Atrazine Ecological Monitoring Program: Two decades of generating daily or near‐daily monitoring data in highly vulnerable watersheds

Abstract

Surface water monitoring of pesticides ensures adherence to environmental and human health regulatory requirements. This study focuses on an unprecedented monitoring program spanning two decades with daily or near‐daily sampling across 13 states in the US Midwest and Southern United States, targeting watersheds in the upper 20th percentile of runoff vulnerability based on the United States Geological Survey watershed regressions for pesticides model. The Atrazine Ecological Monitoring Program (AEMP), mandated by the United States Environmental Protection Agency (USEPA), aims to collect extensive high‐frequency atrazine exposure data alongside key environmental parameters to better understand the dynamics of atrazine fate, transport, and concentrations in these watersheds. Note, the AEMP is also referred to by the USEPA as the Atrazine Ecological Exposure Monitoring Program, or AEEMP, though the former is more commonly cited. Analysis of the 322 site‐years of data revealed that sampling frequency is paramount in accurately estimating seasonal chemical concentrations and runoff loads in flowing waters. The AEMP has advanced with improved sampling techniques and a focus on increasingly vulnerable watersheds (i.e., currently representing 97th–99th centile runoff vulnerability), as evidenced by analysis of variance results indicating higher atrazine concentrations in later years, particularly post‐2012. Factors such as soil conditions and precipitation were significant in influencing atrazine levels in surface water. Regression analyses underscored the interaction between agricultural activity and weather patterns as predictors of atrazine concentrations. The AEMP's detailed dataset has notably contributed to environmental risk assessment and the refinement of regulatory models. This study highlights the value of high‐resolution data in vulnerable regions, emphasizing that high‐frequency monitoring and inclusion of detailed environmental data significantly enhance our understanding of pesticide fate and transport in surface waters and informs stewardship efforts.

Core Ideas

• The Atrazine Ecological Monitoring Program (AEMP) is a 17‐year study with daily or near‐daily sampling across 13 states.

• The AEMP monitors increasingly vulnerable watersheds, targeting watersheds representing the upper 20% of runoff vulnerability.

• High‐frequency monitoring and detailed environmental data are indispensable for understanding pesticide fate/transport and validating/assessing mechanistic models.

• Soil conditions and precipitation are significant variables influencing atrazine levels in surface water.

• Stewardship and mitigation efforts have led to decreasing concentrations at many of the most vulnerable watersheds.

Abbreviations

AEMD

Atrazine Ecological Monitoring Database

AEMP

Atrazine Ecological Monitoring Program

CE‐LOC

concentration equivalent level of concern

FIFRA

Federal Insecticide, Fungicide, and Rodenticide Act

LC‐MS/MS

Liquid chromatography tandem mass spectrometry

LOQ

limit of quantitation

NASS

National Agriculture Statistics Service

PRZM

pesticide root zone model

SAP

scientific advisory panel

SWAT

soil and water assessment tool

TSS

total suspended solids

USDA

United States Department of Agriculture

USEPA

United States Environmental Protection Agency

WARP

watershed regressions for pesticides model

1. INTRODUCTION

Environmental monitoring of pesticide residues in surface waters informs adherence to environmental and human health regulatory requirements and provides a benchmark to compare accuracy and relevance of model‐generated exposure predictions. Monitoring aids in understanding potential runoff from typical use of pesticides such as atrazine under varying environmental and climatic conditions, which can help identify factors influencing off‐target presence/movement. Atrazine, a chlorotriazine herbicide first registered in 1958 by the United States Department of Agriculture (USDA), has been extensively studied in US watersheds due to its widespread application on corn, sorghum, and sugarcane (Norman et al., 2020; Scribner et al., 2005; Solomon et al., 1996; United States Geological Survey [USGS], 1999). Though several programs have focused on monitoring atrazine to characterize potential runoff concentrations or loads into flowing waters, these efforts have often been limited in frequency/scope/logistics to accurately predict potential future occurrences and exposures (Chen, Mosquin et al., 2018; Setzler, 1980; United States Environmental Protection Agency [USEPA], 2001). Specifically, these studies have been constrained by geographical extent, environmental susceptibility, sample frequency, instrumentation availability, and access sites, which complicates characterization of chemical runoff and dissipation curves to assess potential environmental impacts at the local watershed scale (Thurman & Scribner, 2008; United States Environmental Protection Agency [USEPA], 2001).

As part of the reregistration review of atrazine in 2003, the United States Environmental Protection Agency (USEPA) required initiation of the Atrazine Ecological Monitoring Program (AEMP) to address potential ecological risk concerns for aquatic ecosystems with prolonged concentrations of atrazine at 10–20 µg/L, specifically in flowing waters of vulnerable watersheds. This was formalized in the Atrazine Interim Registration Eligibility Decision issued by the USEPA on January 31, 2003 (United States Environmental Protection Agency [USEPA], 2003a) and amended on October 31, 2003 (United States Environmental Protection Agency [USEPA], 2003b). The primary objective was to generate a comprehensive atrazine monitoring dataset by collecting multiple years of daily or near‐daily monitoring data, coupled with key environmental data. These data will inform future analyses of the mechanics and patterns of atrazine transport and concentrations in headwater streams of highly vulnerable watersheds. The AEMP was agreed upon by Syngenta Crop Protection, LLC (Syngenta) and five other registrants by the signing of the Atrazine Memorandum of Agreement (United States Environmental Protection Agency [USEPA], 2005). As part of the agreement, separate studies were implemented to address potential drinking water concerns for community water supplies and rural wells. Previous studies have shown that monitoring frequency significantly affects the ability to capture peak concentrations during runoff events and accurately predict atrazine levels. (Mosquin et al., 2017; Norman et al., 2020; Oelsner et al., 2017; Ryberg et al., 2020). Furthermore, covariates like stream flow and a detailed understanding of application timings and precipitation events are critical for improving predictions (Mosquin et al., 2017).

The AEMP has collected high‐resolution environmental data over a 17‐year period across 13 states throughout the US Midwest and Southern United States within headwater streams of watersheds classified within the upper 20th centile of vulnerability according to USGS watershed regressions for pesticides model (WARP). This retrospective analysis of the AEMP endeavored to address the following objectives: (1) Identify key site selection criteria characterizing watershed vulnerability; (2) assess the validity of the one‐in‐10‐year event phenomenon; (3) evaluate the critical role of monitoring frequency in accurately estimating seasonal chemical concentrations and loads in flowing waterbodies from runoff; and (4) illustrate the value of supporting spatial and environmental data for contextualizing atrazine concentrations at the local watershed level both seasonally and long term. While a comprehensive exploration of the mechanics and patterns of atrazine transport is beyond the scope of this investigation, the data generated by the monitoring program provide a solid foundation for future analyses of these dynamics.

2. MATERIALS AND METHODS

2.1. Watershed site selection and characterization

The AEMP was initiated in 2004 and operated in three distinct monitoring phases over a period of nearly two decades with four distinct watershed site selection efforts in 2004, 2005, 2007, and 2010. Watershed site selection (Table 1) was completed by the USEPA, Syngenta, and Waterborne Environmental.

TABLE 1.

Number of Atrazine Ecological Monitoring Program (AEMP) reporting sites monitored by year.

Sampling period	Number of reporting sites	Number of states represented
2004	20	6
2005	45	11
2006	31	10
2007	16	6
2008–2009	16	5
2010	39 a	10
2011	36 a	9
2012	31	9
2013	11	5
2014–2021 b	9	5

^a Count includes two nested reporting sites within the larger MO‐01 watershed. These two nested reporting sites were monitored from 2010 to 2011 as part of a stewardship research program.

^b The AEMP sampling activities were suspended in 2020 due to the pandemic and travel restrictions (USEPA, 2020a).

Phase 1 was conducted from 2004 through 2007 and consisted of two distinct watershed site selection processes. Vulnerable watersheds across major corn and sorghum states in the United States were identified using the USGS WARP, which uses factors such as cropped area use intensity in the watershed, precipitation intensity, and soil erodibility to predict concentration statistics in unmonitored streams, and the USEPA‐adopted Generalized Random Tessellation Stratified (GRTS) method to ensure spatially balanced selection of vulnerable sites (Larson et al., 2004; Williams, Harbourt, Matella, et al., 2004; Williams, Harbourt, Ball, et al., 2004). A filtering process was used to select hydrologic unit code 10‐scale watersheds with high atrazine use and runoff vulnerability across a 37‐state extent. Watersheds intersecting counties in the upper 45th percentile of atrazine corn‐sorghum cropped area use intensity (5860) became the baseline watershed dataset for site selection. Further refinements from ground‐truthing, drainage area, urban land use accumulation, crop accumulation, and selection of delineated watersheds that fell in the upper 20th percentile of runoff vulnerability for atrazine as predicted by the WARP model resulted in 1172 eligible watersheds for selection. Subsequently, a specific GRTS method selected by USEPA was applied using two stratum groups to balance a final selection of 40 watersheds that were both geographically dispersed and distributed across the range of WARP predicted values. Individual review of the potential watersheds within a geographical information system was then used to assess synthetic stream segments meeting specific criteria from USEPA and Syngenta, resulting in eligible headwater stream segments.

Core Ideas

• The Atrazine Ecological Monitoring Program (AEMP) is a 17‐year study with daily or near‐daily sampling across 13 states.

• The AEMP monitors increasingly vulnerable watersheds, targeting watersheds representing the upper 20% of runoff vulnerability.

• High‐frequency monitoring and detailed environmental data are indispensable for understanding pesticide fate/transport and validating/assessing mechanistic models.

• Soil conditions and precipitation are significant variables influencing atrazine levels in surface water.

• Stewardship and mitigation efforts have led to decreasing concentrations at many of the most vulnerable watersheds.

Additionally, consultation with landowners, county, and state entities alongside visual inspections of potential watersheds for safety and watershed access confirmed the subset of watersheds presented to USEPA for final approval to monitor. This resulted in 40 reporting sites on 1st–3rd order streams, representing watersheds in the upper 20th centile of runoff vulnerability for atrazine water concentrations as predicted by the WARP model (Supporting Information Section 1 and Figure S1). Sampling of watersheds in corn and sorghum areas was staggered across 2 years to ensure proper study setup, landowner, county, and/or state communications with locations across nine Midwestern states. In 2007, three additional reporting sites were added to provide greater context surrounding two of the original watersheds in Missouri to assist in watershed characterization.

Phase 1 also included site selection in sugarcane use areas, which was different than corn/sorghum because of different farming practices. Similar methods were used to determine the necessary WARP parameters, but the data to determine atrazine use intensity were specific sugarcane growing areas, which were concentrated in Texas, Louisiana, and Florida. Given the extensive sample programs in Florida and Louisiana, two reporting sites in Florida and three reporting sites in Louisiana were selected to begin sampling in 2005 (Williams, Harbourt, Ball, et al., 2004).

For each reporting site monitored in Phase 1, a minimum of two sampling years was completed prior to evaluating for continuation or dismissal by USEPA waiver from the AEMP. Reporting sites with relatively low exposure that met the criteria for dismissal outlined by the United States Environmental Protection Agency USEPA (2008) were waived (decommissioned) from the program. Following 2007, nine reporting sites were retained in the program: six of the original 40 corn/sorghum locations and all sugarcane reporting sites (United States Environmental Protection Agency [USEPA], 2008, 2010).

Phase 2 was conducted from 2008 to 2009. This phase introduced two additional reporting sites to the AEMP to enhance the understanding surrounding three high runoff potential watersheds monitored in Phase 1. Four reporting sites were waived from Phase 2 while seven corn/sorghum reporting sites continued (Table 1).

Phase 3, initiated in 2010, introduced 25 new reporting sites selected by USEPA to further refine watershed characteristics that contributed to high runoff vulnerability. Seven of the existing 11 reporting sites from Phase 2 continued plus two nested locations resulting in 34 reporting sites to monitor in the AEMP. The selection of new watersheds in 2010 followed recommendations published by the Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA) scientific advisory panels (SAPs) (SAP, 2007, 2009) reviewing the reregistration of atrazine. In addition to the criteria for nine to 40 square mile headwater watersheds, selection was primarily driven by (1) the occurrence of soils with shallow restrictive layers or (2) soils with shallow restrictive layers on slopes under treated crops (Miller et al., 2009b, 2009a; Prenger et al., 2009; Syngenta et al., 2010). These characteristics have been shown to potentially generate elevated levels of atrazine in runoff (Lerch & Blanchard, 2003). Other factors considered in the final selection of the 25 reporting sites included geographical distribution, potential vulnerability, and safe and effective water sampling capabilities.

Between 2010 and 2021, monitoring continued at six of the 25 new reporting sites and three of the original 40 corn/sorghum reporting sites from Phase 1 and 2. The AEMP watershed selection process for all phases is outlined in Figure S2. The AEMP was suspended in 2020 due to the COVID‐19 pandemic and travel restrictions (United States Environmental Protection Agency [USEPA], 2020a) and resumed in 2021. In total, 77 reporting sites across 13 states were monitored for a minimum of 2 years (Figure 1). See Supporting Information Section 1 and 2 for selection processes and key environmental metrics used in the decision‐making process.

FIGURE 1.

Locations and number of years monitored for all Atrazine Ecological Monitoring Program (AEMP) sites.

2.2. Targeted monitoring design

2.2.1. Sample collection schema

Three primary water sampling schemes were used over the course of the AEMP: (1) 2004–2006 comprised of 4‐day grab samples and/or event‐based composite autosamples, (2) 2007–2009 comprised of 4‐day grab samples and/or daily composite autosamples, and (3) 2010–2021 comprised of daily composite autosamples. In total, 52 reporting sites collected daily monitoring samples for at least 2 years. Daily composite samples were retrieved weekly from autosamplers. Samples collected during the study were analyzed for atrazine and total suspended solids (TSS). All sampling activities and analysis of samples followed the USEPA Good Laboratory Practices (United States Environmental Protection Agency USEPA, 1989).

The start date of each sampling period was based on historical annual and weekly corn and sorghum planting dates from USDA National Agriculture Statistics Service (NASS) reports to approximate timing of atrazine applications. Land use in each watershed was quantified from USDA‐NASS Cropland Data Layer data and provided a means to assess the spatial distribution for planting time variability and potential atrazine use (United States Department of Agriculture [USDA]–National Agriculture Statistical Service [NASS], 2004–2021; Supporting Information Section 1 and Figure S1). Sampling of locations began in advance of the anticipated first application of atrazine in the watershed and continued until at least July (Southwest states) or August (Midwest states) typically spanning 20–30 weeks each year. This period ensured seasonal applications of atrazine would be captured during monitoring.

For the 2004–2006 sampling period, at approximately 25% of the reporting sites, autosamplers were individually configured to initiate event‐based sampling in response to runoff events identified by a predetermined increase in stream stage. When a runoff event was detected in 2004, an initial aliquot (approximately 100 mL) was collected, followed by seven additional aliquots (one every 1.5 h), resulting in a 12‐h composite water sample. For runoff events in 2005 and 2006, the initial 100 mL aliquot was followed by seven additional aliquots (one every 45 min), resulting in an approximately 6‐h composite water sample.

In 2007, the collection of daily composite samples commenced, each containing eight approximately 50 mL aliquots (i.e., one aliquot every 3 h totaling about 400 mL per sample). From 2009 to 2021, water samples were preserved in the field with the antimicrobial agent sodium omadine at a target concentration of 64 mg/L (G. Smith et al., 2008). Adjustments and improvements to sampling methodology were made throughout the program to incorporate USEPA recommendations and outcomes from FIFRA SAPs to ensure accurate estimates of seasonal atrazine concentrations from runoff and assist in contextualizing the impact and magnitude of those concentrations at the local watershed scale (SAP, 2007, 2009, 2012). Additional details related to sampling are included in the Supporting Information Section 3 and Table S1).

2.2.2. Statistical analysis of targeted site selection design

By design, the AEMP intended to concentrate monitoring results toward watersheds recording higher atrazine concentrations. Therefore, annual 95th percentile concentrations were calculated from maximum daily values for each site in the AEMP (R Core Team, 2023) to determine the effectiveness of the sampling design. The maximum annual 95th percentile was then identified for each of the 77 reporting sites, including sugarcane. The corn and sorghum locations were then grouped by their sampling period, and concentrations were compared using an analysis of variance (ANOVA) test followed by a Tukey–Kramer post hoc test (de Mendiburu, 2023; R Core Team, 2023). The concentrations were natural‐log transformed prior to testing to achieve normality and homoscedasticity, confirmed with Shapiro–Wilk and Bartlett's tests, respectively (R Core Team, 2023).

2.3. Characterization of environmental variables

Water quality, meteorological, and streamflow data were collected from the onset of the AEMP. Specific water quality parameters included pH, specific conductivity, temperature, dissolved oxygen, and TSS. In Phase 1, portable hand‐held instruments were used during weekly site visits to collect water quality data. In 2008, newer models of handheld water quality sensor systems were used to measure parameters. Extended deployment water quality sensors were utilized at various locations beginning in 2010, which added chlorophyll‐a, blue‐green algae, and nitrate to water quality data. These data, collected over 10 years, supplemented the weekly water quality data collected during sample retrieval but are not discussed further here. Additionally, meteorological and soil moisture data were collected at each location, including accumulated rainfall, current and peak rainfall intensity, air temperature, relative humidity, wind speed, wind direction, barometric pressure, solar radiation, soil moisture, and soil temperature. Reporting sites were instrumented with various river stage and velocity measurement sensors, as determined by specific hydrologic conditions and evolving technology available through the program's history.

2.4. Analytical methodology

Various methods were used for atrazine analysis, including immunoassay, gas chromatography‐mass spectrometry, and liquid chromatography tandem mass spectrometry (LC‐MS/MS) prior to 2010 (Huang et al., 2003; United States Environmental Protection Agency [USEPA], 2015c, 2014). Methodologies used from 2004 to 2007 had a limit of quantification (LOQ) for atrazine of 0.1 µg/L. Starting in 2007, improved methodologies lowered the atrazine LOQ to 0.05 µg/L for the remainder of the program. Direct aqueous injection liquid chromatography electrospray ionization and mass spectrometry (LC‐ESI/MS/MS) Syngenta Method GRM014.02A (Huang et al., 2006) was used for atrazine residue analysis starting in 2010. This method used stable isotope analogues as internal standards (ISs) for residue quantification. Sample aliquots were either taken with an automated pipetting station or pipetted by hand. To each aliquot, 0.05 mL of IS was added for LC‐MS/MS analysis. The method LOQ was 0.05 µg/L for atrazine. Further analytical details are included in the Supporting Information Section 4 and Figure S3.

2.5. Sixty‐day moving averages

To assess chronic exposure profiles for sensitive aquatic species, 60‐day moving average concentrations were calculated for each of the 77 reporting sites. The moving average concentrations were calculated using a simple moving average formula to determine the average concentration for the previous 60 days, including the current date:

$$\mathrm{SM}{\mathrm{A}}_{60}=\frac{1}{60}{\sum }_{i=n-60+1}^n{c}_i$$ (1)

where n is the number of entries in a dataset and c is the daily forward stair step interpolation atrazine concentration. Atrazine concentrations below the LOQ constituted <10% of the samples analyzed and were replaced with one half of the LOQ, following recommendations issued by the USEPA (United States Environmental Protection Agency [USEPA], 2000). See Supporting Information Section 5 for interpolation details.

2.6. Sampling frequency analysis

Sample concentrations from all reporting sites were subsampled to simulate alternate sampling frequencies: every 4 days, weekly, biweekly, and monthly. Subsampling results in a corresponding number of unique permutations for each frequency (4, 7, 14, and 30). For example, while creating a dataset of weekly samples from the AEMP daily data, there exist seven variations of the dataset, based on the first sample date. Mosquin et al. (2016, 2017) demonstrated that kriging techniques and log‐linear interpolation, when compared to linear interpretation, improved accuracy in unsampled atrazine concentration prediction. Therefore, concentration data between subsampled dates were populated using log‐linear interpolation. Maximum daily values and 60‐day moving averages were calculated for each of the subsets following the formula in Section 2.5. Concordance correlation coefficients were calculated via the DescTools package (Signorell, 2024) to compare yearly maximum daily values and 60‐day moving average concentrations from the subsampled datasets to the original daily monitored concentration for each unique subsampling dataset per subsample frequency (i.e., 4, 7, 14, and 30). This was done per year at every site, and correlative strength was then compared across subsample frequencies.

2.7. Modeling sampling data trends

Initially, a non‐parametric Mann–Kendall test in R (McLeod, 2022) was used to develop a general understanding of directional trends in yearly concentrations. These results then scaffolded the development of parametric multivariate regression models capable of incorporating the environmental conditions not considered in non‐parametric tests. Least‐squares linear regressions were built by location to describe the varying monthly atrazine concentrations and test for potential trends over time. The nine AEMP reporting sites used for the analysis are those included in the extended monitoring years from 2013 onward, which continue to be monitored as of 2025. Each of these sites has been monitored at the daily or near‐daily scale for a minimum of 10 years, sufficiently long to capture a potential one‐in‐10‐year event. This type of event is representative of 90th centile predictions from regulatory models (Brain et al., 2015). Equation (2) describes the form of the regressions:

$$\ln \left(A\right)={\mathrm{\beta }}_0+{\mathrm{\beta }}_1\ln \left(C\right)+{\mathrm{\beta }}_{2}W+{\mathrm{\beta }}_{3}T$$ (2)

where A is mean monthly atrazine (µg/L); C is the combined variable of weighted crop progress (%) × precipitation sum (cm); W is the binary variable of maximum weekly weighted crop progress where one indicates weeks 0–4, T is time in years where 0 = 2004, and β values are statistically calculated coefficients. Further details, including additional considered variables, are included Supporting Information Section 6 and Table S2. Variables were logarithmically transformed as needed to correct for heteroscedasticity, first by adding half minimum non‐zero values per location to account for zeroes. Variables were checked for multicollinearity via variance inflation factor (Fox & Weisberg, 2019), regressions were checked for autocorrelation via Durbin–Watson test using p < 0.01 (Zeileis & Hothorn, 2001), and residuals were graphically checked for normality. Analyses were completed in R software, with graphs created with ggplot2 (R Core Team, 2021; Wickham, 2016).

3. RESULTS AND DISCUSSION

3.1. Sample results

3.1.1. Individual sample analysis

Over the course of the study (2004–2021), samples were collected across 77 distinct reporting sites, including two nested reporting sites, totaling 322 site‐years. During this time, 42,651 samples were collected, 29,534 of which were analyzed. Sample analysis was based on seasonal timing. During peak planting seasons, all daily samples were analyzed, while outside of the peak planting seasons, every 4th day was analyzed. Concentrations less than LOQ (2461) represented <10% of the samples analyzed and were represented using the EPA recommendation of half the LOQ (United States Environmental Protection Agency [USEPA], 2000). The median atrazine concentration across all analyzed samples was 0.73 µg/L, with an average concentration of 4.22 µg/L. The maximum single measured atrazine concentration was 344.26 µg/L, sampled from the IA‐05 location on May 17, 2014. Of all samples exceeding 100 µg/L during the course of study, 43% were recorded in 2014. These exceedingly high atrazine concentrations sampled during 2014 are representative of a one‐in‐10‐year event, affected by dry soil conditions combined with high‐intensity rainfall events across the Midwestern corn‐growing states. These events are further examined in Section 3.3. Figure 2 shows the 95th percentile distribution of maximum daily values across the reporting sites for their respective periods of record. See Tables S3 and S4 and Figure S4–S11 for location‐specific details and a summary of 95th percentile distributions by year across all reporting sites.

FIGURE 2.

Note 95th percentile distribution of maximum daily values for all Atrazine Ecological Monitoring Program (AEMP) monitoring sites.

3.1.2. Analysis of 60‐day moving average

The USEPA established an aquatic plant concentration equivalent level of concern (CE‐LOC) of 18 µg/L for 60‐day average concentrations during the study period of 2004–2011 and revised it to 10 µg/L for 2011–2019 (United States Environmental Protection Agency [USEPA], 2013). Any reporting sites with a maximum 60‐day moving average concentration below the CE‐LOC for two consecutive years were eligible to be waived from the monitoring program. Therefore, 60‐day moving average concentrations were calculated for all site‐years, as outlined in Section 2.5.

There were 66 site‐years with 60‐day average concentrations above the USEPA CE‐LOC. Of the 77 reporting sites, 75 had at least one occurrence of two consecutive years below 10 µg/L, which would qualify them to be waived from the monitoring program; however, some locations continued to be monitored on a voluntary basis. Of the nine locations with continued monitoring during 2013–2021, 8 had at least two consecutive years below 10 µg/L. Five reporting sites were below 10 µg/L for 2021, with the remaining reporting sites having a maximum 60‐day average of 10.74, 13.00, 18.37, and 18.70 µg/L (MO‐01, MO‐08, LA‐04, and MO‐07, respectively) in 2021.

Table 2 provides a summary of 60‐day average concentrations during the sampling period for each year. The overall average, minimum, and maximum of the total 60‐day moving average across all site years were 2.98, 0.03, and 90.83 µg/L, respectively.

TABLE 2.

Distribution of 60‐day moving averages from a daily basis for all 77 sites sampled.

Year	25th Percentile	50th Percentile	75th Percentile	90th Percentile	95th Percentile	Mean	Maximum
2004	0.22	0.92	2.37	6.06	8.12	2.08	16.76
2005	0.20	0.85	2.33	5.26	11.51	2.36	33.42
2006	0.12	0.60	1.82	3.61	7.18	2.05	57.51
2007	0.23	0.69	2.15	4.02	5.22	1.54	12.73
2008	0.29	0.95	2.70	6.18	7.22	2.17	20.94
2009	0.22	0.70	2.59	7.74	9.91	2.27	16.83
2010	0.36	2.07	5.03	8.69	10.48	3.47	34.48
2011	0.28	1.34	3.93	6.51	11.57	2.79	27.38
2012	0.42	1.31	3.34	6.42	9.27	2.39	15.57
2013	0.25	1.91	4.42	7.45	10.02	3.31	24.19
2014	0.91	3.02	9.64	40.96	53.43	11.65	90.83
2015	1.12	3.96	6.85	13.43	23.85	5.70	30.70
2016	0.54	2.73	5.75	10.78	16.84	4.43	22.31
2017	1.00	4.30	7.55	10.81	13.94	4.81	16.82
2018	0.16	2.08	6.50	10.31	15.05	4.13	21.52
2019	1.29	3.37	6.08	8.68	11.54	4.03	13.84
2020	–	–	–	–	–	–	–
2021	0.60	2.76	5.40	11.89	16.64	4.36	18.70
Overall summary *	0.27	1.17	3.69	7.49	11.12	2.98	90.83

^*Represents the summary statistics for daily 60‐Day moving averages of all sites‐years.

3.2. Monitoring frequency analysis

Collecting and analyzing daily samples during the growing season ensures that all peak concentrations are captured. For all AEMP sites, less frequent sampling and interpolation of daily values was simulated to understand the impact on maximum daily values and 60‐day moving averages. As the interval between samples increases, the concordance correlation coefficient (CCC) between the baseline set of AEMP daily values and the subsampled dataset decreases, denoting a larger variance in maximum concentration with less frequent sampling (Figure 3, panels A and B). Yearly maximum values from simulated weekly sampling are shown in panels C and D of Figure 3, highlighting that variance in maximum concentrations is greater in years with high atrazine concentrations. Sampling less frequently will always result in maximum measured concentration equal to or lower than those captured by more frequent sampling. Maximum 60‐day moving average values for the weekly subsampled sets varied as much as 11.8 µg/L higher and 21.8 µg/L lower than baseline data, a concerning level of inaccuracy when compared to the CE‐LOC of 10 µg/L.

FIGURE 3.

Concordance correlation coefficient for all Atrazine Ecological Monitoring Program (AEMP) sites, and maximum daily values for IA‐05 simulated weekly sampling. Concordance correlation coefficient (CCC): (A) value of one represents datasets with equal means and standard deviations, with 0 signifying no agreement between datasets. Red dots in panels (C) and (D) represent values representative of the baseline dataset.

3.3. Spatial and environmental data contextualization

This spatially extensive study reflects a wide range of meteorological, seasonal, hydrologic, soil, and land use conditions. These environmental parameters, collected locally at the watershed, provide valuable context for measured atrazine concentrations. The timing of corn or sorghum planting was observed to assess the potential effect on atrazine concentrations, with the highest atrazine concentrations in Midwestern locations occurring between April and June and Southern locations between March and May. See Supporting Information Section 8 for further details regarding planting timing and atrazine concentrations (Figure S12).

The connection between seasonal peaks in atrazine concentrations after rain events generating runoff, particularly when shortly following applications, is well documented (Scribner et al., 2005; Thurman et al., 1991; Wittmer et al., 2010). Measured rainfall and flow confirmed that precipitation events that led to an increase in magnitude of flow and elevated levels of atrazine in runoff occurred in all watersheds. For example, this relationship is observed for both small and large increases in flow at MO‐08 in 2017, with 3 spikes in atrazine levels during the year. The first two spikes in atrazine levels are associated with large flow rate increases from <0.5 m³/s to >75 m³/s. The third spike in atrazine concentration follows a rainfall event, which causes a smaller flow increase, from a base level of 0.14 m³/s to a peak of 1.4 m³/s (Figure 4). The third spike may be explained by relatively small increases in magnitude of flow often leading to longer residence times when compared to more significant increases in flow rate (Goodwin et al., 2017).

FIGURE 4.

Example of high‐intensity rainfall following a dry period: MO‐08 chemograph of 2017 season. An example of a period of dry days, followed by precipitation and subsequent higher atrazine concentrations. TSS, total suspended solids.

Soil condition is another contributing factor to atrazine levels in runoff. Studies have shown field management practices, particularly those that impact retention of soil moisture (e.g., no‐tillage vs. mulch tillage, particularly in areas with restrictive layers and within the Central Claypan Region of Missouri and Illinois), can have a significant effect on atrazine loss in surface runoff (Lerch & Blanchard, 2003; Ghidey et al., 2010). Prolonged periods of dry days lead to decreases in soil moisture content, which can lead to a soil crusting phenomenon in extreme cases. Precipitation onto dry or crusted soil may not be absorbed readily and can yield higher atrazine levels in runoff (Ma et al., 2004). Soil moisture and precipitation measured at AEMP sites aid in understanding this phenomenon. The three spikes in atrazine concentration at MO‐08 in 2017 (Figure 4) all occurred after rainfall events preceded by a period of at least 7 dry days. The largest of these spikes occurred after 17 dry days in a row, during which soil moisture content dropped 9% to its minimum value during the growing season. The chemograph shown in Figure 4 is representative of other sites with similar conditions: an extended period of dry days followed by a rainfall event.

This same soil crusting phenomenon, in conjunction with high‐intensity, localized rainfall, also led to the highest observed concentrations of the study in 2014. In 4 days of rainfall leading up to the maximum concentration of 344.26 µg/L at IA‐05, NEXRAD rainfall reported <1.25 cm (0.5 inches) of rainfall per day. However, highly localized rainfall events peaked at rates greater than 11.5 cm (4.5 inches) per hour, generating intense local runoff but not significantly increasing stream flow.

In contrast to rainfall after a lengthy period of dry days (Figure 4), rainfall events occurring when soil moisture is relatively elevated or even saturated can result in higher surface runoff, potentially yielding higher atrazine levels, for example, at LA‐04 in 2019. (Figure S13). Soil moisture remained relatively elevated, ranging from 37% to 60% volumetric water content between March and mid‐May, before declining to a baseline near 20% for the rest of the year. The major soil component in corn and sorghum growing areas of the LA‐04 watershed is a well‐drained fine‐silt with clay‐rich horizons. These properties can lead to higher concentrations in surface runoff when the soil becomes saturated and a heavy rainfall event occurs.

While variables such as hydrologic group, soil taxonomy order, yearly total precipitation, etc., can be used by generalized mixed models to predict pesticide concentrations, the local environmental parameters collected by AEMP provide additional context for causality of peaks in concentration. These additional parameters could also be integrated into future modeling efforts to refine predictions and enhance the understanding of atrazine transport.

3.4. Trends over time

3.4.1. Effects of targeted location selection

To assess the effective selection of more vulnerable sites with study progression, data were grouped into four categories by similar sampling years. While sugarcane sites had a different site selection process from corn and sorghum sites, they were included in this analysis. Groups consisted of locations sampled mainly between 2004 and 2006, 2007 and 2009, 2010 and 2012, and those continued in 2013–2021 (Table S5). The maximum annual 95th percentile atrazine concentration ranges for these four groups were 1.4–18.1, 1.6–62.01, 3.5–74.7, and 46.5–282.9 µg/L, respectively, with the latter group of sites 3–16 times greater than the maximum value from the initial group of sites in 2004–2006 (18.1 µg/L). The first three groups were representative of WARP vulnerability predictions in upper 87th, 91st, and 96th centiles, respectively (Table S5). Those sites included in the 2013–2021 group are in the upper 97th centile of WARP vulnerability, with LA‐04 being the only exception (91st centile). Moreover, sites reflected in the 2013–2021 cohort encompass the atypical one‐in‐10‐year event experienced in 2014. Sites monitored from 2013 onward had maximum annual 95th percentile concentrations in the 86th percentile or greater of all locations sampled within AEMP.

Results of an ANOVA test indicated maximum annual 95th percentile atrazine concentrations from the first group were significantly less than for the other three groups of locations (Figure S11). Locations sampled during 2007–2009 and 2010–2012 had statistically greater concentrations than the original locations (p < 1e‐3). The continued monitoring locations (2013–2021) deviated further from the original group (p < 1e‐6) to the extent of being significantly higher than the 2007–2009 and 2010–2012 groups as well (p < 1e‐5). The analysis confirmed that as locations were waived from the AEMP, the remaining monitored locations progressed toward more highly vulnerable locations to generate additional important data for assessment.

Results were compared to data available in the Atrazine Ecological Monitoring Database (AEMD), which includes data from over 100 surface water monitoring programs and is comprised of >243,000 samples (Perkins et al., 2021). The programs included in the AEMD varied in sampling frequency, from daily (targeted) sampling to very sparse sampling, sometimes as infrequent as once per year (ambient), with most characterized as ambient. When comparing maximum values per location, AEMP locations fell within the top 13% of all locations in AEMD. Additionally, 58 locations were in the top 5%, and 26 locations were in the top 1%, including the nine corn and sorghum AEMP locations with continued monitoring (2013–2021). Figure 5 shows the distribution of daily concentrations, as well as several moving averages using the maximum values per location per year. Moving average calculations for AEMD sites are detailed in Supporting Information Section 7.

FIGURE 5.

Distribution of daily and moving average concentrations: comparing phases of Atrazine Ecological Monitoring Program (AEMP) to Atrazine Ecological Monitoring Database (AEMD). 4‐, 21‐, and 60‐day moving averages (MA) based on maximum values per site per year. Boxes extend from the 25th to 75th percentile (interquartile range, IQR) with a line at the median. Whiskers extend up to 1.5x the length of the IQR, with points beyond. *AEMP 2004–2009 does not include MO‐01, MO‐02, or NE‐04. AEMP 2010–2021 includes all monitoring years for MO‐01, MO‐02, and NE‐04.

The distribution of site‐year maximum concentrations for AEMP is higher than those in AEMD across all time frames. As AEMP progressed, the distribution of concentrations increased as less vulnerable locations were waived from the program. This comparison against other monitoring programs further demonstrates the effectiveness of the vulnerability criteria used for AEMP site selection and shows the upper 20th percentile of concentrations were captured.

3.4.2. Insights from regression analysis

Additional statistical analyses were performed to examine potential effects of environmental variables on seasonal and long‐term variability at the most vulnerable sites. These sites include the nine watersheds included in the program since 2013, which were monitored over 11–17 years. Further basis for the inclusion of just these sites can be found in Supporting Information Section 6. Trends of decreasing atrazine concentrations over time were found at six of the nine AEMP locations monitored, based on the statistical significance of the time variable with negative beta coefficients, while the remaining three locations were insignificant at a 0.10 level of significance. Overall, regressions explained 47%–69% of variability in monthly mean atrazine concentrations in eight of the locations and 12% of variability in the TX‐01 location (adjusted R ²). Full details including regression coefficients for each location can be found in Tables S6 and S7. Residuals by location as well as comparisons across time are shown in Figure S14–S15.

Crop progress was explanatory in distinguishing between months within years and served to reduce noise in the regressions. Months with greater crop progress coinciding with greater precipitation (variable C) had significantly greater concentrations at eight of the locations, as shown in Figure 6. Planting week 4 and prior (variable W) had lower concentrations than later weeks on average (Figure 6). With these two variables serving as controls, year emerged as a significant negative correlation in six of the locations. On average, at these six locations, concentrations decreased by 6% (at MO‐02, [e − 0.066] − 1) to 12% (at MO‐07, [e − 0.126] − 1).

FIGURE 6.

All nine sites’ concentrations and regression variables are shown by year. Colors represent variable c, the weighted crop progress‐rain variable, demonstrating that greater values had greater concentrations on average. Circles, representing points that occurred ≤4 weeks from the beginning of planting (variable w), had lower concentrations on average.

The percentage of hectares (2.471 acres) planted with corn and sorghum was not included in the final regression structure due to its correlation to year (Figure S16), its relatively lower significance, and its inconsistent direction among locations. In contrast, year was more significant with p‐values < 0.10 for six of the nine sites and consistently negative for all sites, regardless of changes in corn and sorghum hectarage (Table S7). Of note, though the percentage of corn and sorghum hectares increased over time in MO‐01, MO‐02, MO‐07/N, and NE‐04 watersheds, concentrations still decreased over the years on average (Figure S15).

In contrast, the Texas location concentrations were not well explained by these variables having an adjusted r squared of only 12% (Table S7). This location is known to have backwater conditions and is near the Gulf of Mexico, creating a unique environment compared to the other locations. Additional characteristics and conditions would need to be measured to help explain the variability in atrazine concentrations at this location.

With year being an explanatory variable, ecological conditions are not solely responsible for the decrease in concentrations over time. Over the course of the monitoring study, the labeled maximum single application and yearly total application rates have remained unchanged at 2.24 kg/ha (2 lbs/acre) for corn and sorghum, and 4.48 kg/ha (4 lbs/acre) on sugarcane. Estimated total atrazine usage has fluctuated over the course of the study, with an average application rate under 1.12 kg/ha (1 lb/acre) for corn and sorghum uses and 2.86 kg/ha (2.55 lb/acre) for sugarcane between 2013 and 2017 (USEPA, 2020b). During the study period, some use restrictions, such as a reduced rate on highly erodible lands, have been added to the atrazine label. Active stewardship and mitigation programs in each of the nine watersheds were likely significant contributors to the reduction of concentrations.

3.5. Use of AEMP monitoring data for ecological risk assessment

This comprehensive, targeted water monitoring program was developed under a specific mandate from the USEPA, coinciding with multiple SAPs over the last two decades (SAP, 2007, 2009, 2012). Over the course of multiple evaluations, including an examination of AEMP monitoring data, the CE‐LOC was reduced from 18 µg/L to 10 µg/L in 2011. In October 2019, the EPA revised the CE‐LOC to 15 µg/L but has more recently proposed a CE‐LOC of 9.7 µg/L in July 2024 (USEPA, 2024). Due to the mode of action of atrazine, aquatic plants are most vulnerable to effects of exposure, so the CE‐LOC is also considered protective of fish, invertebrates, and amphibians (P. Smith et al., 2021). In comparison, the CE‐LOC is higher than the maximum contaminant level for atrazine in drinking water, 3 µg/L.

This data‐rich program provides the USEPA with the requisite information to further refine and characterize its ecological risk assessments (SAP, 2012). which generally involve estimating exposure of aquatic organisms to pesticides via computer simulation models. As one of the few monitoring programs with daily sampling or over 100 samples per year, the AEMP data are critical for model calibration to improve environmental fate and transport predictions. AEMP data have been utilized as calibration data for regulatory water models such as the pesticide root zone model (PRZM) (Carsel et al., 2003; Ghebremichael et al., 2020, November 15−19, 2022; Young & Fry, 2014) and soil and water assessment tool (SWAT) (Texas A&M University [TAMU], 2017; Winchell et al., 2018).

The AEMP data were also used by the USEPA in preliminary evaluations for the 2015 SAP on the Spatial Aquatic Model (Thurman et al., 2014, 2016; United States Environmental Protection Agency [USEPA], 2015a, 2015b) and in the SAP for “Approaches for quantitative use of surface water monitoring data in pesticide drinking water assessments” in developing the SEAWAVE‐QEX (trend analysis method with seasonal wave, streamflow adjustment, and extended capability) (United States Environmental Protection Agency [USEPA], 2019; Vecchia, 2018). In addition to its use in the regulatory environment, the program and associated data have been presented in numerous public and industry meetings, such as Environmental Modeling Public Meetings (EMPM) held by the USEPA (Boharty et al., 2018; Hetrick et al., 2014; Chen et al., 2014; Chen, Mosquin, et al., 2018; Chen, Perkins, et al., 2018, 2019, March 27) and American Chemical Society meetings (Harbourt et al., 2009; Miller et al., 2009b, 2009a, 2009c), which are further detailed in Supporting Information Section 9.

4. CONCLUSION

This spatially extensive study reflects a wide range of meteorological, seasonal, hydrologic, soil, and land use conditions, which has already generated 322 site years of highly time‐resolved data from 2 to 17 years of monitoring at 77 reporting sites. The AEMP has evolved over time, utilizing increasingly sophisticated sampling and analytical approaches as well as focusing on increasingly more vulnerable watersheds. The daily or near‐daily sampling regime of the AEMP ensured that peak concentrations were captured, establishing accurate 60‐day moving average calculations for regulatory requirements. High‐resolution environmental parameters, such as the local precipitation, soil moisture, and flow collected during AEMP, provide context for observed concentrations and will be valuable in future detailed analysis of atrazine fate and transport. High concentrations in individual samples often coincided with extremes in soil moisture, such as extended dry periods or saturated soil, followed by a rainfall event. As reporting sites were waived from the AEMP, the remaining locations advanced toward higher vulnerability, confirmed by ANOVA analysis on max annual 95^th percentile concentrations between 2004 and 2006, 2007 and 2009, 2010 and 2012, and 2013 and 2021 reporting site groups. This is not indicative of increasing atrazine concentrations or use rates, but rather that successive iterations of site selection were successful in identifying more vulnerable watersheds. Regression analysis showed that the weighted crop progress x precipitation sum, maximum weekly weighted crop progress (binary), and year are important in modeling seasonal and long‐term trends. Because of stewardship and mitigation efforts, atrazine concentrations in many monitored watersheds have decreased over time, as shown in Figure 6. The unprecedented scope, longevity, and sampling frequency of the AEMP data establish it as a premier dataset and invaluable resource to inform regulatory models. However, caution should be exercised when contextualizing the data given the degree of vulnerability of the watersheds included. The site selection process and sampling regime of the AEMP serve as a model for future studies that they may provide high‐quality data for further refinement of risk assessment.

AUTHOR CONTRIBUTIONS

Zechariah Stone: Data curation; visualization; writing—original draft; writing—review and editing. Sunmao Chen: Conceptualization; project administration. Jennifer Trask: Writing—original draft; writing—review and editing. Sarah Terrell: Formal analysis; writing—review and editing. Megan Cox: Writing—original draft. Nicholas Guth: Visualization. Richard Brain: Project administration; writing—review and editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

The supplemental material includes in‐depth discussion of land use, site selection, sample collection schema, analytical methodology and sample data trends. It contains additional summary tables and further data analysis that was not included in the main document.

Stone, Z. , Chen, S. , Trask, J. , Terrell, S. , Cox, M. , Guth, N. , & Brain, R. (2025). Atrazine Ecological Monitoring Program: Two decades of generating daily or near‐daily monitoring data in highly vulnerable watersheds. Journal of Environmental Quality, 54, 1060–1076. 10.1002/jeq2.70014