The sampling sites were selected based on land cover and land use characteristics. Four watershed types were selected: urban, agricultural, mining, and oil and gas (Fig. 1). W1 is an urbanized watershed (89.9 % developed; 6.10 % forested) and is the smallest in size at 39 km² (Dewitz, 2021). W2 is 660 km² and its land use consisted of approximately 8.31 % cultivated cropland (37.2 % corn, 32.5 % soybeans, and 9.20 % winter wheat; USDA-NASS, 2023), 11.7 % hay/pasture, 67.0 % forested (59.8 % deciduous forest), and 4.97 % developed area. There were approximately 40,700 cattle throughout the three counties that W2 spanned (USDA, 2022). Portions of the watershed also lie within the Western Coal Field of Kentucky resulting in approximately 4 % of the area defined as inactive (3.49 %) or reclaimed mine land (0.749 %). W3 was 484 km² and defined as 78.4 % forested (71.4 % deciduous forest, 6.90 % mixed forest), 8.50 % herbaceous, 5.17 % developed, and 3.51 % barren according to the 2019 National Land Cover Dataset. However, 70.3 % of the watershed is defined as active, inactive, or reclaimed surface mining (Horn, 2021) in which native Appalachian forests do not reestablish on most post-mining landscapes, shifting the land cover from a forested ecosystem to grassland/shrubland (Pericak et al., 2018). W3 also had 58 oil and gas wells equating to 0.120 wells km⁻² (KGS, 2022). The watershed, W4, was the largest in area at 3320 km² and extended into Kentucky, Virginia, and West Virginia. The land cover for W4 consisted of 87.1 % forested, 4.86 % developed, and 6.98 % herbaceous, hay/pasture, and shrubs. Additionally, 44.3 % of W4 fell within a permitted mine boundary (Horn, 2021; WVDEP, 2017) with extensive mining and coal processing, and contained the highest number of oil and gas wells (5040) of the studied sites equating to 1.52 wells km⁻² (KGS, 2022; VaEnergy, 2022; WVDEP, 2016).

The watersheds also varied by geology and topography. W3 and W4 were in the Appalachian Mountain Plateau and had an elevation relief of 378 m and 868 m, respectively (USGS, 2022) whereas W1 and W2 had a relief of 57.9 m and 143 m, respectively, and were located within a karst landscape (Paylor and Current, 2001). The major lithologic constituents for W3 and W4 consisted of approximately 40 % sandstone, 35 % shale, 15 % siltstone, 10 % coal, and < 5 % limestone (Noger, 1988), while W1 was predominantly 90 % limestone and 10 % shale. In contrast, W2 was comprised of 70 % siltstone, 30 % sandstone, and some limestone and coal. As for soil type, W1 was predominantly composed of Bluegrass-Maury Silt Loam along with hydrologic soil group (HSG) B soils with moderate infiltration rates consisting of sandy loam or loam soils and HSG C soils with low infiltration rates consistent with sandy clay loams (NRCS, 2023). W2 was approximately 50 % HSG C (sandy clay loam) soils, 23 % HSG C/D (clay loam, silty clay loam, sandy clay loam) soils, and 17 % HSG B (sandy loam and loam) soils. The main soil classification was Zanesville Silt Loam. W3 was comprised of 40 % HSG A with high infiltration rates consistent with sand and loamy sand soil types, 35 % HSG B (silt loam and loam), and 25 % HSG C (sandy clay loam). The main soil classification for W3 was Shelocta-Highsplint-Gilpin complex in pre-mined or un-mined areas and Kaymine, Fairpoint, and Fiveblock soils in mined regions of the watershed (Williamson and Barton, 2020). The soil type for W4 was defined as 40 % HSG A (sand and loamy sand) soils and 55 % HSG B (silt loam and loam) soils with the main soil classification being Pineville-Berks association.

The average temperatures during the growing season for the study region were 7.20 to 24.4 °C, peaking in July (NOAA, 2023). The average annual precipitation total for Kentucky was 1190 mm and the most amount of precipitation occurred in July with an average of 135 mm (PRISM, 2022). W2 received an annual average of 1300 to 1372 mm of precipitation while the other three watersheds received 1220 to 1300 mm (PRISM, 2022).

Field sampling took place from March to October 2022 to analyze monthly contaminant signatures during the growing season and capture runoff associated with pesticide and fertilizer applications. Polar organic chemical integrative samplers (POCIS) and grab samples were used to collect water samples. POCIS is a type of passive sampler that can be used to determine a time-weighted average (TWA) of concentrations at a given sampling point by binding water-soluble polar organic compounds moving through the water column (Alvarez et al., 2004; Satiroff et al., 2020). The POCIS utilized in this study were hydrophilic-lipophilic balanced (HLB). POCIS canisters were deployed monthly at each site and placed perpendicular to stream flow in the middle of the channel and secured in the stream by chaining to a bridge or a nearby tree. Each sampling period lasted approximately one month (26–35 days, Table 1S). At the conclusion of each sampling period, the POCIS were retrieved and a new set was installed at the same locations. The POCIS were deployed seven times to determine monthly concentrations of neonicotinoids and strobilurin pesticides along with PPCPs (EPA 1694 Group 1). After retrieving the POCIS from the field, membranes were wrapped in aluminum foil to limit photodegradation that could occur post-sampling. To limit the membranes from drying out and further contaminant degradation, the membranes were placed on ice in a cooler and then stored at <0 °C until they were shipped for analysis. Sample holding time was <60 day from collection to analysis.

Grab samples were utilized to determine concentrations of glyphosate, trace elements, dissolved organic carbon (DOC), and nutrients (i.e., ammonia (NH₃), nitrate-N (NO₃-N), total nitrogen (TN), orthophosphate (PO₄-P), total phosphorous (TP), sulfate SO42-, bromide (Br⁻), chloride (Cl⁻)). Grab samples were taken at the beginning and end of each sampling period for a total of eight sample points. Additional grab samples were also taken if a POCIS canister was missing or damaged to help determine the concentrations of neonicotinoids and PPCPs at the time of sampling (Table 1S). Table 1 outlines the sample container, preservation, storage method, laboratory, and analysis used for each contaminant.

Additional stream characteristics (e.g., pH, dissolved oxygen, specific conductance, discharge, depth, water temperature, stream profile) were also measured at each sampling point. Specific conductance (SPC), pH, and dissolved oxygen (DO) concentrations were measured monthly at each site using a YSI handheld multimeter (YSI Incorporated, Yellow Springs, OH). United States Geological Station (USGS) gauge stations were located at W1, W2, and W4 sites and were utilized to determine continuous discharge and water depth. W3 did not have a nearby USGS monitoring gauge; therefore, a HOBO^® U20L04 Water Level Logger (Onset, Cape Cod, MA) was installed to collect continuous water depth measurements and site surveys using a SonTek River Surveyor^® M9 ADCP (Xylem, Washington D.C.) was used to determine stream morphology and velocity. Additionally, HOBO^® Pendant^® Temp/Light MX2202 (Onset, Cape Cod, MA) Loggers were installed at all monitoring locations to determine continuous water temperature and measure light fluxes on the POCIS to account for potential implications for photodegradation on the membranes.

Four POCIS membranes were deployed each sampling period at the four sites. Two membranes were composited and analyzed for neonicotinoids and strobilurin pesticides and the remaining two membranes were composited and analyzed for PPCPs. Pesticides and PPCPs samples were processed and analyzed using liquid chromatography-tandem mass spectrometry (LC-MS/MS; Snow et al., 2020). Isotopically labeled internal standards were used for calibration based on availability of compounds. Surrogates were added at the beginning of the processing to evaluate effects of ion suppression or enhancement from sample matrices. Fortified blanks, fortified matrix, and reagent blanks were run at a frequency ≥5 % each (1 for every 20 field samples). Compound recovery for the spiked standards (Table 3S) along with method detection limits (MDL; Table 2S) for each analyte are presented in the supplementary materials.

The water grab samples were either shipped or hand-delivered to the appropriate laboratories for analysis within two days of collection. The analysis method for each contaminant is listed in Table 1. Briefly, nutrients were analyzed using ion chromatography and followed methods SM 4500 for NH₃ and TN, EPA Method 365.3 for TP, and EPA Method 9056A for NO₃-N, PO₄-P, SO42- and Cl⁻ analysis. Dissolved organic carbon was analyzed with high temperature combustion using a total organic carbon analyzer and EPA Method 415.1. Trace elements were digested according to EPA Method 3015 A (SW-846) and analyzed for total recoverable element concentrations with an inductively coupled plasma-mass spectrometry (ICP-MS) based on EPA Method 200.8. Glyphosate was analyzed using ICP-MS following Guo et al. (2005). The remaining pesticides and PPCPs grab samples were analyzed using a high performance liquid chromatography and tandem mass spectrometry (HPLC/MS/MS) when samples were collected (Table 1S).

The analysis of the POCIS extracts resulted in a measured mass for each analyte per membrane. The TWA concentrations were determined using the recovered mass from extracts and experimentally determined uptake rates for each analyte: (1) Cw=NRs*t where Cw was the ambient chemical concentration (ng L⁻¹), N was the mass accumulation (ng), Rs was the experimentally determined uptake rates for POCIS (L d⁻¹), and t was the exposure time in days (d; Ahrens et al., 2015; Noro et al., 2020). Experimentally determined uptake rates were based on published values in the literature (Table 4S).

A statistical assessment was completed to determine relationships between sites, sampling date, and individual contaminant. All data were log-transformed and analyzed with R studio (RStudio, 2021) using one-way analysis of variance (ANOVA) with post hoc Tukey honest significance difference (HSD). Principal component analysis (PCA) was performed to describe the relationship between the land cover, physiochemical, and contaminant properties of the four watersheds. The log-transformed variables were standardized by subtracting each of the values by its mean and dividing them by the standard deviation to ensure each variable had the same level of contribution to the analysis. Then the covariance matrix, eigenvectors, and eigenvalues were calculated. The top two principal components were selected for data analysis. The square cosine value (cos2) for each variable concerning the first two principal components was used to compare the contribution of each variable.

Over the sampling period, the physiochemical characteristics of the stream (e.g., temperature, discharge, pH, SPC, DO) changed between each site. Overall, W4 had the largest discharge and W1 had the smallest discharge (Table 2), which was unsurprising given their differences in watershed areas (Galster, 2007). It is important to note several flooding events, during March, April, and August when W2 experienced high-water conditions where stream waters were in the floodplain during the sampling events. A major flooding event occurred during July 25th to 30th where an unprecedented 35 to 40 cm of rain fell over W3 (Table 2), equating to a 1 in 1000-year rainfall event (Kohli, 2022). Due to the flood, sampling during August at W3 and W4 did not occur due to limitations in access to the sites and concerns about safety. The POCIS and depth logger for W3 were lost in the flood, resulting in missing discharge and contaminant data for July.

Mean water temperature values for all the sites ranged from 10.6 °C in the early spring to 28.7 °C in mid-summer (Table 3). Water pH values also remained consistent throughout the sampling period ranging from 6.73 in March for W2 to 8.17 in July for W3. Overall, W1 and W2 had the lowest DO values throughout the sampling period (Sep_W2 = 1.78 mg L⁻¹; Oct_W1 = 3.61 mg L⁻¹) which were below a freshwater criteria limit (>5 mg L⁻¹; NREPC, n.d.). W3 had the highest DO values (Oct_W3 = 14.01 mg L⁻¹) whereas larger and faster-moving mountain streams tend to have higher DO with increased mixing and movement of the water (Canadell et al., 2020). There was an inverse relationship between DO and SPC and temperature where increased temperatures and SPC values resulted in lower DO levels. Morosanu et al. (2016) found similar results and concluded that the increases in DO resulted from the elimination of pollutants that cause turbidity, as shown by the decreased SPC values. As for temperature, cold water can hold more dissolved oxygen than warmer water driving seasonal variations (Rajwa-Kuligiewicz et al., 2015).

The SPC for W3 had significantly (α<0.05) higher values for all sampling periods. Surface mining operations in the Appalachian Mountains have been linked to increased salinity resulting in SPC values ranging from 68.0μscm-1 to 3050μscm-1, with the major ionic constituents in surface water receiving mining discharge being SO42-, alkalinity, magnesium (Mg²⁺), calcium (Ca²⁺), Cl⁻, and sodium (Na⁺; Armstead et al., 2016). Additionally, in the Central Appalachians, an SPC threshold of 300μscm-1 is deemed protective of aquatic biota (EPA, 2011), and all the samples collected in W3 and W4 were well above that level. As for W2, dramatic changes in SPC in October can be an indicator of anthropogenic pollution such as agricultural fertilizer runoff with increases in NO3- and PO43- or wastewater pollution with increases in Cl⁻ (de Sousa et al., 2014; Harwell et al., 2008) however, this was not the case. Rather, the site’s high conductivity in October could have been due to low water levels (0.029 m³ s⁻¹) concentrating ions in the stream.

Nitrogen (i.e., NO₃-N) and phosphorous (i.e., PO₄-P, TP) species were significantly (α<0.05) higher in W1 (89.9 % urban, 2.21 % agriculture) and second highest in W2 (4.97 % urban, 22.5 % agriculture; Fig. 2). While TN (0.900–1.82 mg N L⁻¹) was highest for W1, there was not a significant difference between the urbanized (W1) and agricultural (W2; 0.480–2.00 mg N L⁻¹) watersheds. Additionally, W2 had the highest concentrations of NH3+0.036–0.292mgNH3++L-1 but was not significantly different from W1 (0.011–0.109mgNH3+L-1). However, both sites detected concentrations of NH3+ that were above a chronic criterion for freshwater organisms (0.05 mg L⁻¹) except for May (0.036mgNH3+L-1) for W2 and March (0.011mgNH3+L-1), May (0.02mgNH3+L-1), September (0.049mgNH3+L-1) and October (0.049mgNH3+L-1) for W1 (EPA, 2014).

A wastewater treatment plant (WWTP) is present downstream of the W1 site, indicating that effluent discharge may not be contributing to nutrient concentrations in the urban stream, however, leaky sewage pipes and overflows with combined stormwater and wastewater networks have been shown to result in elevated levels of nutrients in the environment (Ellis and Butler, 2015). For all other sampling sites, a WWTP was upstream of the site (Table 1S). Increased nutrient concentrations in W1 are also due to non-point sources such as fertilizer applications to lawns, gardens, and recreational areas. The increased nutrient loads in W2 may also be attributed to fertilizer, livestock, and land-applied manure within the watershed (Gu et al., 2022) as well as point source discharge from wastewater treatment and septic systems (Katz et al., 2010).

W1 had significantly (α<0.05) higher concentrations of Cl⁻ (April =126 mg L⁻¹) compared to the other sampled watersheds throughout the entire study period (Fig. 3). Chloride is commonly found in highly urbanized systems and can occur from road salts, wastewater, fertilizer, animal waste, irrigation, energy production wastes, and landfill leachates (Granato et al., 2015). However, it is likely that elevated levels of Cl⁻ in the more urban watershed (W1) are mainly due to the dissolution of Cl⁻ from the limestone geology in the region (Krawczyk and Ford, 2005; Lang et al., 2006).

The watershed with the highest activity of mining had significantly (α<0.05) greater concentrations of SO42- and exceeded an ambient water quality criterion for freshwater of 500 mg L⁻¹ in 4 of the 8 months sampled (DEP, 2012; Fig. 4). In freshwater, SO42- is typically considered non-toxic, but at high concentrations, it can induce osmoregulatory stress in organisms (Soucek and Kennedy, 2005) while also contributing to the salinization of water bodies (Cañedo-Argüelles et al., 2013) and impact the biogeochemical processes of C, N, and P (Zak et al., 2021). Sulfate has also been found to increase Cl⁻ induced corrosion of water pipes, leading to the release of metals in a drinking system (Sun et al., 2017). The higher SPC values for W3 also corresponded to the variation in SO42-, indicating SO42- is the major ion driving the conductance of the stream. Elevated levels of SO42- in freshwater occur commonly in anthropogenically impacted waters and may be attributed to intensive mining and weathering of sulfide minerals and gypsum (Arkoc et al., 2016; Niu et al., 2018; Runtti et al., 2018). Several studies have reported high SO42- and SPC levels from mining activities in the same mined watershed examined in this study (Hutton et al., 2020; Hutton et al., 2021; Muncy et al., 2014; Price et al., 2018) and noted declines in salamander species abundance and occupancy within streams. Arkoc et al. (2016) measured SO42- ranging from 4 to 756 mg L⁻¹ in surface waters downstream of coal mining in Turkey. Niu et al. (2018) found that SO42- increased above baseline levels in areas where there was intensive coal mining in Pennsylvania, peaking at 90,100 mg L⁻¹ and median values of 100 mg L⁻¹.

The DOC ranged from 1.78 to 10.8 mg C L⁻¹ (average = 4.41 mg C L⁻¹) for W4, BDL to 14.9 mg C L⁻¹ (average =−6.27 mg C L⁻¹) W3, 2.93 to 15.7 mg C L⁻¹ (average = 6.05 mg C L⁻¹) for W1 and 5.40 to13.4 mg C L⁻¹ (average = 6.23 mg C L⁻¹) for W2. DOC concentrations were similar to or slightly higher than what has been reported for Kentucky surface waters (Hoppe-Jones et al., 2010; Lei et al., 2020; Yurista et al., 2001). For all locations, the DOC was highest during summer (July; α<0.05) and lowest during spring (March and April) and fall (October). DOC concentrations in streams have been shown to vary by land cover, stream slope, temperature, and precipitation (Cool et al., 2014). The results indicated a negative correlation between DOC and the 5-day rainfall totals for W1 (−0.29) and W2 (−0.52) and a positive correlation for W3 (0.11) and W4 (0.59). While there is often a positive correlation between precipitation and DOC, it has also been shown that precipitation events can also have a dilution effect where the amount of precipitation added to the stream exceeds the increasing levels of DOC from sediment and tree litter during rainfall events (Liu and Wang, 2022; Ruckhaus et al., 2023). Temperatures positively impact the production of DOC since it is a microbial-meditated process where warmer temperatures increase the activity of organic matter degrading enzymes (Gillooly et al., 2001; Wallenstein and Weintraub, 2008). Deciduous forests are associated with lower DOC concentrations (Cool et al., 2014), and in our study, the lowest DOC concentrations were found in W4 (83.7 % deciduous forest cover). DOC plays an important role in freshwater ecosystems, regulates carbon and energy cycling, and can alter aquatic chemistry by contributing to acidification and the formation of complexes with trace elements (Toming et al., 2020).

Sixteen trace elements were measured in water grab samples (Table 4). Trace elements can result from soil parent material and geologic formations and anthropogenic activities. For W2, iron (Fe) concentrations exceeded the chronic criterion for aquatic organisms(1000μgL-1) during March, July, and August and aluminum (Al) concentrations exceeded their chronic criterion for aquatic organisms (350–1400μgL-1) in March, April, August, and September (EPA, 2023b). The chronic criterion for aquatic organisms for Al depends on pH, alkalinity, and DOC and was calculated each month using the EPA’s Aluminum Criteria Calculator (EPA, 2018). The discharge values at the time of sampling were 70.5 m³ s⁻¹, 44.9 m³ s⁻¹, 24.8 m³ s⁻¹, and 0.290 m³ s⁻¹ respectively, and the mean annual flow for the stream is 9.00 m³ s⁻¹ (Martin, 2002). During high streamflow events, sediments were likely washed into the stream due to erosion. Metals bind to sediments and suspend with the solids during the high flow and reducing conditions (Ekström et al., 2016). Iron and Al are naturally occurring throughout the region due to the rock and soil types present with mean ambient background soil concentrations of 11,000 mg Al kg⁻¹ and 22,500 mg Fe kg⁻¹ (EEC, 2004). Additionally, W2 is predominantly silty clay loams (68 % silt, 17 % clay, 3.6 % organic matter) where metals tend to bind with clay soils rich in organic matter. Elevated levels of trace elements beyond natural sources can also occur from industrial, agricultural (e.g., commercial fertilizers, liming materials, sewage sludges, irrigation, agrochemicals), and mining activities (He et al., 2005; Shan et al., 2019). W2 was located within the Western Kentucky Coal Fields which has been shown to contribute elevated levels of Fe, particularly in groundwater (Fisher and Davidson, 2007). The watershed also had the highest concentrations of manganese (Mn; α<0.05) peaking in July along with zinc (Zn) and lead (Pb) peaking in March.

W1 exceeded the chronic criteria for aquatic organisms for Al and Fe in July (932μgAlL-1;1530μgAlL-1) and September (582μgAlL-1;1040μgFeL-1). The elevated levels of Al and Fe in the watershed were likely due to residential runoff (e.g., roadways, vehicle traffic, roof runoff; Joshi and Balasubramanian, 2010; Reeves et al., 2018). W3 had the highest concentrations of strontium (Sr), a metal associated with mining activity, which was significantly (α<0.05) different from the concentrations detected in W1 and W2 (Clark et al., 2018; Clark et al., 2023). It was expected that W3 would also have elevated levels of Fe from the extensive mining activity in the watershed, however, the solubility and speciation are impacted by pH, which was between 7.3 and 8.2 resulting in lower concentrations of Fe in W3 (Tusher et al., 2015). Cadmium (Cd) was highest both in W3 and W4 and is often associated with mining and oil and gas activities (Clark et al., 2018; Johnston et al., 2019). Barium (Ba) was the highest in W4 and is used in drilling muds for oil and gas production (Ebrahimi and Vilcaez, 2018).

In total, 38 pesticides or pesticide by-products were analyzed, 35 were analyzed using the POCIS and 3 pesticides, glyphosate, AMPA, and glufosinate, were collected via grab sampling. Of the pesticides sampled, only seven (i.e., 6-chloronicotinic aldehyde, 6-chloro-N-methyl-nicotinamide, dimethoate, sulfoxaflor, thiacloprid, tefluthrin, trifloxystrobin), were not detected (Tables 5 and 6).

Watersheds W1 (89.9 % urban; 2.21 % agriculture) and W2 (3.97 % urban; 22.5 % agriculture) had the highest concentrations of imidacloprid and its by-products throughout the sampling period compared to the other sites (Table 5). All imidacloprid formations and by-products peaked in July (Fig. 5; α<0.05). Imidacloprid is a widespread neonicotinoid insecticide commonly found in both urban and agricultural watersheds (Satiroff et al., 2020). In urban settings, imidacloprid is used to manage pests in grassy areas such as parks, golf courses, and lawns (Mourikes et al., 2023) and is also used in flea and tick medicine where urban systems often have a larger density of domesticated household pets and dog parks. More recently, imidacloprid was used as a soil drench to treat emerald ash borer, a common infestation in trees in W1 (Larson and Crocker, 2022). Due to common detection in urban streams, imidacloprid is listed as an urban signature pesticide with increasing toxicity related to an increase in urbanization (Nowell et al., 2021).

The loading of imidacloprid in the environment has increased approximately 50 times in the past 20 years and has been linked to the killing of non-target insects like pollinators (Motaung, 2020). While the nicotinic agonist of the acetylcholine receptor (nAChR) for the target species is assumed to be specific to insects, some studies have shown toxicity implications for mammals (Kimura-Kuroda et al., 2012) including humans (Loser et al., 2021). An aquatic life benchmark for imidacloprid is set as the chronic criterion for freshwater invertebrates is 0.01μgL-1 (EPA, 2023a). W1 exceeded the criterion during May, June, and July, and W2 exceeded the criteria in June. Even more, imidacloprid metabolites (e.g., imidacloprid desnitro, imidacloprid urea, imidacloprid olefin) are often more toxic than the parent compound and can increase the potency in mammals compared to insects (Chao and Casida, 1997; Huang et al., 2021; Liu et al., 1993; Tomizawa and Casida, 2000). The main pathway for the degradation of imidacloprid into its by-products is through photodegradation with a half-life of 0.2 days in water with light exposure, but can also occur through hydrolysis, chlorination, and biological processes (e.g., microbial, fungal, plant; Loser et al., 2021).

The herbicides, alachlor and atrazine, were found in larger concentrations (2190 ng atrazine L⁻¹; 823 ng alachlor L⁻¹) in W2, peaking in June, compared to the other sampled sites. Atrazine is the most common pesticide found in agricultural surface waters due to its widespread use for controlling certain annual broadleaf and grass weeds and is primarily used for corn, but also applied to sorghum, sugarcane, and other crops. Alachlor has similar uses but is predominately used on corn and soybean fields, the most common crops grown in the watershed (USDA-NASS, 2023). Because of the herbicides’ water solubility and association with the pre-emergent application, the high concentrations are likely attributable to the spring flush (Battaglin et al., 2003; Hansen et al., 2019; Mottim et al., 2021; Rinsky et al., 2012; Thurman et al., 1991). For example, Battaglin et al. (2003) found concentrations of 4070 ng atrazine L⁻¹ and 50 ng alachlor L⁻¹ in agricultural streams after the first precipitation event following application where low-flow conditions produced 170 ng atrazine L⁻¹ and <50 ng alachlor L⁻¹. Additionally, Hansen et al. (2019) detected mean atrazine concentrations of 1.29μgL-1 with a maximum concentration of 175μgL-1 predominately in the early growing season in watersheds with a high amount of corn production and heavy precipitation in the United States. Both herbicides can be a threat to non-target aquatic species (Graymore et al., 2001; Hayes et al., 2010; Pereira et al., 2021). However, all concentrations detected throughout the sampling period were below the EPA’s chronic aquatic life benchmarks of 5μgL-1 and 187μgL-1 for fish and 60μgL-1 and 100μgL-1 for invertebrates for atrazine and alachlor, respectively (EPA, 2023a).

Glyphosate, used to manage unwanted weeds in agriculture but can also be used in pastureland, forestry, lawn care, gardens, and paved areas (Hébert et al., 2018), was also detected. It was hypothesized that all herbicides would peak in the early summer months following the first rainfall event post application and would be highest in W2, since it had the most agricultural land cover. However, the largest glyphosate concentrations were detected during August through September in W3, a watershed with little agricultural activity (0.9 % hay/pasture; Dewitz, 2021). As mentioned previously, a 1 in 1000-year rainfall event occurred over W3 from July 25th to 30th, corresponding with the peak glyphosate concentrations. A report written by the Ohio River Valley Institute and Appalachian Citizens’ Law Center (Dixon and Shelton, 2023), estimated that approximately 9000 homes were damaged by the flood. Entire homes and garages were swept away by flood waters, and it is likely that chemicals stored in those buildings were spilled and led to the high concentrations detected.

The last class of contaminants analyzed was PPCPs. In total 47 PPCPs were analyzed using the POCIS samples and 29 analytes were not detected or BDL. The largest detection of PPCPs was caffeine observed in March at 553 ng L⁻¹ in W4 (4.86 % urban) and the second highest (260 ng L⁻¹) in W2 (4.97 % urban; Table 7). The peak concentration in March is likely due to the cooler water temperatures throughout the month since caffeine degrades quicker in warmer environments resulting in lower concentrations of caffeine observed throughout the summer (Table 3; Hillebrand et al., 2012a). Hillebrand et al. (2012a) detected mean caffeine concentrations in freshwater of 10.3 ng L⁻¹ with higher concentrations occurring during the winter months and after a precipitation event. In general, caffeine is removed efficiently during wastewater treatment; however, at times the amount of caffeine that enters the water is greater than what can be degraded during treatment (Zhu et al., 2013). The half-life of caffeine is also shown to be stable in the aquatic environment with 100 to 240 d (Hillebrand et al., 2012b). Another study detected caffeine concentrations ranging from 118 to 735 ng L⁻¹ in August and September in the United Kingdom resulting from untreated or poorly treated wastewater (Nakada et al., 2017).

Acetaminophen (18.6 ng L⁻¹) and cotinine (29.1 ng L⁻¹; α<0.05) were also highest in W2 during March. Sulfamethoxazole was the only PPCP detected that was highest for W1 (89.9 % urban), peaking in July at 10.9 ng L−1 (α<0.05). Sulfamethoxazole is persistent in sediments, potentially limiting its detection in water bodies (Conkle et al., 2012). The two most predominant PPCPs found at elevated concentrations in the environment are caffeine and acetaminophen attributed to their high consumption (Osuoha et al., 2023).

All four of the PPCPs discussed are efficiently removed (>70 %) by WWTP (Buerge et al., 2008; Osuoha et al., 2023; Wu et al., 2023; Zhu et al., 2013). A high proportion of caffeine in the environment is evidence of untreated sewage (Daneshvar et al., 2012) and can be an indicator for the presence of other PPCPs (Hillebrand et al., 2012a). Katz et al. (2010) analyzed groundwater samples in a karst environment, similar to the geology of W1 and W2, and observed that septic systems resulted in the detection of caffeine, acetaminophen, and sulfamethoxazole. The study region has had a long history of wastewater problems and unpermitted sewage discharge, leading to environmental harm (EPA, 2002). An estimated 40 % of Kentucky households are not connected to a centralized sewer system with approximately 33 % of rivers and streams in Kentucky impaired with high levels of pathogens (EPA, 2002). For W4, approximately 11 % of the households in the sampled county are on a public sewer system compared to 97 % of households in W1 (KIA, 2015). Additionally, both W3 and W2 have 90 % and 95 % of the wastewater lines over 20 years old respectively (KIA, 2015). While PPCPs are efficiently removed during water treatment, the use of non-centralized sewer systems and aging infrastructure could be leading to the detection of PPCPs in these rural watersheds.

Furthermore, the only PPCP that occurred at just one of the sites was codeine, an opioid pain reliever, and was only detected in W4. Rodayan et al. (2016) detected codeine in surface (25–100 ng L⁻¹), wastewater effluent (350 ng L⁻¹), and drinking water samples (50 ng L⁻¹) linking drug abuse and the impact on downstream water sources. Among the counties in W4, there is an elevated rate of overdoses from drug abuse with an opioid overdose morality rate of 414.2 (deaths per 100 k population) compared to the national rate of 27.3 from 2018 to 2021 (NORC, 2024). Thus, not only can measuring PPCP in surface waters help to target areas for wastewater infrastructure improvements, but the presence of codeine in this rural Appalachian watershed highlights the potential use to target areas for public health and community resources related to drug abuse.

A PCA analysis was conducted to understand the relationship between land cover, water chemistry, and water quality. The PCA analysis was first conducted utilizing the entire dataset including land cover (i.e., percent urbanized area, percent agricultural area, percent mining area, density of oil and gas wells), physiochemical properties, and contaminant concentrations. Analysis was completed using the top two components which explained 65.1 % and 10.4 % of the variance respectively. The analysis showed that the top contributing variables were percent mining area, imidacloprid and its degradation byproducts (i.e., imidacloprid desnitro, imidacloprid urea), azoxystrobin, Sr, metalaxyl, TP, thiamethoxam, and dinotefuran (Table 6S). The top contributing variables were grouped together, showing that they are all positively correlated to each other, except for mining and Sr which are grouped together opposite of the other variables. Imidacloprid olefin showed a negative correlation with imidacloprid and its other by-products, likely due to the very low concentrations of imidacloprid olefin that were BDL for all site and events excluding W3 in September. The land cover seemed to have a larger influence than the water’s physiochemical properties where most of the fungicides and insecticides positively correlated with the percent agricultural area. The PPCPs also were more positively correlated with the percent urbanized area and density of oil and gas wells. Discharge was positively correlated with glyphosate and percent mining areas, which aligns with the high concentrations of glyphosate following the flooding event in W3.

The PCA analysis was also run with the top four contaminants that had the highest concentrations in each category (i.e., nutrients, trace elements, pesticides, PPCPs) to reduce the number of variables and gain a better understanding of the relationship between land cover and contaminant concentrations (Fig. 6). The results of the analysis identified DOC had a positive correlation with the nutrients, Al, Fe, Mn, imidacloprid, atrazine, alachlor, and cotinine, while temperature (T) had a positive correlation with Cl⁻, caffeine, and sulfamethoxazole. Both pH and SPC had a negative correlation with Al and Fe and a positive correlation with Sr, Cl⁻, and sulfamethoxazole. DO and discharge both had a positive correlation with SO42-, glyphosate, and acetaminophen.