Microplastic removal across ten drinking water treatment facilities and distribution systems

Microplastics (MPs) have been reported in both source and drinking waters around the globe^{1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18}. Drinking water serves as a direct exposure route along with food, air, and skin contact¹⁹, highlighting the importance of their quantification as well as removal during treatment. However, studies which focus on drinking water have employed a wide range of methods for collection and isolation of MP particles, making it challenging to directly compare results regarding treatment process performance across different facilities. Monitoring of MPs in drinking water has been legislated in the States of California (Senate Bill 1422, 2018) and New Jersey (Assembly Bill 4821, 2023), further emphasizing the need for a robust and practical assessment methodology to be developed and widely adopted^20,21. The ability to capture and quantify MPs <30 µm is of particular importance due to potential implications to human health^22,23,24,25. Many methodological challenges exist when sampling MPs in drinking water. For example, deposition of airborne MPs may potentially contaminate samples and should be mitigated by employing enclosed in-line sampling methods²⁶ rather than use of bottles^{1,2,8,9,16,17}. No single study has systematically investigated the occurrence and removal of MPs <30 µm from a wide range of treatment processes in order to obtain a representative estimate of exposure via drinking water.

In this study, an extensive MP sampling campaign included ten conventional (employing chemically assisted granular media filtration), as well as advanced drinking water treatment facilities (employing micro- or ultrafiltration). MP quantification, size measurement, and chemical identification of particles ≥2 µm were conducted using microscopy coupled with Raman spectroscopy. Facilities were selected to represent: (1) a range of processes used to treat water originating from the same source, and (2) similar treatment processes used to treat different source waters. Study sites were geographically diverse with source waters encompassing four lakes and two rivers (Table 1). Selected facilities served by the same source waters were sampled on different days, thus providing an indication of potential MP temporal variability. Treatment processes included those classified as conventional (A, B, C, D, E, and F), as well as advanced which included ultrafiltration (UF) (G, H, and I) or microfiltration (MF) (J). Since the primary purpose of filtration is to remove particulate matter, it is anticipated to contribute substantially to MP reduction, as previously reported in studies regarding conventional¹³ and advanced²⁷ treatment processes. Coagulation, flocculation, and sedimentation (CFS) have been suggested to serve as primary contributors to MP removal rather than filtration^1,9,15,28,29. As such, samples were collected following both CFS and filtration, wherever possible. Other treatment processes included biological activated carbon (BAC), ozone, and ultraviolet (UV) disinfection (Tables S1 and S2). Microorganisms present in BAC contactors have been hypothesized to bioregenerate adsorption sites by consuming adsorbed organic compounds, thus enabling biodegradation and adsorption to continue over extended periods of time³⁰, which may also contribute to MP removal³¹. Previous studies have suggested that ozone may cause fragmentation of MPs, resulting in an increase in concentration^1,15. Collection of samples prior to and following ozonation was conducted to specifically address this question. In contrast, disinfection using UV or UV-based advanced oxidation are not anticipated to result in MP fragmentation: concentrations of surface functional groups of low- and high-density polyethylene as markers of oxidation, were not observed to be impacted when applying UV or UV/H₂O₂ at dosages typically employed in water treatment³², although leaching of organic matter may be possible³³. To examine the performance of specific treatment processes, samples were collected from: (i) untreated source water, (ii) following each major unit treatment process, (iii) finished water (following disinfection), and (iv) the nearest regulatory sampling point in the distribution system (hereafter referred to as “distribution”). To capture MPs, water was filtered onsite using an enclosed, stainless-steel inline system which contained 10 µm and 2 µm polycarbonate (PC) filters arranged in-series³⁴. Prior to analysis using Raman microscopy, samples were subjected to Fenton oxidation, followed by enzymatic digestion using cellulase and trypsin (method adapted from Cheng et al.³⁵) to reduce the presence of non-plastic particulates.

MP concentrations in source waters ranged from 1193 ± 64 MP/L to 7185 ± 64 MP/L (Fig. 1, Table S3). Previous studies which considered a minimum particle size of 1 µm have reported concentrations within the same order of magnitude (multiple thousands of MP particles per liter)^1,8. Concentrations for two treatment facilities which shared the same intake and were collected on separate days (Lake 3, F and H) ranged from 2521 ± 64 MP/L to 4574 ± 64 MP/L. A comparable 2668 ± 64 MP/L was observed for the influent of facility E (Lake 3 at an intake 50 km away from F and H). Varying MP concentrations for the same source water at different times suggest potential temporal and spatial variability. Repeated monitoring over time is required to investigate temporal changes more thoroughly.

A Pearson correlation analysis indicated a weak, non-significant negative relationship between intake depth, which ranged from 3 to 26 m, and source water MP concentration (r = −0.213, p = 0.555). Although intake depth may impact MP concentration, the presence of MP particles in a given water column may also be influenced by polymer density, morphology and size, as well as biofouling^36,37. Stratification, turbulence, and sediment resuspension can redistribute MPs^37,38,39 and reduce depth-related patterns^37,40,41,42. As such, intake depth alone cannot serve as a reliable indicator of MP concentration.

While a surrounding population (Table 1) is often considered to influence MP concentrations at intakes, several other factors may include land use^43,44, proximity to point sources (e.g., wastewater discharges)⁴⁵, and hydrological conditions (e.g., flow velocities, retention of MPs in soils, vegetation, or upstream reservoirs)^37,44,46,47. Several of the source waters included in this study share overlapping watersheds, making it challenging to identify the influence of a specific population center.

MP concentrations in the lakes exhibited greater variability as well as higher maximum values when compared to rivers (Fig. S1). Lakes may act as sinks for MPs⁴⁸, but the extent is influenced by surrounding land use, direct point-source pollution, inflow characteristics, and hydrodynamics^43,44. The facilities considered in this study primarily received water from large lakes where dilution may serve to reduce MP concentrations⁴⁴. As a result of these complex factors, MP quantification must be conducted on a site-specific basis.

Finished water concentrations

When considering all ten facilities, MP concentrations in finished waters ranged from non-detectable (facilities G and H) to 64 ± 64 MP/L (facility C) (Fig. 1, Table S3). Following application of a one-way ANOVA with post-hoc Tukey’s HSD, a significant reduction in MP concentrations was observed when comparing source waters to finished waters (F = 29.1, p = 1.82 × 10⁻⁷; Fig. S2), confirming effective removal during treatment. Finished water concentrations in facilities employing UF (G, H, and I) significantly differed from those employing MF (J) or granular media filtration (A-F) (F = 7.94, p = 0.0226, Fig. S3). Granular media is capable of removing particles 0.1 to 1 µm⁴⁹, similar to MF with a pore size of 0.1 µm (Table S2). Lower MP concentrations (ranging from non-detect to 12.7 ± 64 MP/L) were observed following UF with a pore size of 0.02 µm (Table S2) when compared to post-filtration at facilities that employed granular media filtration or MF (9.3 ± 64 to 400.2 ± 64 MP/L). Theoretically, particles >2 µm should not be present following membrane filtration where nominal pore sizes range from 0.01 to 0.2 µm⁵⁰. However, samples could not be collected directly from membrane permeate, but rather at the first accessible location immediately downstream, typically a clearwell, which can be susceptible to airborne MP deposition¹⁵.

Total microplastic removal

Total MP removal was calculated by comparing concentrations in finished waters to those in source waters (Table S3). For facilities B and G, removal was based on the first sample collected in the treatment process (i.e., post-sedimentation for facility B and post-ozone for facility G), as source water data was not available. When directly comparing conventional facilities (A–F), removals ranged from 98.7% to 99.4%. For the one facility that employed MF (J), removal was 97.5%, whereas the three facilities that incorporated UF (G, H, and I) achieved 99.3% to 100%. A significant difference in removal (one-way ANOVA, F = 6.39, p = 0.0354; Figure S4) was observed when comparing UF facilities (G, H, and I) to MF (J) as well as to conventional facilities (A-F). This difference may be attributed to the smaller UF pore size (0.02 µm). A previous bench-scale study involving similar UF membranes reported complete rejection of MPs > 1 µm²⁷. Removal observed across all facilities was >97.5% despite the significantly greater removal for those which employ UF (Fig. 1). Previous studies have reported treated water concentrations (particles ≥ 1 µm) to range from 18.7 to 1401 MP/L^1,2,8,17,51 with corresponding removals of 50% to 88%. Higher removal efficiencies observed in the current study (97.5% to 100% for MPs ≥2 µm) can be attributed to differences in minimum particle size thresholds, sampling and analytical methods, as well as operational parameters. Several previous studies quantified particles (>1 µm) using scanning electron microscopy^1,8,9,51 and subsequently characterized polymer types based on a subset of larger particles (≥10 µm⁸ or ≥50 µm in size⁵¹) using Raman or FTIR spectroscopy^29,42. In the present study, MP quantification, size, and chemical identification of particles ≥2 µm were all determined via microscopy coupled with Raman spectroscopy.

Removal by individual treatment processes

Filtration, either via granular media or membranes (MF or UF), provided the greatest contribution to removal, ranging from 87.5% (facility B) to 100% (facility G) (Table S3). Statistical analysis could not be conducted with respect to MP removal via individual treatment processes or parameters such as granular media filter depth, membrane age, sedimentation, ozonation, or BAC filtration (Tables S1 and S2) due to variability in treatment design and operation. When considering conventional facilities, qualitative assessment suggested that specific media type (e.g., granular activated carbon vs. anthracite over sand), filter loading rate, bed depth, or effective grain size did not have an impact on MP removal. For advanced facilities, the age of UF or MF membranes did not appear to impact MP removal.

While several previous bench-^28,52,53,54 and full-scale^1,9,55 studies reported coagulation-flocculation-sedimentation (CFS) to provide MP reduction, this was not apparent in the current study (Fig. 1). Studies by others also suggested that CFS does not provide a barrier for small MPs (i.e., <20 µm), as they are less effectively removed by Al- and Fe-based coagulants when compared to particles ≥20 µm⁵⁴. In this study, facilities D, F, and I exhibited increased MP concentrations post-sedimentation (compared to source water), whereas facilities A and C exhibited a decrease (Table S3). A shift in particle size distribution towards smaller particles following CFS was not observed, suggesting that it did not preferentially remove larger particles. In addition, no consistent MP removal trends were observed with respect to BAC age, or the application of ozone or UV.

Wastewater effluent, airborne MPs, and surface runoff are typically believed to contribute MP fibers^56,57,58,59, defined as particles with a length-to-width ratio >3⁶⁰. As such, source waters may be anticipated to contain a larger proportion of fibers (versus fragments), reported by others to comprise 37% to 61%⁸ and 54% to 74%¹ of total MP concentrations. In contrast, among the ten facilities examined in this study, fibers comprised 7.8% and 7.5% of all MPs in source and finished waters, respectively (Table S3). When considering total MPs in all samples, only 6.9% were fibers, with fragments being the predominant morphology. The relatively low observed proportion of fibers may be due to potential fragmentation during Fenton oxidation (especially PET and PA fibers which contain ester and amide bonds, respectively) and/or digestion by cellulase (for cellulosic fibers). In agreement with previous studies^1,8, the relative proportions of fragments to fibers did not change when considering a wide range of source and finished waters, suggesting that morphology did not impact removal during treatment.

When considering size, defined as the major dimension of a given MP particle, 81.1% ± 9.4% of all MPs were <20 µm; 50.6% ± 10.4% were <10 µm (Fig. 2, Table S6). As previously reported, removal of MPs during drinking water treatment is anticipated to increase as a function of particle size^1,8,9,61. In this study, a reduction in the proportion of MPs >20 µm was observed for facilities A, B, C, F, and J, whereas for D, E, G, H, and I they remained the same or increased (Table S6). When considering all finished waters, MPs <20 µm comprised >75% of all MPs observed, suggesting that previous studies (which only characterized particles > 20 µm) may have unintentionally excluded a substantial proportion of MPs present in drinking water. Thus, it is recommended that future studies employ Raman microscopy for polymer type identification, quantification, and measurement of MPs >1 µm. Accurate reporting of MP count, size, and polymer type is especially important given that particles <30 µm may translocate within humans²⁵. The relative abundance of particles in the 2–5 µm size fraction is lower than anticipated based on the overall decreasing trend with increasing particle size (Fig. 2). This discrepancy likely reflected the narrower bin width (3 µm vs. 5 µm in other bins).

Polymer type was dominated by polypropylene (PP), polyethylene (PE), polyamide (PA), and copolymers in all samples (Tables S4 and S5), with PP typically being the most abundant (average relative contribution 61.5% ± 23.3%). Values were below detection limits for facility G. Previous studies have reported PP and PE to be the most abundant types present in source and treated waters^{1,7,8,9,61,62,63,64,65}. Source waters associated with facilities G and I exhibited a large proportion of polyester (31.1% and 25.0%, respectively) when compared to others (1.9% ± 2.5%). This discrepancy may be due to environmental and temporal variability since facilities F and H share the same intake as G but were sampled at different dates. Seasonal flow dynamics, storm events, or variations in MP input to source waters could influence the types and proportions of MPs present at any given time^47,48. As such, the absence of elevated polyester levels at facilities F and H compared to facility G may reflect short-term or seasonal variability.

The impact of polymer type on removal was qualitatively examined (Table S4) for individual treatment facilities by considering both source and finished waters (Fig. 3). Neither conventional nor advanced treatment processes preferentially removed any specific polymer type, implying that removal mechanisms may be similar to those for inorganic particulates and not impacted by chemical composition, as reported by others^1,9. Filtration (by granular media or membrane filtration) removes particles based on physical exclusion or adsorption⁶⁶. This functional similarity likely contributed to the lack of preferential removal of specific polymer types as observed in this study.

No statistically significant difference in MP concentrations was observed when comparing finished water and distribution system water (Tukey’s HSD pairwise comparison_{FW vs DS}, p = 1.00, Fig. S2). It is hypothesized that deposition of airborne MPs into reservoirs may potentially represent a source of MPs within distribution systems¹⁵ but was not observed in this study. In addition, relative polymer type abundance and size distribution were similar when qualitatively comparing finished and distribution system water (Figs. 2 and 3, Tables S5 and S6). It should be noted that hydraulic residence time was not considered; sampling locations represented those employed for regulatory monitoring purposes and in general were supplied by large diameter (≥900 mm) concrete pressure pipes.

The distance traveled through a given distribution system was not observed to have a significant impact on MP concentration when comparing finished and distribution system water (Pearson correlation, r = −0.202, p = 0.5752; Fig. S5). It is suggested that more extensive sampling throughout distribution systems be conducted in future studies to assess the influence of pipe material and age, hydraulic residence time, flushing/cleaning practices, as well as potential exposure to airborne MPs in reservoirs.

Routinely monitored water quality parameters including turbidity, total particle counts, UV₂₅₄, specific UV absorbance at 254 nm (SUVA), pH, total and dissolved organic carbon (TOC and DOC), were measured for all samples (Table S9). Specific fractions of natural organic matter (NOM) in source and finished water were also quantified (Table S10). Only parameters that represent particulate material (turbidity, total particle counts, and TOC) were examined for potential correlation with MP concentration. Prior to analysis, UV₂₅₄, SUVA, DOC, and NOM samples were passed through 0.45 µm filters and as such represent dissolved material.

A Spearman correlation analysis using source water values revealed no significant correlations between MP concentration and turbidity (r = −0.249, p = 0.487) or total particle counts (r = −0.297, p = 0.405) (Fig. S6). A significant negative correlation was observed between MP concentration and TOC (r = −0.758, p = 0.0111) (Fig. S6), which may indicate sites with higher natural organic matter to be less impacted by anthropogenic MP inputs. However, this relationship should be interpreted with caution, as the apparent correlation may be influenced by other site-specific conditions including organic loading, hydrodynamics, and proximity to MP inputs. The absence of a significant relationship between MP concentration and turbidity is likely due to the fact that source waters also contain many particulates of which microplastics only represent a small fraction.

When considering finished water, no significant correlation was observed between MP concentration and turbidity (r = −0.483, p = 0.157), TOC (r = −0.288, p = 0.420), or total particle counts (r = 0.122, p = 0.738) (Fig. S7), again indicating that traditional water quality parameters do not correlate with MP concentrations. Similarly, total MP removal did not correlate significantly with the removal of either turbidity (r = −0.0791, p = 0.828), TOC (r = −0.289, p = 0.418), or total particle counts (r = 0.0669, p = 0.854) (Fig. S8). These findings emphasize that it is essential to implement a dedicated monitoring program in order to quantity MP concentrations, specific polymer types, and sizes.

Microplastics were consistently present in source waters including lakes and rivers with considerable variability in concentration and polymer composition. Most MPs consisted of fragments <20 µm, predominantly represented by PP, PE, PA, and copolymers. Drinking water treatment facilities which employ a range of conventional and advanced processes were capable of removing >97.5% of MPs ≥2 µm. Chemically assisted granular media filtration or membrane filtration provided the highest removal, while coagulation, flocculation, and sedimentation, ozonation, and UV disinfection were observed to have minimal impact. The similarity in removal across a range of particle sizes, shapes, and polymer types suggests that physical exclusion, rather than chemical interaction, is predominant. A lack of correlation between MP concentration and conventional water quality parameters, such as turbidity or TOC emphasizes the need for direct measurement using dedicated analytical methods. Overall, these findings demonstrate the capability of drinking water treatment practices in reducing MP exposure to consumers, while highlighting the necessity for standardized protocols to ensure accurate assessment. Future studies should further explore the impacts of specific distribution system factors such as pipe type, age, water velocity, and hydraulic residence time.

Sampling locations

MPs were collected from ten drinking water treatment facilities located across southern Ontario, Canada, representing a range of source waters and treatment processes (Table 1). These facilities span a range of urban settings and are situated in areas influenced by different watershed characteristics and climatic conditions typical of the Great Lakes Basin, including temperate seasonal variability with cold winters and warm, humid summers. At all facilities, single samples were collected at the intake (defined as “source water”), immediately following each individual treatment process (whenever possible), as well as finished water (after the final stage of treatment). Distribution system samples were collected at the first downstream sampling location typically used for regulatory compliance monitoring. Conventional treatment facilities are classified as those which employ granular media filtration, whereas advanced treatment facilities incorporated micro- or ultrafiltration. Coagulation and flocculation were practiced at all facilities with the exception of E, G, H, and J. Samples were collected from the effluent of individual filtration processes for facilities that employed two different types of granular media filters operated in parallel (anthracite over sand and granular activated carbon (GAC) over sand, facilities A and E, respectively), or biologically activated carbon (BAC) (facility J). Process-specific characteristics which included grain size (granular media filtration), ultra- or microfiltration membrane age, and UV and chlorine dosages are shown in Tables S1 and S2.

Microplastic sampling equipment

Due to the extensive scope of this study, which involved sampling ten different drinking water treatment facilities across southern Ontario, only one sample per treatment step per facility was collected. This approach was necessary to balance the practical constraints of fieldwork logistics, including the considerable travel distances between facilities and the time-intensive nature of MP sampling and analysis. The sampling strategy prioritized capturing a broad representation of different treatment technologies and operational conditions across the region, in lieu of repeated sampling at individual facilities. While this potentially limits the ability to assess short-term variability associated with source water intakes and within treatment processes at a single site, it provides valuable insight regarding the overall pattern of MP removal across a diverse range of full-scale drinking water treatment operations.

To collect MPs, an enclosed in-line filtration apparatus consisting of two stainless-steel, 47 mm diameter filter holders (MilliporeSigma, Darmstadt, Germany) connected in-series was employed as described by D’Ascanio et al.³⁴. Particulates were isolated onto 10 and 2 µm, 47 mm diameter polycarbonate (PC) filters (MilliporeSigma, Darmstadt, Germany) enclosed within the holders. Quick disconnect brass couplings and control valves were located at both the inlet and outlet. Threaded connections were sealed using polytetrafluoroethylene (PTFE) tape. Rubber O-rings were used to seal the interior of the filter holders, and to prevent potential contamination by airborne MPs. Only facility E was sampled using only a 5 µm PC filter (Sterlitech, Auburn, Washington, United States) upstream of a 2 µm PC filter (instead of a 10 µm) filter due to rapid occlusion of the 2 µm filter by particulates during initial sampling trials. Specific polymer types associated with sampling or lab equipment were excluded from analysis, including PTFE, PC, nitrile glove material and rubber O-rings. All “MP-free” reagents (Table S12) were filtered through a 0.45 µm PTFE filter (Omnipore, Millipore Sigma, Darmstadt, Germany) and stored in rinsed, closed glass bottles.

Prior to sampling, the filter holders were sonicated in MP-free water for 30 min and subsequently cleaned using MP-free water which consisted of 0.05% Tween® prepared with ultrapure water (18.2 MΩ·cm, 0.2 µm filtered, Milli-Q EQ 7000 System; MilliporeSigma, Darmstadt, Germany). All cleaning and assembly were conducted within a Class II laminar flow hood to minimize the potential for contamination by airborne particles. All glassware and equipment including glass filtration apparatuses, Pasteur pipettes, 600 mL glass beakers, 1 and 4 L glass bottles, PTFE stir bars, sampling equipment, and stainless-steel forceps were rinsed three times using MP-free water. Following placement of PC filters inside the in-line filtration holders, brass control valves at each end (inlet and outlet) were closed and covered with aluminum foil to minimize any potential contamination by airborne MPs.

At all water treatment facilities, individual samples were collected starting with finished water and ending with source water. Samples were collected directly from the influent or effluent of specific treatment processes via sampling taps, whenever possible. When sampling line pressure was ≥30 PSI, in-line filter holders were connected directly. For pressures <30 PSI, or when an air gap was requested by facility personnel, a rotary vane pump (GA072, Fluid-o-Tech) and a 30 L stainless steel constant head tank (CHT) equipped with an overflow port at 23 cm, inlet port at 17 cm, and outlet at 4.5 cm from the bottom was employed (Fig. S9). The tank was covered with a stainless-steel lid. All tubing upstream of the in-line filtration apparatus to the sampling port consisted of braided stainless steel. A pressure relief valve set to 50 PSI (Apollo Valves, Matthews, North Carolina, United States) was incorporated upstream of the filter holders to ensure PC filter integrity.

Prior to sample collection, all components (including the pump and CHT when required) were initially flushed for 15 min following connection to the sampling port. The in-line filter was then connected to collect a pre-selected volume of water. Due to higher turbidities upstream of granular media filtration or UF, lower sample volumes (typically 0.5 L) were collected when compared to post-filtration (typically 50 L). Sample volumes initially collected from facility G were used to inform those subsequently collected for both pre- and post-filtration at other facilities. Particle concentrations per analyzed filter area obtained from facility H samples were used to inform pre-filtration sample volumes that would result in a number of particles that could be analyzed within 2–3 days based on the Raman microscopy procedure used in this study (Table S3). Volumes ≤ 0.5 L were measured using a 1 L graduated cylinder placed at the discharge end of the in-line filter, whereas volumes > 0.5 L were quantified using a positive displacement low flow water meter (Assured Automation, Roselle, New Jersey, United States) at the outlet of the in-line filtration apparatus. Following collection of a desired volume, inlet and outlet control valves were closed and the in-line filter assembly disconnected. Inlet and outlets were covered with aluminum foil for transport to the laboratory.

Turbidity and pH were measured in the field using handheld instruments (HF Scientific MicroTPW Portable White Light Turbidimeter and Orion Star A121 Portable pH Meter, Fisher Scientific, Ontario, Canada). Parallel samples were also collected for in-lab analyses which included TOC, DOC, UV254, specific UV absorbance (SUVA), particle count, turbidity, and pH. Additional samples of raw and treated water were collected at each facility (except for G) for liquid chromatography organic carbon (LC-OCD) analyses. Complete water quality data is shown in Table S9; LC-OCD results are shown in Table S10.

Analytical methods

Extraction and digestion

A vacuum pump was used to filter any residual liquid in the filter holders (inside a Class II laminar flow hood). The in-line filter holders were then disassembled and the two PC filters placed into 600 mL beakers. All interior parts of in-line filters (including stainless steel underdrains and support screens), were rinsed three times into the same 600 mL glass beakers using a minimum of 5 mL of MP-free water. PC filters and associated water were sonicated for 55 min at 25 °C. PC filters were removed using forceps and rinsed into a beaker using MP-free water. The suspension was filtered through a 1 µm PTFE filter (Omnipore, MilliporeSigma, Darmstadt, Germany) using vacuum filtration. The PTFE filter was then subjected to Fenton oxidation, followed by cellulase and trypsin enzymatic digestion following a method adapted from Cheng et al.³⁵. Reagent specifications and operational parameters are provided in Table S14.

Following digestion, each sample was filtered using a 1 µm, 47 mm diameter PTFE filter for analysis. Prior to analysis using Raman microscopy, PTFE filters were mounted on a custom built stainless- steel stage, specifically designed to ensure that the filters were held taut to provide a flat surface for analysis.

Raman spectroscopy methodology

Raman spectroscopy coupled with microscopy was performed using a Horiba XPlora Plus system with LabSpec 6.6 software (Version 6.6.2.7, Horiba Scientific, Kyoto, Japan). Instrument parameters are provided in Table S15. A spectral range of 150–3600 cm^-1 was employed to include the entire range where peaks characteristic of plastics are known to occur⁶⁷. The laser power at the surface of the filter was 7.54 mW at 100% when using a long working distance (LWD) 50× objective (LMPlanFL N 50x/0.50 BD, Olympus, Richmond Hill, Ontario, Canada). The system was calibrated every 24 h to a 520.7 cm⁻¹ peak obtained from a silicon wafer.

Evaluation of the entire area of a 47 mm diameter filter would result in unrealistic analysis times, exceeding two weeks for a single sample. Initially, 6.6% of the filter area was analyzed (facilities G and H), as adapted from Pittroff et al.¹⁰. The filter area was divided into two zones: an inner circle with a diameter of 25 mm and an outer ring with an inner diameter of 25 mm and an outer diameter of 35 mm; both zones had equal surface areas (Fig. 4). Particles in these zones were analyzed to account for potential variations in particle concentration between the center and edge of the filter. For all other facilities, 0.45% to 5.42% of the total area was analyzed to ensure that a minimum of 40,000 to 50,000 individual particles were considered, which typically could be accomplished over two to three days. For each filter area, a mosaic image was obtained using the ViewSharp tool to create a corresponding topographic map using the 50× LWD objective with bright field illumination. Mosaic images were then analyzed using the Particle Finder tool to identify individual particles and subsequently collect corresponding Raman spectra.

A two-phase spectral acquisition approach was applied to optimize overall analysis time. During Phase I, spectra were acquired, baseline corrected, and screened for the presence of MP particles based on the C-H stretching region observed for most polymer types (except PTFE) from 2800 to 3150 cm^{−1 10,68,69} by employing the parameters and settings described in Table S13. In Phase II, post-acquisition and correction screening was applied to identify potential plastic polymers. Following screening, points on the mosaic corresponding to peak areas >2 were selected for further analysis where higher quality spectra were acquired for subsequent comparison to libraries of known polymers. Following baseline-correction, spectra were imported using WITec TrueMatch (6.1.7, WITec Wissenschaftliche Instrumente und Technologie, Ulm, Germany) for identification using three databases: (i) S.T. Japan (S.T. Japan Europe GmbH, Köln, Germany), (ii) SLOPP, and (iii) SLOPP-E⁶, as well as an in-house database associated with 21 polymers from Polymer Kit 1.0 (Hawai’i Pacific University Centre for Marine Debris Research). In order to reduce any potential influence of the PTFE filter on HQI values of suspected polymers, the region of 709–759 cm^−1 3,4 was excluded from the database search. To reduce potential overestimation of MPs, those identified as polymers matching items used in sample collection/extraction (e.g., rubber O-rings inside the in-line filter holders, nitrile gloves) were disregarded. All spectra were manually confirmed; material types were sorted into groups using a method adapted from Munno et al.⁷⁰ (Table S5). As described in Table S5, anthropogenic cellulose and anthropogenic unknown particles were identified, but excluded from further calculations. Anthropogenic cellulose was defined as particles with a spectrum consistent with dyed cotton or dyed cellulose, whereas the anthropogenic unknown class unites all additives, dyes, pigments and material which indicate anthropogenic origin, or modification without definitive plastic or cotton/cellulose indications.

QA/QC procedures

Contamination control measures

Whenever possible, all laboratory work was conducted in a Class II laminar flow hood. White 100% cotton lab coats and nitrile gloves were worn at all times by lab personnel. All glassware and equipment used for sampling, cleaning, digestion, and spectroscopic analysis was rinsed three times with MP-free water. Areas in the lab where MP sample digestion or analysis were conducted (outside of a Class II laminar flow hood) were equipped with portable high efficiency particulate air (HEPA) filtration systems (Coway, Las Vegas, Nevada, United States) to reduce any potential airborne particulates. Throughout all analyses, samples and MP-free water were kept covered whenever possible.

Laboratory blanks

Three laboratory blanks were prepared by placing new 10 µm and 2 µm, 47 mm diameter PC filters within a cleaned in-line filtration apparatus, removing it from the Class II laminar flow hood (to simulate transporting equipment to the field for sampling), and then returning it to the laminar flow hood for subsequent MP extraction. The same procedures for the extraction and digestion of actual samples were employed for laboratory blanks. To address any potential changes in lab or operating conditions, blanks were conducted at time intervals evenly spread across the entire study period, typically one blank per 25 samples with a laboratory blank approximately every 4 weeks.

Data processing

MPs identified in analyzed filter areas were extrapolated to the entire filtration area (979 mm²). Mean extrapolated MPs measured in blanks were subtracted from the extrapolated MPs within actual samples. Particle concentrations (MP/L) were calculated by dividing blank-subtracted values by the volume of water filtered. Standard deviation of the mean number of MPs for blanks (n = 3) was used to indicate the variation associated with blank subtraction.

Statistical analyses

Quantitative analyses were carried out wherever data availability and consistency permitted. In cases where a dataset was too small or exhibited high variability in operational parameters (e.g., membrane age, sedimentation and filtration parameters), qualitative assessments were conducted to support interpretation.

Statistical analyses were conducted using Python. Normality of the data was assessed using a Shapiro-Wilk test. Results can be found in Table S11. Depending on the distribution of the data, either Pearson or Spearman correlation coefficients were calculated.

Group differences were evaluated using one-way ANOVA, followed by Tukey’s HSD post-hoc testing where applicable. Results are shown in Figs. S1–S8.

Study limitations

Reported concentration values (MP/L) represent extrapolations based on the analysis of 0.45–5.42% of total filter area. Only MPs ≥ 2 μm were considered due to the minimum pore size of the filters employed during sampling. Given that only a percentage of each sample was analyzed via Raman spectroscopy, the resulting sub-sampling introduces a deviation of approximately 17% or more⁷¹. Thus, reported values should be interpreted as estimates rather than precise quantifications.

Raman spectra were acquired using a method that may exclude particles which exhibit fluorescence or degrade via heating and/or photodecomposition resulting in a peak area of <2 within the 2800–3150 cm⁻¹ spectral range, which would result in omission of those particles from the final count. As such, MP/L concentrations reported in this study represent a conservative estimate. Intra-sample variability could not be assessed as only single samples were collected. Environmental factors such as temporal fluctuations in water flow, temperature, and biological activity could not be controlled or systematically assessed. These sources of uncertainty should be considered when interpreting both the concentration estimates and the broader implications of the findings.