Review of trace organic chemicals in urban stormwater: Concentrations, distributions, risks, and drivers

Urban stormwater, increasingly seen as a potential water resource for cities and towns, contains various trace organic chemicals (TrOCs). This study, conducted through a comprehensive literature review of 116 publications


Introduction
Climate change poses a threat to water availability, and meanwhile, the demand for water is expected to rise in urban areas due to population growth and densification (Larsen et al., 2016).Under this context, urban stormwater presents an attractive solution to supplement water supply (Luthy et al., 2019) while simultaneously mitigating floods and redirecting pollutants from entering waterways (Zhang et al., 2020).Understanding stormwater pollution is especially important to ensure the safety of stormwater harvesting or discharge into water bodies.The focus of research efforts in this area has traditionally been on characterizing stormwater quality in terms of suspended solids, nutrients, and heavy metals (Li and Davis, 2014;Liu et al., 2018).Much less attention has been paid to the pollution of trace organic chemicals (TrOCs), also known as priority substances, chemicals of emerging concern, xenobiotic compounds, organic micropollutants (Zgheib et al., 2012;Spahr et al., 2020).Despite being detected at trace levels (ng/L to μg/L), TrOCs may cause significant concerns for the environment and human health (Eriksson et al., 2005;Eriksson et al., 2007;Mutzner et al., 2023).For example, 6PPD-quinone, an acutely toxic substance, has been detected in urban runoff at concentrations of 86 -1400 ng/L, exceeding the reported toxicity levels of coho salmon, and may also have potential risks to other aquatic organisms (Challis et al., 2021).
Recently, efforts have been made to monitor TrOCs in urban stormwater.For example, Masoner et al. (2019) reported an extensive monitoring campaign of stormwater from 50 runoff events across 21 sites in the USA, which detected 215 organic chemicals out of 438 analyzed.Fairbairn et al. (2018) investigated 384 contaminants of emerging concern in 36 stormwater samples in Minnesota, USA, and detected 123 from various resources.Rippy et al. (2017) tested 27 pesticides in multiple Australian catchments over long-term monitoring and detected 19 in the stormwater, many of which were found in over 50 % of the samples.Through high-resolution monitoring and machine-learning approaches, Yun et al. (2023) identified the dynamics of micropollutants and first flush efforts (FFE) in an urban catchment.They found that pesticides and pharmaceuticals showed higher FFE than industrial micropollutants.These studies underscore the growing concern and imperative for investigating TrOCs in stormwater.
A comprehensive review of these existing bodies of research could provide a fundamental understanding of the stormwater quality regarding the TrOCs pollution, which could allow the identification of trends and patterns, awareness of their risks, development of effective mentoring strategies, and informing policy decisions and regulatory frameworks.A recent review by Mutzner et al. (2023) identified 49 persistent, mobile, and toxic (PMT) substances and suggested that unified guidelines are needed to monitor PMTs in stormwater and assess their environmental risks to ensure safe stormwater capture for water reply.Spahr et al. (2020) identified the sources and occurrence of a specific subset of TrOCs in stormwater, i.e., hydrophilic TrOCs, and their toxicological relevance by benchmarking against environmental quality standards.However, quantitative assessments (e.g., risks, distributions) were not performed using the collected data from only 18 peer-reviewed publications.Some important questions remain unclear: (1) What is the expected distribution of TrOC concentrations in stormwater? ( 2) what are the ecological and health risks of these TrOCs for different stormwater harvesting scenarios?(3) how do the land use and rainfall characteristics influence the TrOC pollution levels?
Concentration distributions.Stormwater quality is highly variable in time and space, and thus, knowing the statistical distribution will allow the appropriate parameterization of models to estimate pollution levels, aiding the mitigation decision-making processes.Several stormwater studies have focused on statistical analysis, risk assessment, and modelling of stormwater pollutants based on pre-assumed distributions (Duncan, 1999;Métadier and Bertrand-Krajewski, 2012), highlighting the crucial role of appropriate distribution assumptions in these analyses.Yet, there is no understanding of TrOC concentration distributions in stormwater.Concerns have been raised about data availability, e.g., raw datasets were often inaccessible, and extensive efforts had to be made to acquire the data for systematic analysis (Mutzner et al., 2022).Therefore, evaluating the concentration distributions of TrOCs can be highly beneficial for advancing the statistical modelling of their pollution and enhancing control measures in stormwater management studies.
Risk assessment.Despite many TrOCs being regarded as 'emerging contaminants' with potential environmental and health risks, it is surprising that risk assessment has not been adequately conducted in stormwater runoff.Both Spahr et al. (2020) and Mutzner et al. (2022) evaluated the environmental risks of TrOCs in stormwater by comparing the EMCs with environmental quality standards (EQS) for surface waters.The risk quotient (RQ) was estimated by dividing a precautionary 90 %-percentile of concentrations with chronic EQS (and acute EQS when the chronic value was not available), with the data originating from multiple sources, including EU Directive, USEPA, national water quality standards, etc. Due to limitations in the EMC values (that lump intra-event-variability) and the RQ method (which shows the required dilution factor, not the actual eco-toxicological risk), Mutzner et al. (2022) recommended the assessment of actual eco-toxicological to complement the EQS approach.Ma et al. (2017) used a toxic equivalent factor approach for PAHs and showed that they may pose potential cancer risks when considering direct uptake of stormwater and dermal contact.It's important to highlight that previous risk assessment studies often relied on data from a limited number of past studies or concentrated on a small subset of TrOCs.The go-to document in Australia for practitioners wishing to conduct stormwater harvesting, i.e., Australian Guidelines for Water Recycling: Stormwater Harvesting and Reuse (NHMRC, 2009), considers stormwater "not to be expected to be chemically hazardous".This conclusion, however, was not built on any stormwater quality data on TrOCs.Thus, a comprehensive review of TrOCs in stormwater and specific ecological and health risk assessments are urgently needed.
Urban land use has been shown as an essential factor affecting stormwater quality (Goonetilleke et al., 2005;Simpson et al., 2022).Specifically, for TrOCs, traffic volumes were often found to correlate with concentrations of PAHs (Gasperi et al., 2014;Gunawardena et al., 2018;Wicke et al., 2021), and phthalates (Wicke et al., 2021).Wicke et al. (2021) found that commercial areas contribute to elevated concentrations of flame retardants.Fewer studies have been conducted for other types of TrOCs, and the results often suggest little or no correlation with the land use types.For example, Zgheib et al. (2011) observed no significant differences in the pollutant loads between the land use patterns and a range of priority substances across organotins, PAHs, PCBs, alkylphenols, pesticides, phthalates, etc. Mutzner (2019) found no clear correlations between land uses and TrOCs in stormwater, attributing this to the large spatial variability of TrOCs in combination with insufficient data.Thus, there is a need for a comprehensive assessment of the relationship between land use types and a wide range of TrOCs.
Rainfall characteristic is another critical factor influencing stormwater quality (Yan et al., 2023).Using a data mining framework, Yan et al. (2023) found antecedent dry days, average rainfall intensity, and rainfall duration are the most critical rainfall characteristics affecting the EMCs of common stormwater quality parameters.Parajulee et al. (2017) highlighted the significance of antecedent dry days and rainfall depths in influencing the levels of PAHs in urban stormwater.Indeed, many rainfall parameters have commonly been used as predictors for stormwater quality in previous modelling studies (e.g., (McCarthy et al., 2013;Zhang et al., 2019)).However, it is uncertain whether rainfall characteristics affect TrOCs concentrations in stormwater.
This study aimed to conduct a comprehensive literature review on stormwater quality focusing on the occurrence of TrOCs and compile the literature data for quantitative assessment of (1) the concentration levels, (2) statistical distributions, (3) ecological and health risks, and (4) the influence of catchment land uses, as well as rainfall characteristics.The findings of this study provide a comprehensive dataset that can be used to establish targeted and effective management strategies for TrOCs in stormwater to protect the ecological health of urban water bodies or human health during harvesting practices.The dataset will also be helpful for future data-driven analysis studies.The outcomes can also support further development of evidence-based guidelines to ensure the safety of stormwater harvesting concerning TrOCs.

Systematic literature review
A systematic literature review was conducted to find studies on organic TrOCs in stormwater from urban catchments (Fig. 1).Widely used scientific searching platforms (Scopus and Web of Science) were used to find relevant papers using pre-determined keywords (Search Term 1 in Supplementary Information 1 (SI_1) Tab 1).We also looked for stormwater quality publications that reported untreated stormwater inflow into treatment systems (Search Term 2 in SI_1 Tab2).Thereafter, multiple rounds of screening were conducted to select the relevant publications according to the exclusion criteria shown in Fig. 1.To ensure the robustness of the systematic review, two individuals independently conducted the entire review process and cross-checked the data collection process.
We also reviewed data from the Stormwater BMP Database (https:// bmpdatabase.org/),and the sources were carefully checked to avoid duplication with other publications.Unpublished stormwater quality data (from three sources in Australia) were also acquired from project partners.The final list of the 116 publications/sources is provided in SI_1 Tab 2.

Data collection
Concentration.Most of the publications reported the concentration data in the form of descriptive summary values for specific monitoring sites (based on multiple rainfall events), e.g., mean & standard deviation, medium, minimum, and maximum, or various percentiles (thereafter named "descriptive summary data").Fewer studies reported the raw concentrations of monitoring sites from individual events (termed as "event-specific data").Also, event mean concentrations (EMCs) were frequently reported ("EMC data"), in comparison to intra-event discrete concentrations (named "discrete data").It should be noted that descriptive summary or event-specific data could be based on both EMC and/or discrete data.This made the analysis of the whole dataset more complicated.Still, it also provided opportunities for statistical analysis of the impact of data reporting approaches on concentration levels, something that has not previously been examined for such data in the literature.
Catchment characteristics.This study categorized the land uses into agriculture, commercial, industrial, open space, roads/transportation (light, medium, and heavy traffic roads, as well as parking lots), residential (low, medium, and high density), and roof (various materials) and mixed.In many cases, individual catchment or monitoring sites did not consist solely of a single land use type, except for roofs.Therefore, when the dominant land use exceeded 50 % of the catchment, it was categorized as that specific land use.If no single land use exceeded 50 %, it was classified as a mixed catchment.
For transportation land uses, the classification of heavy, medium, and low-density traffic followed the suggestion by "American Association of State Highway and Transportation Officials", i.e., heavy density with average daily traffic (ADT) of over 50,000 vehicles may be considered high, low density with ADT of less than 2000 vehicles, and otherwise medium density.Not every paper reported ADT; in these cases, the transportation land use was marked as 'unclassified', unless the paper specifically mentioned a low/medium/heavy traffic road (qualitatively without the ADT data).We treated 'parking lots' as an individual type of transportation land use, as many studies looked at stormwater runoff from pure parking lots.
For residential land uses, the classification of low, medium, and highdensity areas often varies depending on the local context used by urban planners, researchers, or government agencies.The specific criteria were usually not reported in reviewed papers.Thus, we used very general criteria to define sub-categories of residential land use: high-density residential areas are characterized by tall apartment buildings, condominiums, or mixed-use developments with multiple floors.Low-density residential areas are characterized by a relatively low number of housing units spread over a larger land area; common examples include singlefamily houses with private yards.Medium-density residential areas -balance between high-and low-density residential zones, which may include a mix of single-family houses, townhouses, and low-rise apartment buildings.In the case when a paper did not mention the land use type, but the accurate monitoring site location was reported, we tried to identify the residential land use type following the above criteria using Google Maps.Otherwise, it was considered as 'unclassified'.
Rainfall characteristics.We recorded rainfall depth, duration, and antecedent dry-weather period (ADWP) where available in the reviewed studies.Although rainfall intensity was also found to be a significant factor for other water quality parameters, it was not regularly reported in the collected papers, and thus insufficient data was collected.

Data analysis 2.3.1. Distribution fitting and statistical analysis
The distribution fitting began using the entire dataset.The concentration data for each TrOC (minimum ten points) were tested for ten common distributions, namely chi2, expon, logistic, gamma, lognorm, norm, powerlaw, weibull_min, weibull_max, and uniform.The Kolmogorov-Smirnov (KS) test was used to determine the goodness of fit of each distribution to the observed data.The KS p-value, which measures the probability that the observed data and the fitted distribution come from the same underlying distribution, was used to compare the fitting of different distributions.A higher KS p-value indicates a better fit.
It should be noted that there is a considerable amount of leftcensored concentration data (i.e., lower than the limit of detection, LOD).These were excluded from the distribution fitting tests, which aimed to compare different distributions.This may lead to potential issues, e.g., biased estimation, loss of information, etc.To account for this, the summary statistics in this study were computed by using regression on order statistics (ROS) (Helsel, 2012).The r 2 of ROS analysis is presented in SI_1 Tab 3.

Ecological and health risk assessment
2.3.2.1.Ecological risk assessment.The ecological risks from TrOCs were evaluated using the ecological risk quotient (ERQ), following similar methods used in Biel-Maeso et al. (2018) and Lu et al. (2018).Briefly, ERQ is calculated as the concentration divided by the predicted no-effect concentration (PNEC) of each TrOC (Eq.( 1)).
This study used the 95th percentile concentration to represent the upper range or high-end concentrations, providing a conservative estimate of risks to account for the risks adequately.Similar previous studies also used precautionary levels of concentrations for risk estimations, e. g., (Mutzner et al., 2022) and (Biel-Maeso et al., 2018).PNEC was calculated using the lowest values of acute EC 50 (effective concentration for causing adverse effects to 50 % of test organisms) divided by a default assessment factor (AF) (Eq.( 2)).EC 50 values from two different trophic levels, e.g., fish (Fathead minnow) and invertebrate (Daphnia magna, Tetrahymena pyriform), were derived from the Toxicity Estimation Software Tool -T.E.S.T. (USEPA, 2020).AF values were set as 100 and 50 when the EC 50 values from one and two trophic values, respectively.
2.3.2.2.Health risk assessment.A four-step risk assessment procedure was used to estimate the health risks of TrOCs, similar to that used in Fang et al. (2021) and Ma et al. (2017).
Step 1 -Hazard identification estimates the TrOCs concentrations in stormwater.
Where C donates the 95th percentile concentration (μg/L), IR is the volume of water ingestion rate during each exposure event (L/event), EF refers to exposure frequency (event/year), ED is exposure duration (years, 70 for an adult), BW is body weight (kg, 70 for an adult), AT is average lifespan, 25,550 days for an adult.IR and RF values for different exposure scenarios are summarized in Table S1 of Supporting Information 2 (SI_2), and details can also be found in Fang et al. (2021).
Step 3 -Dose-response assessment determines reference dose (RfD, μg/kg/day), which was acquired from the Integrated Risk Information System database (IRIS, 2023).When the RfD value was unavailable, it was estimated using the LD50 values (Eq.( 4)) acquired from T.E.S.T. (USEPA, 2020).LD50 is the lethal dose to 50 % of test organisms in the test population (typically based on laboratory rats), representing the acute median lethal dose of the target compound.The same method was also used by Lu et al. (2018) to estimate the health risks from antibiotics.
In the last Step 4 -Risk characterization, the hazard risk quotient (HQ) was estimated by dividing the CDI by RfD values (Eq.( 5)).
This study aggregates the Hazard Quotients (HQ) from all reuse pathways to assess the highest potential health risks.The ecological and health risks were both classified to be insignificant risk with ERQ or HQ < 0.01, low risk with 0.01 ≤ ERQ or HQ< 0.1, moderate risk with 0.1 ≤ ERQ or HQ < 1, and high risk with ERQ or HQ ≥ 1.The values of 95th percentile concentrations, EC 50 , AF, LD50, and RfD for all TrOCs are provided in SI_1 Tab 3 (TrOC Summary).

Land use impact analysis
We used the Kruskal Wallis Test to investigate if the TrOC concentrations were significantly different between different types of land uses.In the case when a significant difference was identified, Dunn's test was then used to evaluate whether there was a statistically significant difference between two land uses.Each catchment was treated as equally important; therefore, to avoid the unbalanced number of data points taken from different catchments, only the statistical summary data was used for land use impact analysis.In addition, a minimum of five data points were required for a specific land use to be included in this analysis, otherwise the land use type was excluded from the analysis.

Rainfall impact analysis
We conducted both Pearson correlation (indicating a linear relationship) and Spearman correlation (indicating a non-linear relationship) analyses to assess the relationships between TrOC concentrations and rainfall characteristics (including rainfall depth, duration, and dry period).As rainfall characteristics were event-based, this analysis used only the event-specific measurements of the data from each catchment (a minimum of five data points were used for analysis).It should be acknowledged that pollution levels may be correlated with multiple rainfall parameters (e.g., (Yun et al., 2023)).However, multi-regression analysis was not performed due to insufficient event-specific data.

Occurrence of TrOCs in stormwater
This review found a total of 629 TrOCs detected at least once (i.e., >= limit of quantification) across multiple categories, reported in sources (Fig. 2).SI_1 Tab 3 provides descriptive statistics of all the TrOC concentrations, source references, identities (InChiKey and CAS), and uses.
Pesticides.Due to the widespread application of various pesticides (including herbicides, insecticides, fungicides, biocides, etc.) in urban areas (Chen et al., 2019), pesticides have become a concern in urban stormwater over the past decades, with their presence frequently reported in many studies (Polkowska et al., 2009;Weston et al., 2009;Rippy et al., 2017).We identified that, in total, 228 pesticides have been detected at least once in urban stormwater (Fig. 1).Among those, pesticides have a considerable amount of data documented in the literature (> 50 points, SI_2 Tab 3), with the top five being diuron, simazine, atrazine, metolachlor, and 2,4-D.Rippy et al. (2017) monitored pesticides in various urban catchments in Australia and observed that diuron, 2,4-D, and simazine were detected in >50 % of stormwater samples.Mutzner et al. (2022) also found that diuron, atrazine, metolachlor, and 2,4-D were frequently detected in urban stormwater of various catchments worldwide.However, the median concentrations of these five pesticides were relatively low (< 0.1 μg/L).Nonetheless, their 95th percentile concentrations, except for Metolachlor, exceeded 0.1 μg/L, which serves as the precautionary threshold for pesticides in the EU Drinking Water Directive (EU Regulation 98/83/EC).Bifenthrin, reported in multiple catchments (Weston et al., 2009;Ensminger et al., 2013;Masoner et al., 2019), was found to have the highest estimated 95th percentile concentration (71.3 μg/L) among all pesticides with sufficient datasets (>10 points).Bifenthrin is a pyrethroid insecticide that is commonly used in urban areas for pest control and was previously found to cause toxicity in urban stormwater wetlands (Jeppe et al., 2017), as well as river water column toxicity following stormwater events (Weston and Lydy, 2012).
PAHs were of great interest among the previous studies, especially relating to road runoff, as transportation activities are one of the primary sources of PAHs (e.g., (Flanagan et al., 2018;Beryani et al., 2023)).We found 29 different PAHs in stormwater (including total PAHs) reported in over 40 papers, with 20 of them having a considerable amount of data (> 50 points, SI_1 Tab 3).Among all PAHs, O terphenyl had the highest median concentrations (80.4 μg/L), though it was reported in only one study that investigated the stormwater runoff from porous asphalt pavement (Jayakaran et al., 2019).Benzo(a)pyrene, one of the most carcinogenic PAHs, has been detected in many studies, and its reported medium and 95th percentile concentrations were 0.112 μg/L and 4.039 μg/L, respectively.These levels exceed the threshold of many standards for drinking water, e.g., in Australia (NHMRC, 2011) and EU (EU, 1998) -both set as 0.01 μg/L.Flame Retardant.Numerous flame retardants are also detected in stormwater (in total 28 -SI Tab 3).Among them, many are polybrominated diphenyl ethers (PBDEs), with BDE-209 displaying the highest concentrations (0.251 μg/L) reported in Gasperi et al. (2014).
Plasticizer is another group with a decent number of compounds found in stormwater (in total 24, SI Tab 3), most of them are phthalates.DEHP, the most common plasticizer, has been detected in numerous studies (23 studiesrefer to SI Tab 3 and Tab 2), has its highest concentration of 130 μg/L observed in Flanagan et al. (2018).Its 95th percentile concentrations (37.9 μg/L) also far exceed the threshold value of multiple drinking water guidelines (from 6 to 10 μg/L) (USEPA, 2009;NHMRC, 2011;WHO, 2011).Bisphenol A is another plasticizer commonly used in the production of various plastics, and it is widely found in many stormwater monitoring studies, e.g., (Kalmykova et al., 2013;Flanagan et al., 2018;Masoner et al., 2019) (full list refer to SI Tab 3 and Tab 2).Its 95th percentile concentration was 2.43 μg/L, with maximum value (12.7 μg/L) found in (Kalmykova et al., 2013;Flanagan et al., 2018).
PCBs (polychlorinated biphenyls) are toxic pollutants that are persistent in the environment and thus have been prohibited worldwide since the 1970s.Despite this, PCBs were still found to be present in stormwater (Hwang and Foster, 2008;Jaradat et al., 2010;Zgheib et al., 2012).Masoner et al. (2019) It was found that PCBs were detected less frequently and at lower concentrations than other types of organic chemicals in stormwater, possibly due to their limited use in recent years.The most frequently detected one was PCB 138 (18 % samples), with a maximum concentration of 0.108 μg/L.
Other industrial / Chemical intermediates / Solvent.A vast number of other organic TrOCs are present in urban stormwater.Those included many monocyclic aromatic compounds (e.g., benzene, ethylbenzene, toluene), halogenated aliphatic hydrocarbons (e.g., Chloroform, 1,4-Dichlorobenzene) that can also be highly toxic to human health and the environment.For example, 6PPD-quinone, a rubberderived compound that has been recently discovered and was reported to be acutely toxic, was also found in stormwater (Challis et al., 2021).Challis et al. (2021) detected mean concentration levels of 600 ng/L in stormwater samples (57 % detection rates), and between 80 -370 ng/L in snowmelt samples (80 % detection rates), posing risks to many aquatic organisms.

TrOCs concentration distributions
Following the Kolmogorov-Smirnov (KS) test of the reported concentrations for all TrOCs, we found that log-normal distribution had the best fit, evidenced by the highest KS p-values compared to the other nine different distributions (Figure S1 in Supporting Information 2 (SI_2); also refer to SI_1 Tab 4 for the fitted log-normal distribution parameters for TrOCs -including only those TrOCs with over 10 data points).This is consistent with many previous studies (e.g., Duncan (1999) and Métadier and Bertrand-Krajewski (2012)) found that event mean concentrations for TSS, nutrients, and chemical oxidation demand (COD) were approximately log-normally distributed.This current study supports that concentrations of TrOCs also follow log-normal distributions well.When evaluating the influence of data reporting on specific lognormal distributions, we identified differences for some TrOCs (e.g., DEHP and Benzo(a)pyrene), while very minor differences or some other ones (PFOA and diuron) (refer to SI_2 Section 2 for detailed methods and results).Nevertheless, log-normal distributions typically provide the most accurate descriptions of the concentrations of various TrOCs.

Ecological and health risks from TrOCs in stormwater
Varying levels of ecological risks (ERQ) and health risks (HQ) were estimated, as shown in Fig. 3 (full data presented in SI_1 Tab 5).As expected, ecological risks were more severe than the health risks caused by the TrOCs in stormwater.82 TrOCs were found to pose high ecological risks (ERQ ≥ 1.0).These include 42 pesticides, 10 PAHs, 8 plasticizers, 8 PCBs, 6 flame retardants, 5 Other industrial / Chemical intermediates / Solvent and 3 PPCPs.A previous study by Mutzner et al. (2022) found that 24 organic TrOCs had high environmental risks, with PAHs having the highest risk quotient values.This current study further reveals that a much wider range of organic compounds can cause high ecological risks.For example, very severe ecological risks could be caused by a range of pesticides, i.e., bifenthrin, bromophos, diazinon, dichlorvos, malathion, sulfotep, as well as other TrOCs such as DINP (Bis (3,5, phthalate; a plasticizer) and BED 209 (decabromodiphenyl ether; a flame retardant), with estimated ERQ all greater than 100 (Fig. 3).The highest ecological risk was observed for bifenthrin (ERQ = 2400), a widely used insecticide in urban areas for pest control, that was detected in a large number of urban sites (Weston et al., 2009;Ensminger et al., 2013;Masoner et al., 2019).Hence, these results show that organic TrOCs are frequently released into the environment, meaning that they have a high likelihood of exposure to a diverse range of wildlife species (Spindola Vilela et al., 2022;Gkotsis et al., 2023).In fact, many previous studies have found the adverse effects of organic chemicals (e.g., pharmaceutical compounds) on aquatic organisms, such as fish, invertebrates and algae, even at trace levels (Biel-Maeso et al., 2018;Lu et al., 2018).
On the other hand, three reported pesticides were found to cause high health risks (HQ≥1.0,Fig. 3): Heptachlor epoxide, EPN (Ethyl pnitrophenyl benzenethionophosphonate), and Aldrin (all these three also had EQR≥1.0).While concerns exist regarding organic TrOCs in drinking water, human exposure to these compounds is relatively low compared to environmental exposure.It's worth noting that a value of 2000 mL/event was employed for direct drinking in the risk assessment, aiming to depict a worst-case scenario (refer to Table S1 of SI_2).This suggests that the actual health risks associated with Trace Organic Contaminants (TrOCs) would typically be minimal.Additionally, these high-risk pesticides were detected in just a few stormwater catchments, e.g., Aldrin and Heptachlor epoxide were found in Poland (Polkowska et al., 2001;Polkowska et al., 2009), USA (Cole et al., 1984), and Australia (data not published), while EPN was only detected once in an Australian catchment.These results strongly support the feasibility of stormwater harvesting, as even for activities involving human contact, the probability of health risks is generally minimal in most catchments.Additionally, harvesting stormwater is known to bring multiple benefits, as it minimizes the release and exposure of pollutants into the environment, thus reducing the ecological risks of receiving water.
Nevertheless, it should be noted that the relevant health risk data for TrOCs is quite limited.In fact, Reference Dose (RfD) values used in this study for only 141 TrOCs (SI_1 Tab 3), were from the Integrated Risk Information System (IRIS) database (IRIS, 2023).These values were derived through careful evaluation of available toxicity data, including K. Zhang et al. animal studies and human health studies, considering factors like uncertainty and variability in the data.For the other TrOCs, the RfD values were estimated using the LD50 values (acute median lethal dose of target compound), which were typically estimated through controlled laboratory experiments on animals (Strenge and Peterson, 1989).Thus, this study recognizes the generally constrained availability of health risk data (e.g., the RfD values) for TrOCs.In addition, our knowledge of the occurrence of many toxicologically relevant chemicals in urban stormwater is also very limited, as shown in Mutzner et al. (2023).Thus further research is required to understand potential risks to human health better.

Influence of land use on TrOCs levels
Across hundreds of TrOCs, 68 had sufficient data for statistical analysis of the land use impact on the concentrations.The results showed that land use had statistically significant impacts on the concentrations of 14 TrOCs (Kruskal Wallis test, p < 0.05) (distribution plots shown in Specifically, six polycyclic aromatic hydrocarbons (PAHs), including acenaphthylene, acenaphthene, fluorene, naphthalene, phenanthrene, and Total PAHs, had their highest or second highest median concentrations in heavy traffic road areas, and the differences were often statistically significant compared to other land uses (p < 0.05) (Fig. 4), while no statistical differences were found across the remaining 47 TrOCs (p > 0.05).
Fig. 4).It was found that major roads were the dominant source of traffic-related substances such as PAHs (Wicke et al., 2021).Burant et al. (2018) detected higher concentrations of total PAHs in a commercial site due to the higher vehicle traffic volumes compared to a high-density residential site.Gasperi et al. (2014) observed significant differences in total PAHs across three different urban catchments, with a light-density residential catchment (but with the highest reported traffic loads, >60,000 vehicles/day) showing higher concentrations than another residential catchment and one industrial catchment.The current study also observed that the mix land use was among the land uses displaying statistically higher concentrations, except for total PAHs.Additionally, concentrations of phenanthrene in commercial and high-density residential areas were not statistically different from those in heavy/medium traffic roads and mix land uses (p > 0.05) but were significantly higher than other land uses (p < 0.05).For total PAHs, their levels in runoff from parking areas were close to heavy traffic areas (p > 0.05) and significantly higher than all other land uses (Fig. 4).Significant impacts of land-use types were also found for five pesticides, namely atrazine, bifenthrin, DEET, simazine, and triclopyr (p < 0.05, Fig. 4).This study found that mix land use was always within the group with significantly higher concentrations than other groups of land uses, with the highest median concentrations found for atrazine and simazine.High-density residential areas had concentrations of bifenthrin and triclopyr that were not significantly different from the Mix land use, but higher than other land uses.Interestingly, the low-density residential areas had significantly lower levels of three pesticides (atrazine, bifenthrin, DEET) while also having statistically higher concentrations of simazine (P < 0.05).In fact, these five pesticides have been commonly used within urban areas, e.g., grassy weed control (atrazine and simazine), woody plants and broadleaf weed control (triclopyr), and insect repel (bifenthrin and DEET), which can probably explain their presence in different land uses at varying concentrations.Wicke et al. (2021) found the highest isoproturon concentration in catchments dominated by K. Zhang et al. one-family homes (e.g., low-density residential), which was likely from the use in private gardens.Mecoprop (commonly used in roof materials) and diuron (typical additives to exterior paints) exhibited higher concentrations in residential blocks with 4-6 storeys old buildings (classified as high-density in this study), and glyphosate/AMPA (used to control weed at sidewalks) appeared in most land uses at similar concentrations.By monitoring a high-density residential site and a commercial strip, Burant et al. (2018) observed higher EMCs of atrazine, metolachlor, and carbendazim in the residential site, but higher DEET in the commercial site.However, no statistical differences were found for total pesticides in these two sites.
DEHP (Di(2-ethylhexyl) phthalate) was observed to have significantly higher concentrations in commercial, high-density residential, mix and heavy-traffic road areas than that in low-density residential and roof catchments (p < 0.05, Fig. 4).As a common plasticizer, it was not surprising to detect high levels of DEHP in commercial areas, due to the high consumption of plastics products.DEHP can also be found in automotive compounds such as car interiors, dashboard materials, and wiring, and thus high heavy-traffic areas could also have large amounts of DEHP.Indeed, the study done by Wicke et al. (2021) showed high correlations between phthalates (including DEHP) and traffic counts based on data from five different sites (versus 63 sites analyzed in this study, including 11 sites with transportation land use).Large buildings in commercial areas, as well as high-density residential areas with plastic-based interiors or exteriors, can also be a source of DHEP, which can explain its elevated levels in commercial and high-density residential areas.
TBEP and TDCPP, both used as organophosphate flame-retardants (OPFRs) and plasticizers, were found to have significantly higher concentrations in mix and high-density residential (only for TBEP) areas (Fig. 4) than other land uses, such as commercial and low-density residential areas.In a monitoring study done by Burant et al. (2018), significantly higher concentrations of OPPFs (total and individual ones, i.e., TCPP, TCEP, TDCPP, TBEP and TPP) were found at a high-density residential site than at a commercial site.TDCPP was an exception (higher concentration in the commercial site), but it was detected at much lower concentrations than other chlorinated OPFRs).Opposite results were found by Wicke et al. (2021), who showed commercial areas having the highest concentrations of flame-retardants.Their study, however, also found densely built-up land use types (e.g., high-density residential) contributed to higher TBEP concentrations than low-density areas and road catchments.

Impact of rainfall on TrOCs levels
The Pearson and Spearman correlation analyses identified diverse effects of rainfall characteristics on certain organic chemicals (Fig. 5).Rainfall depth exhibited both positive (r > 0) and negative (r < 0) influences on the concentration of 12 organic chemicals (p < 0.05), primarily demonstrating non-linear relationships (Fig. 5a).Specifically, it is interesting to find that many wastewater-derived PPCPs (e.g., Methyl 1H benzotriazole, cotinine, beta Sitosterol, and nicotine), as well as MCPA and 2,4-D, were negatively correlated to the rainfall depth (r < 0, p < 0.05).This may indicate a possible dilution of those compounds under high rainfall.For example, Benotti and Brownawell (2007) observed that heavy precipitation, leading to a combined sewer overflow event, resulted in decreased or even eliminated concentrations of certain pharmaceuticals in an urban estuary.On the other hand, three PFAS (PFHxA, PFHpA, and PFOS), two pesticides (Dithiopyr and DEET), and total PAHs were positively correlated to rainfall depth (r > 0, p < 0.05).Previous studies also demonstrated that stormwater could induce pollutants loads of trace organic contaminants, with higher concentrations observed during urban rainfall (Parajulee et al., 2017;Burant et al., 2018;Yun et al., 2023).In particular, for PFAS, one major pathway to urban waterways is through atmospheric deposition (Kim and Kannan, 2007) and thus, higher rainfall could potentially lead to high concentrations in runoff for PFAS (Xiao et al., 2012).
Rainfall duration displayed a significant correlation with only three TrOCs (p < 0.05) (Fig. 5b).Rainfall duration was previously found to be an influential factor affecting the concentrations of common stormwater quality parameters, such as suspended solids and nutrients (Yan et al., 2023).It is often noted that particle-associated pollutants in stormwater are generally more affected by rainfall duration (Yuan et al., 2017).Indeed, these three TrOCs, dithiopyr, benzo(k)fluoranthene, and benzo (a)pyrene, are all hydrophobic compounds (i.e., LogK oc >4.5) and can easily attach to particles in stormwater.
The length of dry periods exhibited significant correlations with seven organic chemicals (p < 0.05) (Fig. 5c).The length of dry periods or antecedent dry periods has always been regarded as a major driver of stormwater pollution (Parajulee et al., 2017;Zhang et al., 2021), e.g., based on data-driven analysis, Yun et al. (2023) identified that antecedent dry hours significantly influenced micropollutant dynamics.Often pollutants accumulate over dry periods, and thus a positive correlation would be expected, e.g., in this study, PFNA, PFNA, PFOA, MCPA, 2,4 D, diuron (Fig. 5c).An exception was found for anthraquinone (negative correlations observed, r < 0), which was reported in three studies (Burant et al., 2018;Fairbairn et al., 2018;Masoner et al., 2019).Anthraquinone is commonly used in dye production, and thus may not be a stormwater-derived chemical.
Overall, it is noted that the majority of the correlations were relatively weak (abs(r)<0.5),except for the PFAS (e.g., PFOA and PFOS), where the correlations with rainfall depth and dry period were stronger r > 0.8) (Fig. 5).This is likely due to the diverse sources of TrOCs, and their transportation into stormwater through complex rainfall-runoff processes.Thus, further investigations are still encouraged to reveal specific correlations.

Enhancing stormwater quality management
The occurrence of over 600 TrOCs in stormwater across multiple categories demonstrates the ubiquitous presence of many TrOCs in every catchment and provides evidence that stormwater is a significant pathway for introducing TrOCs into the water environment.The descriptive summary data of all the identified TrOCs (SI_1 Tab 3), or the generated log-normal distributions of TrOCs (only for 110 TrOCs with over 10 data points, SI_1 Tab 4), could serve as a starting point to aid in setting appropriate treatment targets, designing effective treatment systems, or conducting further data-driven analysis.This study highlights potential ecological and human health risks associated with TrOCs pollution in stormwater.Communicating our findings to the public and stakeholders can increase awareness about this issue.Specifically, the dataset could inform the development or revision of stormwater harvesting guidelines to specifically address the concerns of TrOCs, like the Australian Guidelines for Water Recycling: Stormwater Harvesting and Reuse (NHMRC, 2009), which currently overlooks chemical risks due to data limitations.
We also acknowledge the limitations of the systematic literature review.While it identified over 600 TrOCs in urban stormwater, this likely represents only a portion of potential contaminants in stormwater runoff.Emerging pollutants, including transformation products, may not have been captured in previous monitoring studies.Future studies should adopt more comprehensive monitoring to account for the evolving landscape of contaminants.

Removing TrOCs through stormwater treatment systems
This study highlights the need for effective strategies and pollution control measures, to safeguard the quality of our water resources and protect the health of aquatic ecosystems and human health.Researchers have started investigating the treatment of TrOCs in different naturebased solutions (NBS), e.g., bioswales (Flanagan et al., 2018;Fardel et al., 2020), biofilters (Zhang et al., 2015;Gu et al., 2021;LeviRam et al., 2022), constructed wetlands (Mauffrey et al., 2017;Zhao et al., 2019) and porous pavements (Charlesworth et al., 2017;Jayakaran et al., 2019).Although many TrOCs may have high risks (e.g., DEHP: ERQ = 21.5),some compounds are characterized as high hydrophobicity (i.e., logK oc >4.0) and relatively short half-lives (i.e., <40 days in freshwater), and thus could be effectively removed in stormwater control measures (e.g., stormwater biofilters (Zhang et al., 2014)).Nonetheless, highly hydrophobic TrOCs with slow degradation, such as PFAS, may result in long-term accumulation, necessitating thorough investigation.Some TrOCs could be highly mobile and/or persistent, and thus less likely to be removed by currently available treatment measures, as found by Spahr et al. (2020).By further looking into the properties of the 82 TrOCs of either high health or ecological risks (full list in SI_1 Tab 5), 76 could be identified as persistent, mobile, and toxic (PMT) substances, based on the criteria of mobile (logK oc <4) or persistent (degradation half-life in soil > 120 days) according to the REACH guidelines (Arp and Hale, 2019).Even further, 40 were either very mobile (vM, log Koc <3.0) or very persistent (vP, degradation half-life in soil > 180 days).These raise concerns about the effectiveness of current nature-based treatment measures (or Nature-Based Solutions, NBS) in removing those pollutants.Further studies are recommended to develop specific NBS for the effective treatment of these PMT substances that pose high health and ecological risks identified in this study.This can include potential studies on identifying promising media amendments to existing stormwater treatment systems, e.g., bio-char (Mohanty et al., 2018;Ashoori et al., 2019) and activated carbon (Ulrich et al., 2015).Future studies are also required to explore the feasibility of advanced oxidation processes for further purification of stormwater post-NBS, which could be quite effective for specific types of TrOCs (Duan and Sedlak, 2021;Ochir et al., 2021;Zheng et al., 2021;Ferreira et al., 2022;Zheng et al., 2022).

Catchment specific management
The identification of the significant impact of land use in 14 TrOCs (out of 68 ones that had sufficient data for analysis) emphasizes the potential need for catchment land use targeted management strategies.Our findings generally have good agreement with the monitoring studies, with the source of contaminant usage being a plausible predictor for their presence in stormwater (Wicke et al., 2021).Specifically, catchments with heavy traffic are generally sources of various PAHs, diuron, and DEHP.High-density residential areas often had elevated levels of several PAHs and pesticides, as well as DEHP and TBEP, while commercial areas showed similar findings but for different TrOCs.Given the presence of multiple sources for TrOCs in mixed land uses (found to be of great concern for all the 14 TrOCs), there is a need for integrated management approaches that consider the diverse sources of contamination.Often, low-density residential land use was found to have significantly lower levels majority of the TrOCs, except for simazine.This could indicate that harvesting stormwater from low-density residential areas could be quite promising, but care should be considered with regard to the use of some specific pesticides in private green areas.

Linking to rainfall characteristics
The correlation analysis between rainfall characteristics and pollutant concentrations reflected the complex interactions between rainfall-runoff processes and the transportation of diverse TrOCs.This study was limited by data availability; future research endeavors could aim to provide further insight into the interplay of multiple rainfall variables on stormwater TrOC concentrations.Formulating specific strategies that link rainfall characteristics with stormwater management would need additional investigations, especially at the catchment level, to understand these multifaceted interactions between the use and application of these TrOCs and the mechanisms of their mobilization and transport.Understanding these interactions may help design sophisticated real-time controlled treatment systems, capable of bypassing, for example, flows likely to be particularly polluted.

Conclusion
This paper sheds light on the extensive presence of trace organic chemicals (TrOCs) in urban stormwater, emphasizing the importance of systematic assessment.The detection of 629 TrOCs, encompassing pesticides, pharmaceuticals and personal care products, PAHs, PFAS, flame retardants, plasticizers, PCBs, corrosion inhibitors, and industrial chemicals/intermediates/solvents, underscores the complexity of urban stormwater contamination.TrOC concentrations reveal a log-normal distribution pattern, while risk assessment identifies 82 TrOCs with high ecological risk and three with potential health risks.Notably, specific TrOCs exhibit significant associations with land-use types, emphasizing the need for targeted management strategies for catchment land use.Despite relatively weak relationships between rainfall characteristics and pollutant concentrations, this study provides essential

Fig. 1 .
Fig. 1.Key steps and exclusion criteria of the systematic review of urban stormwater quality.

Fig. 4 .
Fig. 4. Concentration distribution plots of 12 TrOCs that had significant impacts by land uses.The numbers next to each land use in each sub-plot indicate the number of data points from each land use.The red asterisk '*' next to a pair of land use categories indicates the significant difference following Kruskal-Wallis and Dunn's test (p < 0.05).Taking Fluorene as an example, the '*' next to Mix and Road_H, means that the Fluorene concentrations in these two categories were significantly different from each other compared to other categories (i.e., Com, Res_H and Res_L).When only one '*' is presented, it refers to that specific land use type significantly different from others.Each box shows the interquartile range (IQR, Q1 to Q3), with the median value represented by the vertical line within each box; the whiskers extend to show the rest of the distribution, from 'minimum' (Q1-1.5*IQR, the left line) to 'maximum' (Q3+1.5*IQR, the right line).

Fig. 5 .
Fig. 5. Results of the correlation analysis for concentrations of organic TrOCs : correlation coefficients (r values) are provided exclusively for significant correlations (p < 0.05).The number in parentheses denotes the quantity of points employed for analysis.Positive r values denote a positive correlation, implying that higher parameter values align with higher TrOCs concentrations.Conversely, negative r values indicate an inverse relationship.