Towards quantifying atmospheric dispersion of pesticide spray drift in Yuma County Arizona

Abstract

While pesticide vapor and particles from agricultural spray drift have been reported to pose a risk to public health, limited baseline ambient measurements exist to warrant an accurate assessment of their impacts at community-to-county-wide scale. Here, we present an initial modeling investigation of the transport and deposition of applied pesticides in an agricultural county in Arizona (Yuma County), to provide initial estimates on the corresponding enhancements in ambient levels of these spray drifts downwind of application sites. With a 50 × 50 km domain, we use the dispersion model CALPUFF with meteorology from the Weather Research and Forecasting (WRF) to investigate the spatiotemporal distribution of pesticide abundance due to spray drift from a representative sample of nine application sites. Data records for nine application days in September and October 2011, which are the peak months of pesticide application, were retroactively simulated for 48-h for all nine application sites using an active ingredient lambda-cyhalothrin, which is a commonly-used pesticide in the county. Twenty-one WRF/CALPUFF simulations were conducted with varying emissions, chemical lifetime, deposition rate, application height, and meteorology inputs, allowing for an ensemble-based analysis on the possible ranges in modeled abundance. Our results show that dispersion of vapors released at time of application heavily depends on prevailing meteorology, particularly wind speed and direction. Dispersion is limited to thin plumes that are easily transported out of the domain. The ensemble-mean vapor concentrations of the 48-h average (> 90 percentile domain-wide) range from 0.2 nanograms (ng)/m³ to 200 ng/m³, and the peak can be as high as 1000 ng/m³ near the application sites. Pesticide particles are mainly deposited within 1–2 km from the application sites at an average rate of 10⁶ ng/km²/h but vary with particle mean diameter and standard deviation. While these findings are generally consistent with reported ambient levels in the literature, the associated ensemble-spread on these estimates are in the same order of magnitude as their ensemble-mean. At the two nearby communities downwind of these sites, we find that peak vapor concentrations are less than 50 ng/m³ with exposure times of less than an hour, as approximately 99.4% of the vapors are advected out and 99.5% of the particles deposit within the domain. Results of this study indicate pesticide spray drift from a sample of application sites and representative days in Fall may have a limited impact on neighboring communities. However, we strongly suggest that field measurements should be collected for model validation and more rigorous investigation of the actual scale of these impacts when the bulk of pesticide applications across the county, variation in active pesticide ingredients, and potential resuspension of deposited particles are considered.

1. Introduction

Pesticides are applied in numerous agricultural activities worldwide and can come to exist in residential ambient air by evaporation from crops and residential surface soil. In addition, pesticides can be transported from agricultural fields by wind with some fumigants released into the air in vapor and particle forms (e.g., Degrendele et al., 2016; Figueiredo et al., 2021; Li et al., 2011; Li and Jennings, 2017; Nascimento et al., 2018). Exposure to pesticides poses a potential danger to public health, especially when dispersed from areas having intensive agricultural activities to population centers. The World Health Organization (WHO) estimates that there are 3 million cases of pesticide poisoning each year and up to 220,000 deaths, primarily in developing countries (Thundiyil et al., 2008). Studies have also shown that pesticide exposure can cause a range of neurological health effects such as memory loss, loss of coordination, reduced speed of response to stimuli, reduced visual ability, altered or uncontrollable mood and general behavior, and reduced motor skills (Shelton et al., 2014; Kori et al., 2018; Furlong et al., 2020a,b; Lucero and Muñoz-Quezada, 2021).

Agriculture in Yuma County, located in southwestern Arizona, plays an important role in Yuma’s economy. There are 237,742 acres of farmland with 224,160 harvested (Tickes & Zerkoune et al., 2002). The Colorado River is the source of irrigation water for the Yuma Mesa and surrounding valleys. Yuma County is responsible for 90% of all leafy vegetables in the United States, from November to March (www.yumachamber.org/agriculture). Data reported by Sugeng et al. (2013) showed that an annual average of 5.7 million pounds of active ingredients of pesticides was applied in Yuma County from January 2006 to June 2011 (Sugeng et al., 2013). The pesticide applied per area of cropland per year in Yuma County is much higher than average value in the United States reported by FAO (2021). This may be because warmer temperature enables farmers in Yuma County to harvest multiple times in one year. There have been harvests even in December (Ottman et al., 1996). These applications have direct implications to public health in the county. O’Rourke et al. (2000), for example, showed that one or more urinary metabolites to organophosphate pesticides were observed in 33% of children younger than 6 years old in Yuma County. It is therefore critical that investigations on potential exposure from pesticide applications be considered, especially that associated ambient levels of pesticide spray drifts (vapor and particles) are poorly known and have yet to be quantified.

Few worldwide jurisdictions have regulated pesticide air quality standards compared to regulations on pesticide residuals in pesticide-contaminated food and water. This means that people around the world are probably not protected by some form of pesticide air quality monitoring and regulations, especially for farmers and workers who frequently work in the agricultural fields (Li and Jennings, 2017). Studies have shown that pesticides may volatilize and drift in the atmosphere during applications. The fraction of the non-target area exposure ranges from a few percent to 50 percent (Gil and Sinfort, 2005; van der Werf, 1996; van den Berg et al., 1999). Chemical techniques and tracer monitoring are typically used to help measure pesticide emissions into the air (Nascimento et al., 2018). But with active samplers, liquid and gaseous phases are difficult to separate, and the sampling method has to be adapted to the type of polluting agent under consideration. It requires meticulous and expensive laboratory analysis (Gil and Sinfort, 2005), which makes it more difficult to deploy regular monitors in affected areas.

Spray models and numerical simulations, which are easier and cheaper to operate, now have been widely used in studying the atmospheric dispersion of pesticide particles. Spray emission modeling, in combination with field tests in particular conditions, could be a suitable solution to understanding pesticide dispersion and conducting impact assessments (Gil and Sinfort, 2005). For example, Woods et al. (2001) used AgDRIFT to simulate the spray drift of pesticides arising from aerial application in cotton, while Teske et al. (2009) introduced the development of an analytical ground boom sprayer model, based on a Lagrangian approach to aerial spraying, in the computer model AGDISP (AGricultural DISPersal model). AGDISP is designed to be simple and practical for its use on initial assessments of potential impacts during the design of new pesticide applications and spray drift near the fields (Nsibande et al., 2015). AGDISP may be limited when used in monitoring pesticide spray drifts since its model design is simplified. It assumes for example a flat terrain and near-neutral atmospheric stability (Bird et al., 2002), which can be a problem when used to simulate emissions over a desert climate and more complex terrain like that in Yuma County compared to the Great Plains. On the other hand, more sophisticated modeling studies have been conducted in recent years. For example, Ellis et al. (2010) revised the ADMS (Atmospheric Dispersion Modelling System) to simulate the volatilization of pesticides after applications. In this study, meteorological data from a single location was used. AERMOD (American Meteorological Society/Environmental Protection Agency Regulatory Model) is another commonly used Gaussian model for atmospheric dispersions of pollutants. It is a plume model assuming uniform meteorological conditions for data files. Some studies coupled GIS (Geographic Information System) and AERMOD for pollutants emitted by complex sources (Teggi et al., 2018).

Another option for dispersion modeling that can represent both complex terrains and varying meteorological backgrounds is California Puff (CALPUFF). The CALPUFF modeling package is a prominent air-pollution model for pollutant dispersion (US EPA, 2012). It is a Gaussian puff dispersion model that simulates the effects of time- and space-varying meteorological conditions on pollution transport, transformation, and removal (Cui et al., 2011; Dresser and Huizer, 2011; Guo et al., 2020; MacIntosh et al., 2010; Wang et al., 2006; Zivan et al., 2016). It can simulate local-scale dispersion, especially when coupled with high-resolution gridded meteorological data to improve representation of mesoscale-to-microscale meteorology. Dresser and Huizer (2011) used both CALPUFF and AERMOD to simulate the dispersion of sulfur dioxide (SO₂) near Martins Creek, PA, and reported that the performance of CALPUFF was superior to that of AERMOD. CALPUFF was also employed in recent studies on pesticide spray drift. Pivato et al. (2015) used CALPUFF to assess the airborne concentrations of pesticide drift from vineyards in Italy. Zivan et al. (2016) simulated airborne organophosphate pesticide drifts in the Mediterranean climate using CALPUFF and evaluated the importance of secondary drift.

This study builds upon the recent use of CALPUFF on pesticide spray drifts (e.g., Exponent, 2019). Our goal is to explore whether and how much the spray drift of pesticides disperses to nearby communities through use of CALPUFF and high-resolution gridded meteorology data. We used CALPUFF in an ensemble approach to simulate 9 pesticide applications in Yuma County over a two-month period in 2011. To our knowledge, this is the first study using an ensemble approach to simulate and evaluate the potential risks of pesticide spray drift in Yuma County. Unlike previous research, instead of focusing on several applications, an ensemble matrix for several model parameters (e.g., emission rate, application height, species lifetime, among others) is developed and incorporated within CALPUFF. We then evaluate the ensemble model outputs of pesticide abundance in the region to assess the potential mass loading of pesticide particles and/or concentrations of pesticide vapors from spray drift at nearby residential communities.

2. Method and materials

2.1. Study site

We show in Fig. 1a the locations of application sites in this study. Yuma County is located in southwestern Arizona with an elevation of about 43 m. Most agricultural fields in this study are located east of downtown Yuma, Arizona. The center of our study site is at 114.2869 °W; 32.6939 °N to cover most fields which are in the valley between the Gila and Muggins Mountains. This study domain covers two residential communities: Fortuna Foothills (Fig. 1b) and Wellton (Fig. 1c). There are more than 20,000 residents in Fortuna Foothills and about 3200 residents in Wellton. Several application sites are located both east and west of Wellton while three sites are located northwest of Fortuna Foothills. We design our domain (50 km × 50 km) to be larger than the spatial extent in Pivato et al. (2015) and Zivan et al. (2016) of about 1 km × 1 km since we are assessing the micro-to-mesoscale (1–50 km) plume features of pesticide spray drift from Yuma agricultural fields to nearby communities surrounding these fields.

Fig. 1.

Google images of the study area: a) model spatial domain showing 9 application sites (red markers) and community locations, b) Wellton community and c) Fortuna Foothills community.

2.2. Information on pesticide applications

The pesticide applications and characteristics considered for this study are listed in Table 1. These are taken from the Arizona Department of Agriculture (AZDA, agriculture.az.gov). In Arizona, many types of agricultural pesticide applications are reported to the state as required by State law (Farrar et al., 2016). Based on AZDA’s 25-year record, the amount of pesticide applications has increased from an annual average of 1340 lbs in 1992 to 4800 lbs in 2016 and have peaked at 5362 lbs in 2015. The amount of pesticides applied in 2011 was 2600 lbs (annual average), which is also the average amount across the 25-year record. Applications vary with season, with highest amounts during Fall (September: 5450 lbs and October: 7660 lbs) and lowest amounts during summer (June: 385 lbs and July: 550 lbs). Most applications are conducted between 18:00 and 22:00 with a duration longer than 1 h. Pesticides are typically applied via small airplanes and sprayed at approximately 3 m (m) (or 10 feet) high. Common pesticide carriers include water, petroleum distillate and once-refined vegetable oil concentrate. From AZDA records, water was used as the pesticide carrier for the 9 applications and the active pesticide ingredient was lambda-cyhalothrin (NCBI, 2022). These 9 sites represent the locations of agricultural fields having relatively large amounts and more frequent applications. Pyrethroids like zeta-cypermethrin, pyrethrins, lambda-cyhalothrin, bifenthrin, carbamates like methomyl, and organophosphates like chlorpyrifos, bensulide, and dimethoate comprise the most applied active ingredient in Yuma, with zeta-cypermethrin having the highest number of applications (104,000), lambda-cyhalothrin (61,341) and bifenthrin (26,650). We chose lambda-cyhalothrin as it represents the average number of applications among the top ten ingredients applied in Yuma. This pyrethroid insecticide has a broad insecticidal spectrum and rapid efficacy (Wang et al., 2021). Lambda-cyhalothrin may cause irritation to the skin, throat, nose, and other body parts if exposed. Skin tingling, burning, and prickling feelings, particularly around the face, are unique temporary symptoms of exposure. Other symptoms may include dizziness, headache, nausea, lack of appetite, and fatigue. In severe poisoning, seizures and coma may occur (He et al., 1989). It is important to note that lambda-cyhalothrin has a relatively longer lifetime in soils (30 days) and in air (0.5–7 days) given its weak volatilization effects and solubility (Guedegba et al., 2019; National Center for Biotechnology Information, 2022). While this implies a larger spatiotemporal extent of its presence in the atmosphere, the bulk concentration is expected to be localized given the relatively large size of the spray droplets during application. Here, the median diameter and standard deviation for the pesticide droplets is assumed to be 195 μm and 88 μm (Table 1), respectively. This is within the recommended size range of 106–340 μm (fine: 106 to 235 μm, medium: 236–340 μm) by the American Society of Agricultural and Biological Engineers (ASABE) standard (ASABE S572.1). To facilitate comparison with other modeling studies, we report the median emission rate per area per simulation period per hour (EMAH) of 8.6 × 10⁻¹ kg/km²/h. Simulation time for each ~85-min application is set to 48 h for this study.

Table 1.

Information on pesticide application and pesticide characteristics.

Application start time	18:00–22:00
Median duration time	85 min
Median area of application	92,673 m2
Median amount of pesticides applied	3.5 kg
Median amount of active ingredient applied	0.1 kg
Median application height	3 m
Active ingredient	Lambda(λ)-Cyhalothrin
Median diameter of particle	195 μm
Standard Deviation of particle	88 μm
Half-life in soils	30 days
Half-life (via OH and O3 reactions)	12 h to 7 days
Median Emission rate per simulation hour (EMAH)	8.6× 10−1 kg/km2/hr
Diffusivity (λ-Cyhalothrin)	0.0569 cm2/s
Solubility (λ-Cyhalothrin)	0.005 g/m3
Vapor pressure (λ-Cyhalothrin)	0.2 μPa
Henry’s law constant (λ-Cyhalothrin)	0.000014

2.3. Potential routes during and after the pesticide applications

Following Pivato et al. (2015), we also consider potential routes during and after pesticide applications. As summarized in Fig. 2, the bulk of pesticides is in the form of droplets intended to fall on the application site designated as Route 3. A small fraction of the pesticides is introduced in the atmospheric boundary layer via Route 1, 2, and 4. Route 1 (vapor) represents the pesticide that volatilizes immediately during application. The fraction of pesticides in Route 1 can range from a few percent to 50 percent (FOCUS, 2008; Gil and Sinfort, 2005; Holland et al., 1997; Werf, 1996; Van Dijk and Guicherit, 1999). Zivan et al. (2016) assumed 10% of the chlorpyrifos would evaporate during the application, while Pivato et al. (2015) assumed 30% of the pesticides with an active ingredient having small vapor pressures (like metiram) would evaporate. It should be noted that many factors influence the fraction of the pesticide that volatilizes during the application. Aside from spray operator issues, Hofman and Solseng (2001) identified the following three categories for this wide range in Route 1: (a) volatility and viscosity of the pesticide formulation, (b) equipment and application techniques, and (c) weather conditions. Although the semi-arid conditions in Arizona may enhance evaporation, lambda-cyhalothrin do not quickly evaporate. Considering that Zivan et al. (2016) conducted a more in-depth investigation of secondary drift than Pivato et al. (2015), we assume a conservative 10% of the total mass will evaporate during the applications similar to Zivan et al. (2016).

Fig. 2.

Potential routes for pesticide spray. Numbers associated with each route represent percentages in total mass of pesticides used in this study. Spray drift calculations include Route 1, 2, and 4.

Some small droplets during spray can be also treated as particles (or aerosols; Route 2). A portion of spray droplets are fine enough for them to stay in the atmosphere for a longer time, with stronger wind enabling transport of these small droplets farther downwind of application sites. Although many previous studies neglected Route 2 (e.g., Zivan et al., 2016), we propose to include this route since this fraction to the total size distribution can be a significant contributor to potential dispersion. As illustrated on Fig. 2, we assume 0.136% of the total mass of pesticide applied becomes small particles (i.e., aerosols) via Route 2. This is based on the size distribution of droplets from a separate AGDISP experiments that we conducted. We note that although some literature claimed that these small particles may also volatilize shortly after the application, we still take them into consideration for baseline calculations. Since we are focused on studying the spray drift of these pesticides, the percentage of mass depositing on the target area (Route 3) is not included. For evaporation of pesticides in the target area (Route 4), we estimate a volatilization rate for lambda-cyhalothrin of 0.0033% per hour and we simulate emissions of Route 4 from 8am to 11am following the application from the prior evening. This estimate is based on two species (copper oxychloride and mancozeb) with similar vapor pressures as those reported in Pivato et al. (2015). Volatilization from the target area of these two species contribute about 0.001% of the average concentrations. In addition to the physical properties, the volatilization effect is estimated to be very low because there is no sunlight in the evening. During these hours, about 70% of the pesticide is expected to be absorbed (Bouldin et al., 2006). For species with stronger volatilization ability (e.g., chlorpyrifos), the evaporation from the soil can be greater (Leistra et al., 2006).

2.4. Meteorological setting

Since Yuma County has an insufficient number of surface meteorological stations for our study period and domain, we use gridded meteorological data from a regional atmospheric model (Weather Research and Forecasting or WRF) as input to CALPUFF. In particular, we use a more recent version of the dynamically downscaled WRF reanalysis originally based on studies by Carrillo et al. (2017) and Luong et al. (2017). The WRF model is a meso-scale numerical weather prediction (NWP) system designed for both atmospheric research and operational forecasting needs (Cipagauta et al., 2014). A detailed description of physics, governing equations, and dynamics incorporated into WRF model are detailed in Skamarock et al. (2008).

The average temperatures in September and October in Yuma are about 304 K (31 °C) and 297 K (24 °C), respectively. Arizona is influenced by the North American Monsoon (NAM) in early September (Adams and Comrie, 1997; Rowson and Colucci, 1992). During the monsoon season, the prevailing winds in Arizona shift to a southerly or southeasterly direction. Modeled wind directions for September 2011 and October 2011 are shown in Fig. 3a and b. We can see the wind direction is southerly or southeasterly in some areas in Arizona and a thermal low is located northwest of Yuma County. Since the monsoon usually ends in mid-September, the southeasterly or southerly wind direction is not so apparent. In addition to the high heat capacity of the desert, the higher temperature in southern Arizona in October compared with other places at the same latitude is associated with the Foehn effect induced by the Colorado Plateau located northeast of Yuma, cutting through central Arizona from northwest to southeast. When the air coming from the north climbs up the Colorado Plateau, the water vapor will condense. The air becomes warmer and dryer when it descends from the Colorado Plateau to the south. In particular, there is relatively low specific humidity in Yuma (September: 10.3 g H₂O/kg air; October: 6.5 g H₂O/kg air). According to Zivan et al. (2016), such relatively warm and dry weather conditions are reported to enhance spray drift.

Fig. 3.

Monthly-averaged 2-m temperature and 10-m wind velocities from WRF reanalysis. Spatial maps a) and b) show the distribution of temperature (color contours) and wind (arrows) for September 2011 and October 2011, respectively, with the CALPUFF domain shown as a blue box. Time series (c) and (d) show the mean diurnal cycle of temperature (blue) and wind speed (orange) for September 2011 and October 2011, respectively.

Modeled wind speeds at 10 m above ground surface usually ranges from 3 to 5 m per second (m/s) throughout the day in both September and October. These values are consistent with ERA5 reanalysis but biased high (by about 1.5 m/s) compared to surface sites in Yuma from AZMET (Arizona Meteorological Network, https://ag.arizona.edu/azmet). The average diurnal variation however is consistent with ERA5 and AZMET. In Fig. 3c and d, we find that the peak of the wind speed across the diurnal cycle occurs around 17:00 in both September and October, almost the same time as the peak in temperature. This is because, with a warmer surface, there will be more vertical mixing due to turbulence. The average wind speed is around 4 m/s from 18:00 to 22:00 during both months. United States Environmental Protection Agency (EPA) requires a wind speed of less than 4.44 m/s (10 mph) during the pesticide application (US EPA, 2017). Given our domain of 50 km × 50 km, it takes about 2 h for most vapors emitted at the center of the domain to be transported 25 km and out of the domain. In fact, most vapors are estimated to be transported out of the domain 3 or 4 h after the pesticide application. Overall wind rose patterns (Fig. 4) for September indicate mostly southerly flow with a large fraction of wind speed at 2–5 m/s. On the other hand, during October, there appears to be a shift to calmer conditions and more variable wind direction (e.g., northwesterly).

Fig. 4.

Windrose pattern for the first 4 h of each application period relative to the center of the application sites.

In terms of rainfall, Sheppard et al. (2002) summarized the historical records of precipitation in Arizona which dates to the mid and late 1800s. From that record, the average annual rainfall of Arizona is only 322 mm (Sheppard et al., 2002). About 45 percent of the precipitation is in July and August (Adams and Comrie, 1997). Less precipitation in September and October is good for pesticide application. EPA recommends conducting applications when there is no predicted precipitation in the next 24 h (www.epa.gov/reducing-pesticide-drift/).

2.5. Dispersion modeling

We used CALPUFF, which is based on studies and reports by Scire et al. (2000a, 2000b) and Exponent (2019), to model pesticide spray drift dispersion. CALPUFF is suitable for regulatory use for long-range transport and on a case-by-case basis for short-range applications involving complex and non-steady-state flows such as in complex terrain, (Abdul-Wahab et al., 2011). It has been recommended and is preferred by the EPA for modeling atmospheric dispersion of pollutants from various emission sources (Abdul-Wahab et al., 2018). The CALPUFF package used included: CALMET version 6.5.0, CALPUFF version 7.3.2, and CALPOST version 7.2.0. As mentioned, we also use WRF to model meteorology instead of using empirical meteorological data from a set of surface stations. Three-hourly WRF model outputs (at 35 km resolution) are extracted and written to a specific format that can be used as input to CALMET using a utility CALWRF provided as part of CALPUFF model package. These are then horizontally and vertically interpolated to 1-km resolution using CALMET, which is a diagnostic wind field processor with micro-meteorological models. CALMET creates hourly wind and temperature fields which serve as the main meteorological data input to CALPUFF (Guo et al., 2020; Scire et al., 2000a). The use of model outputs for meteorology has been shown to provide a more consistent micrometeorology in CALMET relative to studies using data from meteorological sites (Cui et al., 2011; Pivato et al., 2015; Song et al., 2006; Zivan et al., 2016). This is particularly the case for predicting pesticide dispersion from spray drift. In fact, an increasing number of studies in recent years have coupled CALPUFF and WRF for investigating and predicting atmospheric dispersion of pollutants (Cui et al., 2020; Deb et al., 2014; Guo et al., 2020; Lee et al., 2014; Wu et al., 2018).

With regards to emission inputs to CALPUFF, we assume that the pesticide application is a square agricultural field with emissions represented as four spray line sources for each simulation. Source receptors are set every 500 m to provide smoother concentration and deposition fields. We use the MESOPUFF II scheme in CALPUFF (Scire et al., 1984) to represent chemical transformation of pesticides volatized as vapors during application (Route 1 and Route 4 in section 2.3). In particular, we treat this vapor as a sulfur dioxide species and use MESOPUFF for the oxidation of sulfur dioxide (Guo et al., 2020). A description of the MESOPUFF II dispersion model can be found in Scire et al. (1984). Here, we modify MESOPUFF to simulate the half-life of lambda-cyhalothrin by prescribing a chemical loss rate consistent with its half-life of about 7 days. All the other physical parameters were changed to those of lambda-cyhalothrin. Small particles in Route 2 (see Fig. 2) are also treated as particles with a median diameter of 42 μm and a standard deviation of 1 μm. We ran CALPUFF for 48 h after application. If there are pesticide applications on two continuous days, we set them in two separate simulations. Lastly, we use CALPOST to calculate and sort the average and peak concentration and deposition from CALPUFF.

As mentioned, we ran an ensemble of CALPUFF simulations to represent potential range in key model parameters. For vapors (Route 1 and 4) and small particles (Route 2), we vary the following: a) application height, b) emission rate, c) month and year of WRF meteorological data, and d) particle median diameter and its standard deviation. These details are summarized in Table 2 with baseline simulation settings denoted in italics. The baseline simulation, which uses our best estimate of these model parameters, serves as our basis for examining pesticide dispersion in the area with the ensemble spread to represent the possible range of enhancements to ambient levels. Variations in emission rate have been previously described in section 2.3. We also use a median application height of 3 m with standard deviation of 2 m based on AZDA pesticide application records. Release heights for aerial application are relatively low because planes fly as low as possible to reduce spray drift. For meteorological conditions, we vary the meteorological input to CALMET by adopting a simulated meteorology for a similar month or year. Here, we assume that uncertainties in meteorology can be represented from the inherent month-to-month and year-to-year variability. For the Route 2 small aerosol particles, we conducted simulations with different median diameters and standard deviations. With this ensemble approach, we expect to quantify the range of spatial and temporal patterns in resulting abundance of pesticide spray drift representative of late September/early October. This can then help in assessing whether and how much the distribution of vapors and the deposition of small particles reach nearby communities as well as guide potential measurement sampling and monitoring strategies in the area.

Table 2.

Ensemble matrix of model variables and parameters varied in CALPUFF. Italics denote setting for baseline simulation.

Vapor (Route 1 and 4)
Emission rates	Height	Year	Month
5%	1 m	2010	November
10%	3m	2011	September/October
30%	5 m	2012	December
Small Particles (Route 2)
Height	Median Diameter	Standard Deviation
1 m	4.2 μm	0.6 μm
3m	42 μm	1.0 μm
5 m	0.42 μm	0.3 μm

Note: baseline simulation settings are italicized.

3. Results and discussions

3.1. Spatiotemporal evolution

The transport of vapors is strongly dependent on wind speed and wind direction. This is illustrated in Fig. 4 and succeeding figures (Figs. 5 and 6). From Fig. 4, simulated wind patterns relative to the center of application sites show significant shifts in dominant wind speed and direction for each application. This is manifested as variations in the spatial pattern of pesticide abundance and its evolution (Fig. 5). In the simulation for two applications (west side of the domain) during September 19, 2011, the wind is southwesterly at ~3 m/s transporting pesticides towards the northeast side of the domain. Concentrations of pesticide spray dropped significantly within the first 4–8 h of application (from 500 ng/m³ to 0.05 ng/m³) as it spread across to the northeast quadrant of the domain. In contrast, the wind in the simulation for the application (east side of the domain) during September 27, 2011, is easterly and northeasterly at 1–2 m/s. As such, the concentrations of pesticide spray spread from the application site only very slowly to about 15 km in 4 h. These variations are evident in other application periods based on the wind roses shown in Fig. 4. Our analysis for all 9 application days (Fig. 6) also show that wind direction and wind speed have significant influence on the local dispersion of these pesticides. Note that all applications are only allowed when prevailing wind speed is less than 4.44 m/s (10 mph) (US EPA, 2017). In some instances, especially in September, simulated wind speeds in the application sites are higher than allowed. It is unclear whether actual wind speeds are within 4.44 m/s.

Fig. 5.

Time evolution of pesticide near-surface abundance (in ng/m³) for select application periods (09/19 and 09/27). The application site(s) is (are) indicated as yellow stars. Note that the start of application for 09/19 is 01:00 UTC while for 09/27 it is 04:00 UTC.

Fig. 6.

Ensemble mean of the 48-h average pesticide (vapor) concentrations (in ng/m³) resulting from each set of pesticide application(s). Note that these concentrations are plotted in log10 scale to show the extent of pesticide abundance with minimum and maximum of the colorbar corresponding to < 0.1 and > 100 ng/m³, respectively.

3.2. Period average vapor concentrations

A lambda-cyhalothrin background vapor concentration of 0.1 ng/m³ is set for all simulations. We based this background value from the study of Coscollà et al. (2010), where they reported concentrations of many currently used pesticides in ambient air in France are around 0.1 ng/m³ (Coscollà et al., 2010). Similar levels of a few pg/m³ have also been reported by Nascimento et al. (2018) from various locations and across several active and passive sampling approaches with detection limits of less than a pg/m³. The resulting enhancement of pesticide abundance for each application period across the ensemble simulations are shown in Fig. 6 (ensemble mean) and 7 (ensemble spread). The calculated mean concentration near the emission site is about 10–100 ng/m³. These results are consistent with the results by Zivan et al. (2016) when differences in applications are appropriately accounted for. The median EMAH in this work is about 0.86 k g/km²/h (Table 1) while the EMAH in Zivan et al. (2016) is about ten times higher (9.38 kg/k m²/h). We attribute this difference to pesticide emissions in Zivan et al. (2016), where a larger fraction came from volatilization of pesticides after applications since they used chlorpyrifos as the active ingredient, which has a high volatilization rate. The mean concentration of chlorpyrifos in Zivan et al. (2016) is around 3000 ng/m³ while the mean concentration of lambda-cyhalothrin near the application sites in each simulation is 10 times lower (around 300 ng/m³). We note that such consideration needs to be incorporated in model comparisons.

As can be seen in Fig. 6, the distribution of the simulation period average vapor concentrations mainly depends on the wind, consistent with USEPA regulations on wind speed during pesticide application. These spray drift plumes are relatively thin especially when prevailing wind direction is relatively uniform. Spreading out occurs mostly in areas of lower wind speed and/or when low level advection is limited by terrain (e.g., Sep-19 and Sep-27). Some of these variations in ensemble mean (i.e., mean across the ensemble CALPUFF simulations) of the period-averaged vapor concentrations can also be attributed to variations in meteorology. Simulations with different month or year have winds blowing in more than one direction. This is clearly seen as different plumes resulting from the same application. It is therefore important that a range in wind values is taken into consideration as these thin plumes can be ‘misplaced’ in the simulation due to errors in wind direction. Overall, for these 9 application periods, most spray drifts are directed to the east. As such, Wellton is more likely to be affected by the spray drifts than Fortuna Foothills.

This likelihood is supported by Fig. 7, wherein we show the ensemble spread of the simulation period-averaged vapor concentration enhancements in these simulations. It is evident that the spatial pattern of the vapor plumes from these drifts is mostly influenced by meteorology. However, the spread (standard deviation) around the baseline simulation is larger than its mean. Variations are more pronounced in ensemble spread than in the ensemble mean. We attribute this difference in magnitude to variations in emission rates applied in some ensemble members (Table 2). Analysis of each ensemble simulation show that height does not have a significant influence on the mean concentrations of these vapors. Hence, the similarity in spatial pattern between ensemble mean and spread suggests that differences in plume direction for the same application is caused by the use of different meteorology realization. From Fig. 7, we can see that for most cases, the direction of pesticide dispersion changes when the meteorological background is changed. With relatively complex terrain (see Fig. 1), the wind direction may change a lot even within the domain.

Fig. 7.

Same as Fig. 6 but for the ensemble mean spread of the 48-h average pesticide (vapor) concentrations (in ng/m³). Min/Max of the colorbar correspond to < 0.1 and > 100 ng/m³, respectively.

3.3. Period peak vapor concentrations

In Fig. 8, we present the ensemble means of the identified peak (maximum) vapor concentration across the two-day simulation period for each set of applications. The spatial pattern of the vapor plumes in terms of peak concentration are also very similar to those of period-average concentrations in Fig. 6. However, we can see much higher magnitudes in Fig. 8 with some simulations reaching concentrations of 1000 ng/m³ near the application sites and beyond. While these values are significant relative to ambient concentrations of 0.1–6 ng/m³ for most types of pesticides reported in literature (e.g., Degrendele et al., 2016; Figueiredo et al., 2021; Li et al., 2011; Nascimento et al., 2018; Van Dijk and Guicherit, 1999; White et al., 2006; Zivan et al., 2016), note that in most of these simulations, the peak values tend to drop very quickly since transport by advection dominates with higher wind speed, bringing these plumes more than 10 km away just within an hour.

Fig. 8.

Same as Fig. 6 but for the ensemble mean of the 48-h peak pesticide (vapor) concentrations (in ng/m³). Min/Max of the colorbar correspond to < 0.1 and > 100 ng/m³, respectively.

Most peak concentrations occur when the pesticide application is about to be finished. This is evident in Fig. 8 for the Sep-17 and Sep-27 applications where we can see a more pronounced spatial gradient in peak concentration near its source. The peak concentration decreases away from the application site. For example, we can see from Fig. 8 that the peak concentration over Wellton is much higher than 10 ng/m³ for pesticide applications in Sep-27 compared to Sep-17.

Vapor plumes are also more likely to pass through Wellton. As shown by the concentration time series data in Fig. 9, 48-h peak vapor concentrations range from 1 to 10 ng/m³ in Fortuna Foothills while Wellton may experience a peak concentration of approximately 50 ng/m³. However, for both communities, the peak values usually occur after 21:00, when there are fewer outdoor activities. In addition, after one or 2 h, the concentrations will return to ambient levels. This can be clearly seen in Fig. 9 where the width of the time series is quite narrow. Therefore, higher concentrations caused by pesticide applications will not have a significant influence on exposure in the residents living there. It should be noted that since these plumes are very thin, the concentrations may significantly drop several receptors (> 1 km) away.

Fig. 9.

Timeseries of 48-h peak pesticide vapor concentrations over two nearby communities (Fortuna Foothills and Wellton) resulting from 9 pesticide applications across the domain.

Furthermore, Wellton is more likely to experience higher concentrations because of its location. Wellton is situated in the valley where the mountains on both sides may alter wind direction with more convergence near Wellton (see Fig. 1). Since most application sites are north of Fortuna Foothills, it is unlikely for the vapors to be transported to Fortuna Foothills, unless the wind directions are northerly. In addition, the mountains located northeast of Fortuna Foothills can inhibit pesticide spray drift from the application sites northeast of Fortuna Foothills.

However, it should be mentioned that these are just the result of baseline runs. Although the duration of peak concentrations is short, we expect that areas nearby may still have exposure. With changing application time, these two communities may still experience higher concentrations. More notably, these high concentrations also have implications for the amount of pesticide particles that can be deposited and accumulate in the area with greater potential to be resuspended at a later time.

3.4. Period average particle deposition flux

Fig. 10 shows the ensemble mean of the 48-h average pesticide particle deposition flux resulting from pesticide spray drifts for all 9 application periods. Comparing Fig. 10 with Fig. 6, we can see that the spatial distribution for vapors and particles are similar. The transport of particles is also heavily dependent on winds. Unlike the vapors, these particles (although small) can settle due to gravity and can be deposited more easily than vapors. We can calculate the terminal velocities (Eq (1)) for particles of different particle sizes and roughly assess their characteristic settling times and travel distances (Hinds and Zhu, 2022). That is,

$$V_{TS}=\frac{{\rho }_0{d_a}^{2}g}{18\eta }$$ Eq. 1

where the terminal velocity, $V_{TS}$ is proporional to particle density ${\rho }_0$ and gravity $g$, to the square of particle diameter $d_a$ and inversely proportional to the dynamic viscosity. Using Eq (1), terminal velocities, settling times, and travel distances for particle sizes of 0.1 μm to 100 μm are shown in Table 3. We present the time for the particles to fall from 3 m high and the range they can transport, assuming the horizontal flow has constant wind speed of 4 m/s and uniform wind direction. Note that from Table 2, the particle median diameter for the baseline simulation is 42 μm. For a 3 m release height, it takes about 5 min for particle of this size to fall to the ground. This translates to a travel distance of about 1.2 km, which is close to the model grid resolution for all simulations. Hence, assuming no other factors (i.e., chemical transformation), small particles (Route 2) are heavy enough to fall near the application site. This is consistent with the highest mean deposition value of around 10⁶ ng/km²/h that can be seen in Fig. 10. Dispersion of these small particles can be assumed to be limited. However, it should be noted that only nine applications of pesticides with a total of > 20 ensembles are simulated in this study. We expect more applications of pesticides and more applications in different application sites than we simulated. With more emissions, we expect that there will be more small particles that can accumulate. Potential resuspension of these small particles may pose a potential risk to nearby residents.

Fig. 10.

Ensemble mean of the 48-h average pesticide (particle) deposition flux (in ng/km²/h) resulting from each set of pesticide application(s). Note that these concentrations are plotted in log10 scale to show the extent of pesticide deposition.

Table 3.

Deposition estimates for different particle sizes.

Median Diameter (μm)	Terminal velocity (m/s)	Time needed to fall from 3 m high	Transport range (km)
0.1	3.01 × 10−7	2768.55hr	> 50
0.42	5.31 × 10−6	156.95hr	> 50
1	3.01 × 10−5	27.69hr	> 50
4.2 (baseline)	5.31 × 10−4	1.57hr	22.61
10	3.01 × 10−3	16.61min	3.99
42	0.0531	56.60s	0.23
100	0.301	9.967s	0.04

In both Figs. 10 and 11, the ensemble mean and standard deviation can be mainly divided into three different levels. These levels are mainly governed by three diameters in the ensemble matrix for small particles. The standard deviation of the droplet size also plays a role on the spread since the mass $\left(m\right)$ distribution can vary a lot with variations in diameter, $d$ or radius, $r$. That is,

$$\frac{dm}{dr}=4\rho \pi r^2$$ Eq. 2

Fig. 11.

Same as Fig. 10 but for ensemble spread of the 48-h average pesticide (particle) deposition flux (in ng/km²/h).

Eq. (2) implies that both large diameters and large standard deviations can restrict the area of the spread for small particles. This conclusion is consistent with the estimates of Nsibande et al. (2015). In this case study, off-target pesticide drift was monitored during ground application of a pesticide mixture to a sorghum field in South Africa. Results of their sensitivity analysis support our finding (while expected) that droplet size distribution on spray drift is an important factor to consider in modeling dispersion.

3.5. Sensitivity test on half-life of active ingredient

Many previous studies on spray drift neglect the chemical transformation of pesticide species (Pivato et al., 2015). The half-life of lambda-cyhalothrin in air is 12 h to 7 days (Table 1), which means a loss rate of about ~0.4% per hour. The fraction of the chemical transformation is therefore small in this work. However, there are some short-lived active ingredients like pyrethrin. In addition, some chemically active species will be involved in the transformation. For example, chlorpyrifos can turn to chlorpyrifos oxon by photo-oxidation (Zivan et al., 2016). It is necessary to account for this loss in the overall budget of pesticide dispersion. Here, we reconfigure CALPUFF to facilitate a more convenient method to account for variation in chemical reactivity, specifically to incorporate differences in loss rates across pesticide species. As previously described, we used the MESOPUFF II scheme to simulate the loss rate of pesticides. We conducted a short sensitivity analysis by varying the loss rate, mimicking the half-life in air of a shorter-lived species, bifenthrin. We show in Fig. 12 the difference in mean vapor concentrations of two species: lambda-cyhalothrin (half--life: 7 days) and bifenthrin (half-life: 2 days), using the same model settings for all other parameters in the baseline run. The half-life of bifenthrin is about two days, which means a loss rate of about 1.4% per hour. This translates to a difference of about ~1.0% of the original concentrations. As expected, we see very similar spatial pattern on these difference plots due to our prescription of loss rate. Though the difference shown Fig. 12 is small, the approach of prescribing the loss rate may be useful in future studies on short-lived species, especially when used within an ensemble framework. In addition, simulations for active species that can react with other vapor constituents in the air can employ a similar MESOPUFF II chemical transformation scheme.

Fig. 12.

Differences in 48-h average pesticide (vapor) concentrations resulting from pesticide applications using lambda-cyhalothrin and bifenthrin.

3.6. Mass balance

Part of the diagnostic in CALPUFF is a data file on domain-wide mass balance of modeled constituents. We present the mass fraction of emitted pesticides (all routes) that either advected out, chemically transformed, and/or deposited through in-cloud scavenging and rainout (wet) and settling and inertial impaction (dry). Table 4 lists the summary of mass balance for both vapors and particles for all 9 application periods and 9 application sites. These statistics can help us to track these pesticide spray drifts. Since lambda-cyhalothrin is relatively long-lived with a half-life of 12 h to 7 days, most of its vapor is advected out of the domain. It only takes five to 6 h for the vapors to travel 50 km. In these simulations, the deposition of the vapors occurred in the first hour. The deposition of the vapors is mainly contributed by dry deposition since there was little precipitation for these periods.

Table 4.

Domain-wide mass balance (in gram units) of 48-h simulation (baseline run).

	Emissions	Advected Out	% Deposited Within Domain	Chemical Transformation	Deposition		Burden
					Wet	Dry
Vapors	3770.59	3770.59	0.54%	−12.75	0.20	7.51	0.00058
Small Particles	51.18	0.24	99.5%	0	0	50.94	3.09e-05

In the baseline simulations, we consider a median diameter for the small particles as 195 μm and the standard deviation is 88 μm (Table 1). As previously discussed and can be seen in Fig. 10 and Table 3, most small drops of this size can only be transported less than 1 km Table 4 shows that about 99.5% of the small particles will deposit in the study site. Given the limited particle deposition range (section 3.4), we turned off chemical transformation and wet deposition in the simulations for small particles. It should be noted that if the median diameter becomes smaller, larger number of small particles can be transported farther, potentially increasing exposure to nearby residents.

4. Summary and discussion

In light of the lack of measurements on ambient levels of pesticides in Yuma County (despite increasing studies in other parts of the world on the health impacts of pesticide spray drifts), we explore in this study the potential levels of atmospheric exposure through dispersion modeling. We conducted an ensemble of CALPUFF/WRF model simulations to represent the pesticide spray drift in the area and assess its potential short-term 48-h influence on exposure in surrounding areas. Building on recent work on pesticide applications by Pivato et al. (2015) and Zivan et al. (2016), we configured CALPUFF to incorporate a capability to represent the processes affecting local dispersion of pesticide spray drifts, including a capability to vary key model parameters and inputs in CALPUFF (i.e., emission rate, application height, particle diameter and standard deviation, meteorology, and lifetime). This is to provide a more robust assessment of the potential risks of these spray drifts by representing the variability of model parameters in these simulations. We selected 9 application periods on 9 application sites in September and October 2011, for pesticides with lambda-cyhalothrin as the active ingredient and emission rates based on pesticide application records. We then investigated the spatial and temporal distribution of pesticide abundance enhancement across a 50 km × 50 km domain centered around Yuma County, with two nearby communities (Fortuna Foothills and Wellton) as potential areas of exposure to these pesticides. Our results show that the influence of spray drift from these nine sites for September and October in 2011 on these nearby communities is limited but strongly depends on prevailing wind patterns.

The ensemble-mean and period-average concentrations are about 300 ng/m³ in the area close to the application sites. We note however that we do not have sufficient epidemiological studies to gauge whether this level is problematic to cause health outcomes. The highest mean deposition rate for the small particles is only 10⁶ ng/km²/h, but the potential risks of resuspension of accumulated deposition need to be assessed in the future. Using droplets of larger sizes can shrink the dispersion area. In some simulations, the peak concentrations can reach 1000 ng/m³ near the application site just after the application finishes, with the vapors tending to spread but concentrations dropping quickly as well. The time series of concentrations in the baseline simulations at Fortuna Foothills and Wellton are no more than 50 ng/m³. The time at which these spray drifts reach these two communities are after 21:00 local time in most simulations. We also implemented a simple capability to represent chemical transformation for active pesticide species based on MESOPUFF II chemical scheme. Since lambda-cyhalothrin is relatively long-lived with an atmospheric half-life of 12 h to 7 days, most of its vapor equivalent is advected out of the domain, while pesticide particles mostly deposit in the domain. When the vapors are advected out of the domain, the concentrations usually range from 0.1 to 50 ng/m³ in the simulations, depending on the location of application sites and emission rates. Our results show that the influence of exposure from pesticide spray drifts from 9 application sites to nearby communities is not significant. However, if the applications are more intensive and number of applications within the country are larger, the concentration enhancements may be higher and pose a potential risk to nearby residents. For species with stronger volatilization rates, secondary drift might also be important (Zivan et al., 2016).

Although the resulting concentrations and depositions are relatively small in this study, it should be noted that only several applications of insecticides are simulated in this study. We emphasize that this study presents an initial assessment of a possible modeling configuration which can be scaled up in the future to account for more applications. Hence, findings from this study should be interpreted as indicative (rather than definitive) of potential influence. As earlier mentioned, there are more applications of herbicides and more applications in different application sites. With possible increases in pesticide applications beyond what was simulated, exposure is expected to also increase. In fact, Sugeng et al. (2013) reported that 5.7 million pounds (lbs) of pesticides were applied annually in Yuma County. We find however a lower estimate of 24,641 applications per year (amounting to 3115 lbs of active ingredient per year) averaged across 1995–2016 from recent AZDA records. In 2011 alone, there were 5808 applications (13,740 lbs) for September and October, 824 (162 lbs) of which used lambda-cyhalothrin. Note that for this initial modeling study, we only simulated 2.5% of total applications (~4 lbs) using lambda-cyhalothrin in the county. Hence, we expect significantly higher concentrations of these pesticide spray drift even if we scale our results to applications using lambda-cyhalothrin alone. Even more significant concentration enhancements can be expected if we consider all applications (and active ingredients) across Yuma County. More simulations on a more complete set of applications are clearly warranted to provide a more realistic assessment of the influence of intensive pesticide applications in Yuma County. We envision that the ensemble approach with CALPUFF/WRF will also aid in quantifying ranges on these exposure estimates. Like many other deterministic models, interpretation of results can be limited by our understanding of key physical properties and processes (Gil and Sinfort, 2005). This is especially the case on the evaporation processes. In light of increasing association of potential health outcomes from agricultural pesticides (e.g., Furlong et al., 2020a, 2020b; Rull et al., 2006; Shelton et al., 2014), improved capabilities to monitor pesticide abundance, including field measurements for quantifying baseline and ambient concentrations, especially with resuspension, as well as to calibrate and evaluate dispersion models, are strongly recommended, given the increase in pesticide usage, recent reports on their alarming health outcomes, and potential impact of warming in recent years on volatilization of these spray drifts.

HIGHLIGHTS.

• An ensemble-based model assessment of pesticide spray drift over Yuma County was carried out for select application sites.

• Model results show low pesticide vapor concentrations over nearby communities but are strongly dependent on prevailing wind.

• More detailed studies, including field measurements, are imperative to better assess the real scale of these impacts.

Data availability

Data will be made available on request.