaxyl, metolachlor, propazine, propiconazole, simazine, tebuconazole, tebufenozide, and thiabendazole. Of these 17 pesticides, all but 6 were previously detected in urban stormwaters by other studies in Canada or the United States [14,15,53,54,55]. These six exceptions were chlorantraniliprole, clomazone, fluopyram, metalaxyl, tebufenozide, and thiabendazole. Chlorantriliprole and fluopyram are both relatively new pesticides, and clomazone, metalaxyl, tebufenozide, and thiabendazole are more commonly associated with agricultural runoff [24,56,57,58,59]. Ubiquitous and other commonly detected pesticides will be discussed in greater detail in a forthcoming publication. Identifying commonly occurring pesticides in stormwaters is important to gain a better understanding of common sources, landscape drivers, and the potential toxicity of pesticide contamination in urban areas [14].

3.1.1. The Influence of Analytical Sensitivity on Pesticide Detections

Overall, the o-DGT samplers detected substantially more pesticides and had more overlapping detections with the other matrices than the water or biofilm samples. Notably, pesticides that were widespread in o-DGTs were not frequently detected in the other matrices. Of the fourteen pesticides with a detection frequency of 100% in o-DGTs, only four were also detected in one or both other matrices. Moreover, 64 pesticides were detected only in o-DGT samplers (Table 2), and would have been missed without the use of this passive sampling device. Yet we must question whether these results are due to the sensitivity of o-DGT devices themselves, or due to differences in analytical sensitivities.

Pesticide analyses took place at two different analytical laboratories utilizing different analysis techniques and equipment. Prior to considering the drivers of differences in detections among the three matrices, we first inquired into whether the analytical differences may have contributed to the results. To do so, we compared the MDLs of pesticides in water and o-DGT samplers. Unfortunately, we were not able to include biofilm MDLs in this investigation on account of the discrepancy in units, as these are measured on a per mass basis (µg kg⁻¹) whereas the MDLs for water samples and o-DGTs are measured on a per volume basis (µg L⁻¹). The MDLs for analysis in water samples were 1 to 6 orders of magnitude higher than the MDLs for analysis in o-DGTs (Table S1). For example, atrazine, metolachlor, propazine, and tebuconazole all had detection frequencies of 100% in o-DGTs and 0% in the water samples (Table 3). The MDLs for these four pesticides were five orders of magnitude higher in water samples than the MDLs of o-DGTs (Table S1). It is therefore plausible that the differences in detection limits between the water and o-DGT samples are highly influential in the resulting differences in pesticide detections.

To explore the potential magnitude of impact that the analytical sensitivity had on pesticide detection, we standardized by removing detections in the o-DGT samplers that were below the MDL used for the analysis of water samples. As shown in Table S3, only 12 of the 78 pesticides detected in o-DGTs would have been detected if the o-DGT analytical detection limits were as high as those used for water analysis. When differences in analytical sensitivities are factored out, we can see that more compounds were detected in water samples (n = 14) than in o-DGTs (n = 12) and at higher detection frequencies. This begs the question, if the detection limits for analyzing water samples were as low as those in the o-DGTs (down to the ng L⁻¹ or pg L⁻¹ level), what pesticides would have been detected, and at what frequency? Although o-DGTs have an inherent sensitivity because they accumulate contaminants over time, therefore sampling a greater volume of water than the grab water sampling, the analytical conditions are evidently influential on pesticide detections. Without the relatively low detection limits used for the analysis of o-DGTs, our sampling would have missed the majority of pesticides present in the stormwater ponds. We can resolve that monitoring programs aiming to characterize the presence of pesticides could benefit substantially from allocating money and resources into lowering the detection limits of their analytical methods as we infer that the majority of pesticides in urban aquatic systems are present at trace levels. However, programs concerned with assessing hazards arising from elevated contaminant concentrations would not need to quantify concentrations at such low levels and would alternatively benefit from allocating resources toward increasing the temporal resolution of water grab sampling to more accurately describe contaminant exposure.

In addition to these important analytical differences, other factors may have contributed, albeit less significantly, to the differing detections in water and o-DGT samples. Pesticides may be undetected in water samples because of their physico-chemical properties (e.g., hydrophobicity) or because at the time of grab sampling, they were not present or occurred at trace levels at which the method (i.e., the volume of water and the approach of analysis) was not sensitive enough to detect them. The highly dynamic nature of water and contaminant influxes from the stormwater system means that instantaneous water grab sampling may miss sporadic pulses, which may have short yet potentially serious effects on exposed biota [28,60]. The apparent absence in water samples of compounds detected in o-DGTs could imply that pulses in pesticide occurrence took place out of sync with our weekly sampling regime. It is important to note that our water sampling approach is atypical: traditional sampling involves the submission of one grab sample, whereas our submissions were composed of a time-integrative composite sample of nine weekly grab samples. In either case, passive sampling may be preferred for detecting pesticides whose occurrences are too episodic to detect with instantaneous sampling, such as those entering stormwater ponds (discussed further in Section 3.3).

Biofilm sampling is also a form of passive sampling, yet there are a few key differences with o-DGTs that may explain their differing detections, in addition to an assumed difference in instrumental detection limits between the analyses of the two matrices, which is likely the key driver of these results. Even with standardized biofilm samplers, biofilms are as variable as any living being: their growth and composition are influenced by the conditions of their surroundings [61], which are also variable across sites [6]. Biofilms are subject to grazing, and new growth throughout the growing season may result in a dilution effect. Biofilms may also support the microbial degradation of pesticides [62]. Although some biofouling can occur on the o-DGTs, it was noted that such fouling does not have significant effects on measuring organic contaminants in water [63,64]. Although both biofilm samplers and o-DGT samplers were deployed at the same time, the commencement of the accumulation of contaminants within biofilms is delayed because biofilm colonies must first establish, whereas o-DGTs have a defined mass of binding material that can start collecting contaminants upon immersion. Conversely, the large surface area of the biofilm samplers (837 cm² per side per slide) can increase the opportunities for interactions with compounds, compared to the relatively small surface area of o-DGTs (3.14 cm² each), although our results suggest this factor does not necessarily increase sensitivity. Finally, binding and transport processes of chemicals differ between the two matrices, with binding and transport within the EPSs of biofilms likely more variable than within the standardized, well-characterized binding materials of the o-DGTs.

3.1.2. Sampling Pesticides with Three Different Environmental Matrices

Optimizing contaminant monitoring techniques requires the assessment and comparison of currently used methods. Yet, research comparing the performances of biofilms, passive sampling devices, and water grab samples is uncommon, and we do not know of any existing literature comparing DGT-type passive samplers to biofilm sampling. Rheinheimer et al. [36] compared biofilm and water sampling with polar organic compound sampling devices (POCIS; a passive sampling device that does not contain a semi-permeable membrane or diffusive gel layer [65]) in agriculturally affected rivers in Brazil, finding that the combined use of biofilms and POCIS allowed for a more holistic characterization of pesticide contamination compared to water grab sampling. Biofilms detected several contaminants entirely absent in POCIS samplers and were better able to describe contamination on a smaller scale. The authors concluded that because pesticides encompass a range of differing molecular properties (e.g., some are highly soluble in water while others are insoluble), integrating different types of matrices helps to address the variety of compound–matrix interactions (e.g., adsorption, accumulation, mineralization) that are specific to each compound [36].

Our results differ from the aforementioned study as the contribution of detections in biofilm samples to the overall contamination characterization was not as pronounced. This could be explained by the substantial differences in detection limits among the sampling approaches. The o-DGTs in our study, aided by relatively low MDLs, far outperformed the other matrices in terms of sensitivity, and all but four of the compounds found in water and biofilm samples were also detected by the o-DGTs. We discuss these four exceptions below.

Flonicamid, imazapyr, and triclopyr were only detected in water samples. These pesticides have a relatively high solubility in water and a low log octanol–water partition coefficients (Log K_ow) and effective log octanol–water distribution coefficients (Log D_ow), the latter of which accounts for the ionizing behavior of a pesticide in water (Table S4; see [66]). Compounds with low Log D_ow values are less likely to be sorbed by the o-DGT samplers [66]. However, a study by Morin et al. [67] found that the uptake behavior of organic micropollutants into passive samplers can be variable, often specific to the compound, and not consistently predicted by physico-chemical properties. Flonicamid is a selective systemic pyridine insecticide used to control greenhouse pests, such as aphids [68], and would have been missed at 19% of pond sites without time-composited weekly water sampling. Imazapyr is a nonselective imidazolinone herbicide used to control broad-leaf weeds, perennial grasses, and woody plants in turfgrass, urban forestry, ponds and wetlands, and industrial sites. It is the only pesticide approved for use in Canada over standing water to control invasive wetland plants, such as Phragmites australis [69]. Imazapyr was only detected at one site. Triclopyr is a selective pyridine herbicide used to control broad-leaf and woody weeds in uncultivated areas, such as on industrial sites, turfgrass, or rights-of-way [68]. Triclopyr was widespread in water samples (Table 3), all of which would be missed without water sampling. The apparent absence of these three pesticides in o-DGTs, despite the enhanced detection limits in the analysis of o-DGTs, suggests there may benefits to including multiple matrices in pesticide monitoring depending on the objective and scope of the study.

Thiophanate-methyl was detected only in biofilm samples. This compound has low solubility, which may explain its apparent absence in water samples, and its low Log D_ow may explain its absence in o-DGTs (Table S4). It is a parent compound of carbendazim, which was detected in o-DGTs at all sites, suggesting it is only present momentarily prior to its rapid breakdown in water via hydrolysis [70]. Thiophanate-methyl is a carbamate fungicide used to control fungal diseases in turfgrass and greenhouses and is also used in veterinary medicines [68]. However, this fungicide was only detected at one site. In our study, biofilms did not contribute significantly to the overall detection of contaminants, and did not detect notable contaminants that were missed by other matrices, unlike the findings of Rheinheimer et al. [36]. However, biofilms provide a unique and biologically relevant insight into contaminant bioconcentration and fate, as well as an alternative route of entry for contaminants into the food web. This is discussed further in Section 3.3.

3.2. Pesticide Concentrations

Of the 82 total pesticide detections, 10 of 14 detections were quantified in water, 4 of 9 detections were quantified in biofilm, and 76 of 78 detections were quantified in o-DGTs (

Table 4. Minimum and maximum pesticide concentrations in water, biofilm, and o-DGT samples. Concentrations are reported in ug L-1 to ease comparisons with ecotoxicological thresholds, which are commonly reported in same units. Notation of “<MQL” refers to pesticide that was detected (i.e., >MDL) but concentration was lower than MQL.

	Water (µg L−1)		Biofilm (µg kg−1)		o-DGTs (µg L−1)
Pesticide:	Min	Max	Min	Max	Min	Max
2,4-D	0.0036	0.87	<MQL	<MQL	0.0002	0.019
acetamiprid					<MQL	0.00016
ametryn					<MQL	0.0011
atrazine					0.0056	0.087
azoxystrobin			<MQL	35	0.000016	0.022
benalaxyl					<MQL	0.000019
bentazon	<MQL	<MQL			0.000022	0.000031
bromacil					<MQL	0.0021
bromoxynil					0.00031	0.00089
bromuconazole					0.00024	0.00024
bupirimate					0.000075	0.000075
carbaryl					0.00029	0.025
carbendazim	0.52	0.52	<MQL	<MQL	0.000026	0.019
chlorantraniliprole					0.000036	0.01
chlorpropham	<MQL	<MQL	33	33	0.44	1.36
clomazone					0.00011	0.00074
clothianidin	<MQL	<MQL			0.00024	0.0011
cyantraniliprole					0.0004	0.0011
diazinon					0.00042	0.00044
dichlorprop	0.0022	0.028			0.00097	0.00098
difenoconazole					0.000039	0.043
dimethenamid					0.00017	0.00038
dimethomorph					<MQL	0.00023
dithiopyr					<MQL	0.00035
diuron	<MQL	<MQL	<MQL	23	0.00014	0.423
epoxiconazole					<MQL	0.00063
ethiofencarb					0.00017	0.0009
etoxazole					0.000056	0.000056
fenazaquin					0.00029	0.00029
fenobucarb					<MQL	0.000072
fenpropimorph					0.00031	0.00046
flonicamid	<MQL	0.014
fluazifop-p-butyl					<MQL	<MQL
fludioxonil					0.0013	0.004
flufenoxuron					0.0021	0.0035
fluopyram					0.00095	0.0043
flupyradifurone	<MQL	0.014			0.00026	0.01
flutriafol					0.00067	0.0017
hexaconazole					0.00016	0.00016
hexazinone					0.000032	0.00016
imazapyr	0.0089	0.0089
imazethapyr					<MQL	0.000054
imidacloprid	<MQL	0.012			0.00031	0.0084
indaziflam					0.00018	0.00025
indoxacarb					<MQL	<MQL
MCPA	<MQL	1.2	<MQL	<MQL	<MQL	0.0077
mecoprop	<MQL	0.47			0.000054	0.01
metalaxyl					0.000038	0.0023
methiocarb					0.000086	0.000086
metolachlor					0.00073	0.032
metribuzin					0.00062	0.0024
myclobutanil					0.00038	0.0011
ofurace					0.0007	0.0007
paclobutrazol					0.000068	0.0026
picolinafen					0.00065	0.0011
picoxystrobin					0.000043	0.000043
prometon					<MQL	0.00026
prometryn					0.000033	0.00009
propazine					0.000035	0.00044
propiconazole					0.00042	0.049
propoxur					0.00037	0.00062
pyraclostrobin					0.000053	0.00014
pyridaben					0.0005	0.007
pyridate					0.00094	0.01
pyrimethanil					0.00015	0.0028
pyriproxyfen					0.000023	0.00025
quizalofop-ethyl					0.0012	0.0012
simazine					0.00014	0.0015
Simetryn					<MQL	0.000042
spiromesifen					0.00074	0.003113124
spiroxamine					0.000016	0.000039
sulfentrazone					0.00014	0.0096
tebuconazole			<MQL	29	0.0051	1.723
tebufenozide					0.00064	0.0039
tebuthiuron					<MQL	0.00059
terbuthylazine					0.00039	0.00074
terbutryn					<MQL	0.0002
tetraconazole					0.00064	0.00064
thiabendazole			<MQL	<MQL	0.000032	0.02
thiacloprid					0.000055	0.00013
thiophanate-methyl			<MQL	<MQL
triclopyr	<MQL	0.065
trifloxystrobin					<MQL	0.000082

). We compared concentrations in water samples and estimated TWA concentrations in o-DGT samplers to the US EPA’s Aquatic Life Benchmarks (ALB [71]) and found two instances of exceedances. The maximum concentration in water samples of imidacloprid, 0.012 µg L⁻¹, surpassed the ALB of 0.01 µg L⁻¹ (EC₅₀) for chronic exposure to freshwater invertebrates. The maximum TWA concentration of diuron estimated from o-DGTs (0.42 µg L⁻¹) surpassed the ALB for acute exposure to vascular aquatic plants (0.13 µg L⁻¹; IC₅₀) and was within the same order of magnitude as the ALB for chronic exposure to freshwater invertebrates (0.83 µg L⁻¹; EC₅₀). In several other instances, maximum concentrations either measured in water samples or estimated in o-DGT samplers approached but did not surpass benchmarks or guidelines. For example, maximum concentrations of carbaryl (0.025 µg L⁻¹), clothianidin (0.0011 µg L⁻¹), difenoconazole (0.043 µg L⁻¹), pyridaben (0.007 µg L⁻¹), and tebuconazole (1.7 µg L⁻¹) were within one order of magnitude from their respective ALBs for chronic exposures to freshwater invertebrates (0.5, 0.05, 0.86, 0.044, and 11 µg L⁻¹, respectively). We were unable to systematically compare concentrations in the biofilm samples to appropriate thresholds for dietary exposure due to the lack of available information for this type of exposure. Given the diversity and large number of pesticides detected at each site, an evaluation of mixture toxicity would be necessary to assess the risk of toxicity to aquatic organisms living in stormwater ponds or in aquatic habitats receiving stormwater runoff.

The extrapolation of concentration data from o-DGT samplers to inform risk assessment is challenging because it provides a TWA of the chemical concentration in the water during the exposure period. It is well-known that stormwater systems experience dynamic fluctuations in water quality and contaminant loads [2,72], where relatively short pulses of relatively high concentrations of contaminants can enter into the water body where the o-DGTs were deployed. The calculation of a TWA concentration of exposure over the course of the deployment period from the quantities of contaminants present in the o-DGTs does not provide insight into the frequency, duration, or magnitude of any spikes in concentration. This creates a scenario where the TWA concentrations may not exceed a threshold of toxicity, but pulses could have exceeded that threshold, resulting in adverse effects to biota. When conducting a risk assessment, using TWA concentration data from o-DGTs could result in type II errors (i.e., false negatives, where a risk is predicted to be absent when in fact it may exist). However, this same limitation exists for grab samples of water and biofilm sampling as a measure of exposure to biota in surface water.

Pesticide Concentrations in Water Samples vs. o-DGTs

We predicted that if a compound is detected only in o-DGTs, the calculated TWA concentration in water should be lower than the method detection limit (MDL) for water samples. Of the 67 compounds that were detected in o-DGTs but not in water (Table 3), 63 had maximum concentrations in o-DGTs that were lower than the MDLs for water samples. This suggests that the o-DGTs are able to detect pesticides at lower concentrations than water grab samples due to their lower detection limits and the time-integrated accumulation of pesticide compounds from the surrounding water. The exceptions were atrazine, difenoconazole, metolachlor, and tebuconazole. Both the maximum TWA concentration in o-DGTs and the MDL values for water samples were in the same order of magnitude for atrazine, difenoconazole, and metolachlor; however, the maximum concentration of tebuconazole (1.723 µg L⁻¹) is much higher in the o-DGTs than the MDL for water (0.1 µg L⁻¹). This suggests that a pulse of tebuconazole was missed by water sampling but captured by the o-DGTs.

We also predicted that we would observe a positive relationship between the concentration of contaminants measured in the water, and their measured quantities in o-DGTs. We expected that pesticides that are not detected in water samples appear to be absent because they are either present at very low concentrations (below the detection limits) or not present at all. In this case, the more sensitive o-DGT samplers (with relatively lower detection limits) should have correlated quantifications with the water samples. Of the five pesticides we investigated (flupyradifurone, imidacloprid, mecoprop, 2,4-D, and MCPA; Figure 4), all had a significant positive relationship (p < 0.05) with a relatively high coefficient of determination (adjusted r-squared > 0.60; Table S5). We can reasonably conclude here that the detection of these five pesticides in o-DGTs was likely due to the increased sensitivity from lower detection limits or from time-integration. The significant fit of the linear regressions on the log-transformed concentrations of flupyradifurone and imidacloprid could indicate a saturation effect in o-DGTs at higher concentrations; however, previous work has shown that effective binding capacities are rarely reached during deployment periods of up to 43 days, even for high contaminant concentrations [45]. The uptake of chemicals from water into passive samplers is known to be variable; compounds can accumulate linearly, with an inflection point, randomly, or not at all [67], and we require more data points to properly test these relationships as we are constrained by the majority of concentrations being near the detection levels. It must be stressed that clarification of these relationships is also difficult due to the differences in analytical sensitivity, small sample size, and because differences in pesticide behavior in biofilms may be compound specific; all are areas requiring further investigation.

3.3. Practicality and Reproducibility

3.3.1. Variability in o-DGT Duplicates

An advantage of sampling with o-DGTs is the ability to easily incorporate duplication to increase the reliability of detections (biofilms, on the other hand, require substantially more surface area to collect adequate biomass for chemical analysis). We predicted that due to their identical deployment and exposure conditions, duplicates should have a high level of agreement in both the detection and quantification of compounds. Here, we summarize our findings on the variance between duplicates deployed at the same instance at the same sites. Importantly, this is not the variance between two deployment periods at the same site.

We found relatively high levels of agreement in the detection of pesticides across the duplicate o-DGTs (average detection agreement ranged from 65 to 100% between duplicates, with an average agreement of 81 ± 8.7%); these results are encouraging and may suggest high reproducibility for presence–absence analysis (Table S6).

We expected that time-integration would also result in good correlations between concentrations in the duplicates, especially with their close spatial proximity and matching deployment orientation; however, the large range in concentration agreement (mean concentration differences ranged 3.16–24,290 pg/sampler, averaging 676 ± 3115 pg/sampler) may signify uncertainty in the reliability of concentration measurements (Table S7). The coefficient of variance (COV), representing the between duplicate variability in target analyte recovery, averaged 86 ± 74%, and for pesticides detected in both duplicates, COV values ranged from 7 to 295%. Also, variation between duplicates did not decrease with an increase in the mass of analyte present (Table S8), although at such trace levels, uncertainty around concentrations makes sense as even the tiniest difference in dilution volume or exposure history can have a relatively large impact on concentration. As suggested by Challis et al. [43], uptake variability may be caused by physical interference of particles settling from the water column, and it is apparent that stormwater ponds are subject to relatively high concentrations of suspended sediments ([73]; also see Figure 2). The relatively small surface area of the exposed face of o-DGTs may also be a contributing factor [50]. In either case, deploying replicate samplers at a site is clearly important to characterize the variation in resultant concentrations measured across o-DGTs.

3.3.2. Limitations and Recommendations for Pesticide Monitoring in Urban Areas

This study aimed to improve the understanding of urban pesticide contamination by assessing the performances of three sampling approaches and to inform how pesticide monitoring could be implemented in urban areas of the Great Lakes watershed. To our knowledge this is the first study to report on the performance of o-DGT samplers in stormwater ponds for the detection of pesticide contamination. Although the o-DGT samplers clearly detected more pesticide compounds and had lower limits of detection than our composite water samples or biofilm samples, we cannot dismiss the value of water and biofilm sampling. Each matrix has benefits and limitations, and each sampling approach has different purposes for which they are suited.

Instantaneous water sampling is likely insufficient for describing such temporally complex pollutant fluxes as seen in urban stormwater systems [74,75]. Yet, the sample volume required for analysis (about 1 L per site for all three pesticide screens) is easily achieved, and low equipment and labor costs can allow for more samples to be taken across a wider spatial or temporal scale (e.g., the use of automatic high-frequency water sampling [76]). Importantly, direct comparisons in concentrations can be made with other studies and with ecotoxicological risk thresholds and guidelines, of which most report exposure in terms of the concentration in the surrounding water (e.g., refs. [77,78]).

Biofilms are important integrators of pollutants, and the biofilm sampling approach is ecologically relevant, giving insight into potential exposures through the food chain [35]. Compared to o-DGT samplers (which are not consumed), biofilms represent a direct dietary exposure pathway and a more realistic route of entry for contaminants [39]. However, we currently have very little understanding of the impacts of pesticides through dietary exposure routes in aquatic ecosystems, which limits our ability to draw comparisons to other research or to ecotoxicological risk guidelines. Additionally, the biofilm community can be variable—with different compositions even under the same environmental conditions [79]—thus, its standardization can be difficult to achieve from a monitoring perspective, although this may be irrelevant if the focus is on dietary exposure characterization. The surface area required to meet the mass requirements for chemical analysis can make the duplication of biofilm samplers much more difficult. For example, despite the total surface area of 1.67 m² for each biofilm sampler, adequate biomass for analysis was often difficult to obtain. The total amount of biofilm harvested at each site ranged widely from as little as 0.0018 g cm⁻² to 0.098 g cm⁻², and due to the high water content of most biofilms, between 300 and 500 g of wet biofilm was needed to meet the required 4–6 g of freeze-dried biofilm to conduct all three pesticide screens. The construction and deployment of biofilm samplers was time-consuming in our case because each sampler was made by hand and assembled on site (we are not currently aware of adequately-sized commercialized biofilm samplers for deployment in lentic systems). The harvesting of biofilm was also more labor intensive than water sampling or o-DGT collection; for example, at each site, we spent 45–60 min thoroughly scraping the 10 acrylic plates on both sides to obtain our composite sample, whereas the collection of water samples or o-DGTs took less than 5 min. However, both passive sampling techniques (biofilm and o-DGT) reduce the number of required site visits, enabling time-integration without added field labor costs.

What o-DGTs lack in biological relevance, they make up for in sensitivity. Based on our findings, these devices are capable of concentrating compounds present at trace levels across a variety of molecular properties to allow for their detection and quantification otherwise missed by other sampling approaches. This is supported by other similar work with o-DGTs for monitoring pesticides [43,47]. o-DGTs are highly applicable in urban settings due to the wide variety of contaminant types they can detect [80] and allow for a more holistic picture of the mixture of chemicals present. o-DGTs allow for field blanks and duplication, both of which are either not possible or more effort to conduct in other methods. However, given the high degree of variability in concentration between duplicate samplers, there remains uncertainty in pesticide compound quantification extrapolated from o-DGTs. Inferences about pesticide occurrence (i.e., presence/absence) are much more reliable than inferences about pesticide concentrations. An additional trade-off with time-integrated sampling is that because pulses are averaged over the exposed period, extremes are not captured in an interpretable way, and this may lead to inaccurate perceptions of ecological risk.

Analytical differences drove an increase in sensitivity to pesticide detection in the o-DGT samplers compared to water and biofilm samples. Analytical methods are critically important for determining the profile of pesticide contamination because so many pesticides existed at trace levels. For the effective characterization of pesticide contamination, programs clearly need research-grade methods with lower detection limits (i.e., in the ng L⁻¹ range) to detect what pesticides are present. Because most pesticides detected in our stormwater ponds were detected by o-DGTs, the costs of analyzing more than one matrix seem to outweigh the benefits of the additional insight provided by these.

We recommend careful consideration of the monitoring goals prior to selecting any sampling method. When the objective is to cast a wide net to survey the range of pesticides present in the environment and to track trends in their relative usage/release, o-DGTs should be considered because of their high sensitivity to a large range of pesticides and capacity for time-integration. Allocating funds and resources toward lowering detection limits, rather than sampling more than matrix, would better equip these types of programs. When the goal is to characterize contaminant exposure for the purpose of comparison with water quality guidelines and ecotoxicological risk thresholds or to contrast with other research studies (given how commonly environmental monitoring relies on water grab samples), then high-frequency water sampling is best selected. To support the assessment of contaminant exposure and ecological risk from data obtained from o-DGTs, we suggest targeted, high-frequency water sampling, particularly for specific areas where there is concern about contaminant pulses surpassing ecological thresholds. When the objective is to understand the environmental fate of pesticides or potential risk to the aquatic food chain, biofilm sampling should be included in the program. Biofilms are also useful as a substrate for use in bioassays to assess the dietary exposure of pesticides (e.g., ref. [39]). When resources are available and objectives are manifold, we recommend measuring multiple matrices. The collaborative monitoring of stormwater management ponds undertaken in this research provided a more comprehensive understanding of the diversity of pesticides present in the urban environment.

4. Conclusions

In our study of urban stormwater management ponds, we found widespread pollution with pesticides, including an assortment of 82 different herbicides, insecticides, and fungicides in water, biofilm, and using o-DGT samplers. These findings raise important questions about the ecological impact of chronic exposure to these chemical mixtures. Such extensive contamination of the urban environment concerns the well-being of both residents and wildlife in cities and must be addressed to ensure the sustainability of urban water systems and the health of urban biota.