1.1 Scale

The first objective of the Environmental Release Tool (ERT) is to identify the applied toxicity of pesticides over a large extent to promote targeted pesticide toxicity reduction strategies. While governments largely regulate pesticides at the national/multinational level [18, 19], few tools are available to model pesticide impacts across the large and heterogeneous scales managed by regulators (e.g., SYNOPS-WEB [13]). Existing watershed fate and transport models adaptable to specific regions, such as SWAT [20–22] or HSPF [23–25] often require global calibration methods due to limited data for model parameters, and global calibrations can mask toxicity sources. Global calibrations also require monitoring data [26], and for many watersheds, no or limited observations are available for pesticides. Moreover, a fate and transport model’s parameterization and calibration for a single pesticide and watershed can take hundreds of hours, and its uncertainty is compounded over large extents [26–28]. Collectively, these factors can obscure important variations and hinder the identification of toxicity reduction opportunities over a large extent.

1.2 Sources

Another key obstacle to mitigating applied toxicity is identifying pesticide sources and application timing [29–31]. Pesticide use data are often not recorded and reported, making it difficult to predict watershed-level, applied toxicity. While the Pesticide National Synthesis Project [32] provides the best dataset for pesticide use across the United States, the data are too coarse for this purpose. However, the state of California is unique; its Department of Pesticide (CDPR) has the most comprehensive pesticide use database in the world, with daily reports of agricultural applications since 1990 [33]. Currently, no tool is available with which to efficiently use the database to summarize or map the spatiotemporal distribution of pesticide toxicity. The second objective of the ERT is to automatically load and process data to prioritize toxicity reduction while providing the flexibility to quantify the applied toxicity distribution of pesticides in the United States and beyond.

1.3 Cumulative applied toxicity

Pesticide-contaminated soils and surface waters typically contain mixtures of active ingredients [3, 34–36]. The cumulative applied toxicity is the sum of the toxicity of all pesticides applied at a given site within a particular period (e.g., daily, monthly, annually). Often, the pesticide with the highest toxicity is considered the most relevant one, and it is common to assume that the most toxic pesticide can approximate the cumulative applied toxicity in a given sample [37, 38]; however, to reduce pesticide toxicity in a water body throughout the year, it is imperative to understand the cumulative applied toxicity, which can exhibit significant temporal variation in the pesticides responsible for most toxicity [31].

Most watershed models, even where high-resolution pesticide use data is available, cannot accommodate the evaluation of a pesticide mixture within a single simulation; nor are pesticide mixtures regulated in the United States [18, 37], with few state-level exceptions for specific pesticide classes [38]. The variability of mixtures [39] and knowledge gaps related to the synergism or antagonism of a pesticide mixture to non-target organism toxicity [40] often make quantifying their combined effect challenging. Although tools are available that simulate and summarize pesticide mixture toxicity, such as regression models [41] and cumulative risk maps [12], they have limited spatiotemporal extent. To address the cumulative risk knowledge gaps, our third aim is to quantify the spatiotemporal applied toxicity of various pesticides at specific application site types to facilitate toxicity reduction initiatives.

1.4 Economic and health scores

Reducing pesticide-applied toxicity can affect the economic prosperity of agricultural stakeholders [42, 43]. Many options exist for considering the economic benefits and health concerns of pesticides, and consideration of economic impacts is mandated by current US legislation [37]. However, no tools are known to the authors to dynamically quantify health and economic indices for pesticide application sites. Therefore, the fourth objective of the ERT is to generate scores for the various application site types that prioritize toxicity reduction opportunities while also considering the economic benefits relative to health impacts. Though external factors affecting treatment areas, such as water and stress to crop yields, make it difficult to quantify pesticide use benefits, simple health, and economic scores can be leveraged to explore effective strategies.

1.5 Toxicity reduction targets

The ERT is a novel tool for targeting toxicity reduction designed to be user-friendly for experts (researchers) and non-experts (e.g., the public), with results presented in an interactive heatmap with graphical summaries. The development of the ERT sought to overcome the limitations of existing tools for prioritizing pesticide mitigation opportunities by integrating features to 1) assess variation in applied toxicity at the regulation extent (1,000+ watersheds), 2) identify primary sources of environmental toxicity, 3) calculate cumulative toxicity, and 4) generate impact scores that consider economic and health aspects of toxicity reduction.

To identify toxicity reduction strategies in diverse landscapes, California is an ideal study site due to available information on over 400 agricultural application site types and the substantial pesticide use, accounting for 20% of the mass of pesticide sales in the United States and 3% worldwide [44, 45]. Our study leveraged the ERT to identify toxicity reduction targets for aquatic taxa in California’s watersheds. It aimed to answer four questions: 1) How is toxicity distributed among pesticides for diverse taxa? 2) What are the opportunities for toxicity reduction in specific application site types? 3) Does quantifying the cumulative toxicity enhance our understanding of environmental toxicity? 4) Which application sites have the greatest applied toxicity, and what economic benefits do they have?

2.1 Scale

To enable evaluations of the variability of pesticide toxicity over large extents, the tool summarizes pesticide applications and toxicity by watershed. The data is summarized by watershed, and applications sites as well as pesticides within since pesticide losses via runoff and eroded sediments share a common outlet. Summarizing applied toxicity by watersheds is important to conceptualize areas that share common hydrologic routes for pesticide transport. Though the Environmental Release Tool does not simulate loss processes, it is the first stage of development of the Pesticide Mitigation Prioritization Model, and the product of the second stage of development is a companion, mechanistic fate and transport tool where loss processes are simulated.

Watersheds in the ERT are delineated using the Watershed Boundary Dataset [48], a data product of the United States Geological Survey. Each watershed is assigned a hydrologic unit code (HUC), which is based on the hydrologic connectivity and scale of the watershed. Watersheds with shorter HUCs, such as HUC 2-digit codes, are large watersheds encompassing hundreds of thousands of square kilometers, while longer HUCs such as HUC 8-digit codes (HUC8) represent subwatersheds of the shorter digit codes (e.g., HUC2). The assignment of pesticide use data to watersheds of various spatial extents is facilitated by the tool (see S1 Text).

2.2 Sources

To evaluate the spatiotemporal distribution of pesticides, the ERT benefits from the ability to autoload daily pesticide use report data in California from statewide agricultural applicators [33]. The tool internally hosts the data, and using an autoload script, aggregates data for the area of interest to the user, which watersheds or counties may define. Where counties are used, the tool automatically aggregates data to watersheds in the county. For other pesticide input options (e.g., manual inputs or for analyses of other land uses or regions), see S1 Text.

The amount of pesticide applied on application sites (e.g., a specific crop) in California is substantial, millions of pounds for widely cultivated crops, and as high as ~40 million for almonds [49]. To assist efficient analyses, the ERT extracts pesticide usage data for California from CDPR Pesticide Use Reports [33] by active ingredient (AI) and for the 432 agricultural site types for the study area of interest to the user. These reports record daily applications at the County Meridian Township Range Section (referred to as Section) spatial scale (2.6 km²). For Sections where pesticide use data is reported that overlaps multiple watersheds, the area fraction of overlap is used to weight the mass of AI applied. Notably, urban applications were not included in the autoload feature. The reports do not include household applications, and for professional urban applications are recorded at the county level and at a monthly time-step, which cannot be allocated to a specific watershed or date.

For evaluating pesticide sources of toxicity, ERT facilitates the summarization of similar AIs. This feature is useful because many AIs have a similar chemical make-up (e.g., isomers or are produced in several forms, including acids, salts, amines, and esters), but have no or limited toxicity data for the various AI forms. Provided that AI forms can have very different effect concentrations, where possible, the user should provide AI form-specific toxicity. To accommodate specific endpoints where available, but to enable simplification of tool outputs, unique toxicity endpoints are accepted and calculated for pesticides within a user-defined pesticide group, and the group ID reports the group’s total applied toxicity in tool output. In this investigation, we considered AIs detected (2014–2018) within California’s surface waters with available toxicity data (n = 151). From the CDPR’s Pesticide Use Reports, 290 forms of the AIs were observed (e.g., 12 unique esters and 15 salts of 2,4-D).

In addition to pesticide sources of applied toxicity, a key feature of ERT is the ability to preserve information relating to application site types. However, too many application sites make the interpretation of results difficult. The tool thus enables users to group similar application sites (e.g., alfalfa and alfalfa-grass mixture) by assigning the same ID to multiple site types. By default, 432 agricultural application site types from Pesticide Use Reports are simplified to 116 based on the similarity of the crops. Groupings can be viewed and modified in the tool input file for application sites.

To identify pesticide toxicity reduction targets, the ERT quantifies applied pesticide toxicity. Applied toxicity refers to the mass of pesticide applied to an area with the potential to do harm [50]. The applied toxicity for the i^th pesticide in the j^th watershed is calculable from applications to the k^th site type and toxic endpoint of the m^th taxon of interest as: Eq (1) Where TI is the Toxicity Index (kg-m³/kg), M (kg) is the mass of applied AI, and T (kg/m³) is the adverse health-effect concentration of concern (e.g., the lethal concentration of fifty percent of the test organism population) for the species or taxonomic groups of interest. Within a simulation, the tool is suitable for quantifying the applied toxicity to taxa within the same compartment, not across environmental compartments, because variation in the transport of pesticides based on physicochemical properties is not simulated. The tool illustrates applied toxicity within the soil compartment or available for transport to the compartment of interest. While the transport of pesticides from the application site is sensitive to their physicochemical properties [51], property correlation to surface water detection frequencies has been demonstrated to the more robust for pesticide sales data than physicochemical properties in a monitoring campaign of 72 pesticides of diverse properties in over 100 streams [52]. Though this approach is not suitable for risk assessments, it facilitates an understanding of where mitigation opportunities exist [53] without data requirements and uncertainty of fate and transport models over large extents [26, 27, 54].

Our investigation considers the applied toxicity of pesticides for fish, as well as aquatic invertebrates, nonvascular plants, and vascular plants. Toxicity endpoints employed were acute values from the United States Environmental Protection Agency (USEPA) Aquatic Life Benchmarks Database [55]. The USEPA derives Benchmarks from the concentration at which fifty percent of a species sample in single-dose laboratory investigations experience severe effects derived from mortality endpoints or, for plants, significant changes in growth/biomass (LC50 or EC50). A genera endpoint is then calculated based upon a 0.05 cumulative probability of toxicity for represented species, which typically reflects the most sensitive species within the taxonomic group. For fish and invertebrates, the USEPA calculates the final acute value as the product of the taxonomic group endpoint multiplied by a safety factor of 0.5 and does not adjust plants. Where no toxicity endpoints were reported for the pesticide in the Aquatic Life Benchmark database (n = 10), the Pesticide Properties DataBase [56] acute toxicity endpoints were employed, and unverified data were excluded.

The first applied toxicity index reported by ERT for pesticides, sites, and watersheds is the Relative Toxicity Index (RTI) (kg-m³/kg-m²). The index weights the toxicity of the i^th applied pesticide by the size of the application area within the j^th watershed as: Eq (2) where A (m²) is the area affected by pesticide applications.

To estimate the areas affected by pesticide applications, agricultural land use datasets are used. In California, the Pesticide Use Reports can be used to retrieve the impacted area. However, there are known inaccuracies. The planted area is often recorded for all the grower’s land; although reported for a specific crop, and fields are subject to multiple crop rotations within a year. For the applied area, multiple applications are typical for a crop that renders the net-application area unknown. Due to these concerns, alternative land use datasets were evaluated for use [57–59].

After reviewing several datasets, the California Department of Water Resources land use surveys [60] were found to be the most accurate with a median accuracy of 97.5% and positional quality of 8m. However, a limitation of the dataset, as well as the others, is that it provides fewer site types (43) compared to Pesticide Use Reports (432). Using this dataset to determine the affected area of specific application site types would require highly reducing the resolution of pesticide source data. Attempts to recategorize crops to fit available land use data did not obtain reliable results. As a result, we chose to consider the affected area to be all agricultural land in the California Department of Water Resources dataset. The representation of the affected compartment to all agricultural land was deemed appropriate because only 5% of agricultural fields in California employ organic cultivation practices [61] and use non-synthetic pesticides recorded in use reports.

As the quantification of the affected compartment area is frequently limited, and the fraction of a watershed subject to pesticide application is highly variable, we provide a second applied toxicity index independent of area, the Net Toxicity Index (NTI). The NTI is a relative rank toxicity index to determine if the applied toxicity is greater than what is typical for the i^th pesticide in the j^th watershed. As our reference of what is typical, we calculate for the study area the 50^th percentile (perc₅₀) of the applied toxicity for any applied pesticide (pst) in watershed (w). The NTI is calculable from the TI of the i^th pesticide in the j^th watershed as: Eq (3)

The NTI approach can quickly identify applied toxicity above typical levels in the study extent. For example, if the 50^th percentile of the applied toxicity of pesticides to a watershed in the study area is applications of imidacloprid in the San Joaquin Watershed, (e.g., 1000 TI), to calculate the NTI, the TI of the i^th pesticide and the j^th watershed of interest is divided by 1000 TI. Using this approach, pesticide applications within a watershed over the simulation period with an NTI greater than unity have applied toxicity above the 50^th percentile. This normalization provides a unitless applied toxicity index that does not affect the relative rank of the applied toxicity for pesticides, sites, or watersheds, and can identify effective toxicity reduction targets specific to the study area.

2.3 Cumulative applied toxicity

For single taxonomic group investigations with the ERT (e.g., only fish), the cumulative applied toxicity, the potential of all pesticides released to the environment and under investigation to do harm to the taxon, is calculated via the concentration addition method [62, 63]. We calculate the cumulative toxicity indices of pesticides for each index for n pesticides for a watershed (here w₁) as: Eq (4) Eq (5)

The method used in this study relies on the assumption of additive toxicity and non-interacting chemical species. While this assumption is theoretically unsound for chemicals of diverse modes of action, and this limitation is not addressed by the ERT, pesticides rarely express synergism at environmentally relevant concentrations, and cumulative addition has been empirically demonstrated to be a reliable method for quantifying pesticide mixture toxicity [50]. For example, in studies with hundreds of pesticide mixtures, the method has predicted mortality within a factor of 2 for 90% of samples [64–66]. Additionally, the method is robust to independent modes of action [65]. However, the approach is not suited to simultaneously understand the cumulative toxicity to diverse taxa due to the presence of unique organism receptors and responses [67]. To address this issue, we conducted independent simulations for fish, invertebrates, nonvascular aquatic plants, and vascular aquatic plants in addition to the simulation that quantified the net applied toxicity for all organisms simultaneously.

Available literature considers that most acute toxicity in a pesticide mixture can be represented by a single pesticide [68, 69]. However, the frequency of use of pesticides varies throughout the year, as do pesticides detected in surface waters [31]. To evaluate whether the tool’s ability to rapidly quantify cumulative applied toxicity improves understanding of environmental toxicity relative to evaluations targeting the highest impact pesticide, the cumulative toxicity index (NTI) of sites and watersheds was compared to the index of the pesticide with the highest applied toxicity using a one-way, paired t-test. Paired t-tests are commonly employed for samples measured at two-time points and to compare predictions relative to observed data (the cumulative toxicity) [70].

2.4 Economic and health scores

The health and economic impacts of application sites are quantified over the study extent with a Health Score (ha/NTI), an Economic Score (USD/ha), as well as an Economic and Health Score (USD/NTI per ha). These scores are calculated over the study extent (California) rather than in specific watersheds due to the low resolution of reliable land use data (see 2.2 Sources). The health and economic scores with higher values represent more favorable outcomes. Health and economic scores are calculated for the application site areas of the study extent as: Eq (6) Eq (7)

The Economic and Health Score penalizes crops with high applied toxicity and is calculated as: Eq (8)

For health and economic scores, the harvested hectares and gross value of application site types were compiled from the United States Department of Agriculture National Agricultural Statistics Service (https://www.nass.usda.gov/). We considered the median economic value and harvested hectares for a crop in California from 2014–2018 to minimize single-year anomalies.

In addition to numeric scores, users are also provided with categorical scores based upon percentiles for the study area to facilitate interpretation. Scores are divided into 20^th percentiles and range from ‘Very Low’ (0-20^th percentile) to ‘Very High’ (80-100^th percentile).

2.5 Toxicity reduction targets

To demonstrate the effectiveness of the ERT in providing pesticide toxicity reduction targets, we conducted a comprehensive analysis of pesticide use across the 140 HUC8 watersheds in California (mean area ~3,600 km²); see Fig 1(A). Moreover, we leverage the tools’ ability to perform higher resolution analyses to investigate the applied toxicity to the 208 HUC12 subwatersheds (mean area ~100 km²) of the San Francisco Bay Delta Watershed (BDW) with agricultural pesticide applications; see Fig 1(B). The BDW is an area of ecological significance within California and is home to over 90 threatened or endangered species [71]. To evaluate toxicity reduction targets for aquatic taxa in these study areas, we summarized the applied toxicity to fish, invertebrates, nonvascular aquatic plants, and vascular aquatic plants. To explore temporal trends in the net applied toxicity (10 years) of the study areas, we employed a two-sided Mann-Kendall test.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Watershed applied toxicity.

The heat map and legend values represent applied toxicity as the Net Toxicity Index (NTI), the total applied toxicity of pesticide applications to all aquatic taxa investigated over the simulation period, fish, invertebrates, nonvascular aquatic plants, and vascular aquatic plants. Results are displayed for each study extent, a) California’s HUC8 watersheds and b) the HUC12 subwatersheds in the Bay-Delta Watershed. The NTI ranges identify the magnitude of toxicity released during pesticide applications, with values in the upper range, greater than 50,000,000, illustrating areas of applied toxicity that are up to 8 orders of magnitude greater than other watersheds. Base map source:http://goto.arcgisonline.com/maps/Reference/World_Imagery.

https://doi.org/10.1371/journal.pwat.0000124.g001

The evaluations of toxicity reduction targets in the study areas employ pesticide use data recorded at the Section level, which is smaller than a watershed, and a Section may overlap multiple watersheds (see Section 2.1). To evaluate the accuracy of the method used to assign pesticide use report data to watersheds with Section level data, predicted applications were compared to field-level pesticide use data in Kern County, one of the few counties with field-level data. The accuracy of pesticide use assignment to application sites within watersheds was evaluated relative to field-level predictions using the median absolute percent error (MdAPE). While the root mean square error or mean absolute percentage error are more sensitive metrics, they are both sensitive to outliers, whereas the MdAPE has been demonstrated to be more robust [72]. Since watersheds can vary in pesticide loads by many orders of magnitude, the MdAPE was employed.

3.1 Scale

The evaluation of applied toxicity at the statewide management scale in California illustrated that toxicity reduction targets are concentrated in relatively few watersheds. The net applied toxicity (NTI) to all aquatic taxa investigated from all pesticides used over the simulation period showed 80% of toxicity was applied in only 11% of California’s HUC8 watersheds; see Fig 1A. NTI varies by several orders of magnitude across California, from the lower end of the range (1–100,000) to the high end (>50,000,000). This reflects not only the difference in loading, but also the wide range of toxicities for different pesticides. Watersheds in the 97^th percentile, those with NTI values exceeding 50,000,000, received up to 8 orders of magnitude more applied toxicity compared to other watersheds across the state. These findings enable the identification of areas with high applied toxicity within data-limited watersheds during the analysis period. The information can be utilized to determine specific locations where further investigation of pesticide impacts, such as monitoring or simulating fate and transport, should be focused.

In the case of the BDW (agricultural watersheds exclusively considered), the study revealed that 20% of the watersheds accounted for 80% of the applied toxicity; see Fig 1B. For the California and BDW scales, results highlight the effectiveness of targeting a relatively small fraction of watersheds that receive the highest levels of applied toxicity as a strategy for reducing overall toxicity.

Over large scales, the ERT also facilitates the quick identification of changes over time for different resolutions (daily, monthly, and annual). We leveraged this feature to identify if watersheds were increasing in applied toxicity and may benefit from mitigation efforts. For temporal analyses, we extended the evaluation to the most recent 10 years of available pesticide use data. Our analysis covered 2009 to 2018, during which the total NTI of pesticides used in California’s watersheds increased by 150% in the last five years (2014–2018) compared to the first five years. To assess the significance of the changes in individual watersheds, we used a two-sided Mann-Kendall test (α = 0.05). Our analysis found that applied toxicity significantly increased in 63% of watersheds (p-value < 0.001, tau = 0.9). In subwatersheds of the BDW, a similar trend was observed, with 58% of subwatersheds showing a significant increase in applied toxicity (p-value < 0.001, tau = 0.9). Although the ERT does not simulate transport of pesticides to aquatic ecosystems, these results suggest that to preserve environmental health, efforts may be required to manage increases in applied toxicity.

3.2 Sources

The source of 99.9% of applied toxicity (NTI) was identified in the ERT analysis to be attributed to just 14 chemicals, as outlined in Table 1. Among these chemicals, 10 were classified as insecticides/miticides/acaricides, while the remaining 4 were categorized as herbicides. This represents a concise list of toxicity reduction targets, considering 290 AIs were evaluated.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Pesticide applied toxicity.

Summary of the Net Toxicity Index (NTI) from the release of evaluated pesticides which comprise 99.9% of California’s active ingredient (AI) applied toxicity for 2014–2018. The results summarize the applied toxicity across all pesticides and taxonomic groups investigated (fish, invertebrates, nonvascular plants, and vascular plants).

https://doi.org/10.1371/journal.pwat.0000124.t001

The analysis also revealed that the top two pesticides in terms of applied toxicity, cyhalothrin and bifenthrin, accounted for approximately 90% of the NTI, despite constituting only 1% of the applied mass. This indicates that these two AIs have a disproportionately significant contribution to the overall applied toxicity. Similarly, when focusing on subwatersheds of the BDW, bifenthrin and cyhalothrin were found to contribute approximately 90% of the applied toxicity.

Available surface water and sediment monitoring data during the analysis period indicate that pesticide loads identified as having high applied toxicity have been observed at lethal concentrations to the aquatic taxa investigated. For instance, in monitoring data for agricultural ditches in California for 2014–2018, all pesticides contributing to 99.9% of applied toxicity (Table 1), except for indaziflam, were observed at lethal concentrations or at concentrations where plant growth is inhibited (above Aquatic Life Benchmarks) [73]. Furthermore, concerning the pesticides with the highest NTIs, namely bifenthrin and cyfluthrin, it was found that all 121 samples in which they were detected exceeded the Aquatic Life Benchmarks. It is important to note that even samples where no pesticides were detected may still contain concentrations of concern. This is due to the fact that in 99% of the sample analyses (n = 533), the concentrations could not be detected at levels as low as the Aquatic Life Benchmark (limit of quantification too high).

Notably, bifenthrin and cyfluthrin have a high affinity to sediment, and applied toxicity to benthic invertebrates. However, our study did not include an assessment of their effects on benthic invertebrates as the Environmental Risk Targeting (ERT) analysis was specifically designed to analyze taxa residing in the same environmental compartment (see Section 2.2). Nevertheless, to understand if pesticides with high applied toxicity are reaching sediment compartments, we explored the frequency of their detection of bifenthrin and cyfluthrin and if they were observed at hazardous concentrations. For effect endpoints, we considered acute mortality values published for benthic invertebrates in the PPDB [56] (unavailable through the USEPA benchmark database).

Similar to surface water, all detections (n = 161) in sediment samples (n = 268) exceeded life benchmarks, indicating potentially hazardous levels of bifenthrin and cyhalothrin. Furthermore,100% of sample analyses could not detect concentrations as low as life benchmarks. For these pesticides, an ecotoxicological study in a waterbody in California, with little development besides agriculture, has also observed lethal concentrations in sediment [74].

Notably, bifenthrin and cyhalothrin are among the most challenging pesticides to measure at levels of concern to aquatic taxa. Their benchmarks are in the parts per trillion range, and they have the lowest limit of quantification of pesticides studied except for deltamethrin. Earlier work by Parker et al. (2021) also expressed concern that the limits of quantification used in sample analyses are generally too high for a number of pesticides with high toxicity [75], further emphasizing the need to improve our knowledge in this area.

Regarding California’s diverse pesticide application sites as sources of environmental toxicity for the most recent 5-year data, the ERT identified 90% of toxicity (NTI) was concentrated in only 16 of the 116 site classes investigated. Out of the total released toxicity, 25% was applied to almonds, the most widely cultivated crop analyzed, and 19% was applied to pistachios. The other seven crops with the highest applied toxicity contributed 1%-11%. Furthermore, in most watersheds, a few site types contributed most of the NTI, although the sites with the highest applied toxicity varied from one watershed to another. For example, in the Salton Sea watershed, which is one of the most heavily impacted in the state, 80% of the toxicity was caused by 4 of 72 application site types: alfalfa, sweet corn, lettuce, and broccoli.

When examining the sources of applied toxicity in California’s watersheds with increasing NTIs across the state, the analysis revealed that the largest increase in applied toxicity was primarily attributed to two pesticides: cyhalothrin (63%) and bifenthrin (27%). In terms of application site types in these watersheds, the highest NTI was applied to almonds (36%) and pistachios (30%). In the BDW, the rise in applied toxicity was also primarily contributed by cyhalothrin (75%) and bifenthrin (23%) and nut orchards, with the site types with the greatest applied toxicity being almonds (75%) and walnuts (19%). It is noteworthy that these nut orchards are among the most widely cultivated crops, with almond, pistachio, and walnut orchards spanning ~800,000 ha.

3.3 Cumulative applied toxicity

The ERT is unique in its ability to assess the applied toxicity of all pesticides used in specific application sites over time. Typically, when evaluating the acute toxicity of pesticide mixtures, researchers often rely on a single pesticide to approximate within a given time frame, given differences in timing within a given season, application amount per event, and frequency of use of a given pesticide within a mixture [31]. It is also inadequate for large areas with diverse pesticide usage patterns, where the most toxic chemical applied can vary. The cumulative applied toxicity provides a more complete representation of the potential impact. To determine whether the tool’s ability to rapidly quantify cumulative applied toxicity enhances our understanding of environmental toxicity, we compared the cumulative applied toxicity to that of the single pesticide with the highest applied toxicity using one-way, paired student’s t-test, with a significance level of 0.05. This analysis was conducted for all watersheds and sites within watersheds, considering monthly and annual intervals for each of the four aquatic taxonomic groups under investigation. To provide a more robust evaluation of cumulative toxicity, the analysis of significance was performed at the watershed level rather than the overall study extent. This approach was chosen for two main reasons. Firstly, it allowed for a larger sample size (n = 226), enabling a more comprehensive assessment of cumulative applied toxicity. Secondly, within a watershed, the landscape where pesticide applications occur is hydrologically connected with its waterbodies.

Results of the analysis evaluating the significance of accounting for cumulative applied toxicity demonstrated cumulative applied toxicity to be significantly greater (p < 1.0E-16) than the maximum applied toxicity of individual pesticides (Fig 2). These results are also valid when evaluating each taxonomic group at monthly and annual time steps (Table 2), as well as for site types in each watershed. Across taxa, extents, and time steps, the mean cumulative NTI predictions were 118–634% greater than the mean NTI of individual pesticides. The significance observed in accounting for cumulative applied toxicity aligns with previous chronic and sub-lethal mixture studies [76, 77]. However, our findings differ from previous studies that have focused on acute, single-sample mixture toxicity. These acute studies often suggest that toxicity can be represented by the pesticide that contributes the most toxicity [68, 69]. This discrepancy is likely observed because acute mixture analyses of individual samples do not account for seasonal and interannual variability [31, 78–80], while our results capture greater temporal variability in pesticide use and presence.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Watershed cumulative applied toxicity.

Cumulative applied toxicity (red) for the simulation period versus the maximum individual pesticide NTI (blue) for fish in the ten watersheds with the highest applied toxicity in California, 2014–2018. Cumulative applied toxicity to fish was significantly greater than the maximum individual pesticide toxicity for watersheds in the study extent (a = 0.05, p < 1.0E-16).

https://doi.org/10.1371/journal.pwat.0000124.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Taxa cumulative applied toxicity.

Results are presented for the one-way, paired t-test (∝ = 0.05), the percent difference of means, and the median absolute percent error (MDAPE) for the maximum applied toxicity (NTI) of individual pesticides relative to the cumulative applied toxicity across California’s HUC8 watersheds for the analysis period. Analyses were performed for four aquatic taxonomic groups by watershed and by site, at monthly and annual time steps. Cumulative applied toxicity was significantly greater than maximum for all taxa, sites, watersheds, and time-steps (a = 0.05, p < 1.0E-16).

https://doi.org/10.1371/journal.pwat.0000124.t002

In addition to improving our understanding of net environmental toxicity, quantifying cumulative toxicity reveals notable spatiotemporal patterns for targeting toxicity reduction. Here we provide examples of the cumulative toxicity to aquatic invertebrates seasonally and interannually.

Examining monthly cumulative applied toxicity trends to invertebrates revealed that relying solely on single chemical analyses would not capture the chemical with the highest toxicity for a month or season. For example, in the first year of the simulation across all watersheds in California, esfenvalerate had the highest individual monthly applied toxicity at the beginning of the year, followed by bifenthrin in the middle of the year, and cyhalothrin towards the end of the year (Fig 3). Different trends were observed for specific application sites and watersheds. In the case of the widely cultivated crop table grapes, the primary contributors to monthly cumulative applied toxicity during fall and winter were chlorpyrifos and oxyfluorfen, while most of the cumulative toxicity in summer resulted from fenpropathrin applications (Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Temporal trends.

Temporal trends of monthly cumulative applied toxicity to aquatic invertebrates, the Net Toxicity Index (NTI), for pesticides which introduce 99.9% of released toxicity for California’s HUC8 watersheds a), as well as select, high-impact crops which include artichokes b), spinach c), and table grapes d). Annual trend lines (black) illustrate average year-to-year increases in cumulative toxicity to aquatic invertebrates (NTI).

https://doi.org/10.1371/journal.pwat.0000124.g003

The cumulative toxicity contributed by all pesticides to a taxon also varied in intensity over the year. Grapes had the greatest monthly cumulative applied toxicity in the spring and other crops, like artichokes, in the late summer/fall (Fig 3). Since California’s wet season begins in late fall/winter, later pesticide applications may introduce higher pesticide concentrations in runoff than those performed earlier in the year.

Interannually for invertebrates, cumulative toxicity was observed to increase in California’s watersheds. Year-to-year, a significant increase was revealed by a Mann-Kendall test (∝ = 0.95, p<0.001), with an average increase over the study extent of 7.6%. While some sites, such as spinach, exhibited similar trends with a 6.9% increase, there were significant differences observed, highlighting the need for targeted risk reduction. Artichokes showed a mean year-to-year NTI decrease of -6.9%, making them a less effective target compared to table grapes, which had a higher NTI and an average yearly increase in toxicity of 2.8% (Fig 3).

3.4 Economic and health scores

To evaluate the trade-offs between the health and economic impacts of reducing the toxicity of application sites, we calculated health and economic scores. A higher score indicates a more favorable outcome. The Health Score (ha/NTI) considers the cultivated hectares and the applied toxicity of site types. Of application sites contributing 90% of the applied toxicity, crops with the lowest Health Score (greatest applied toxicity per hectare) include strawberries, sweet corn, and pistachios (Table 3). Among these three crops, the Economic Score, as well as the Economic and Health Score, were least favorable for sweet corn and most advantageous for strawberries. Our findings suggest sweet corn may have the lowest health and economic benefit with contemporary pesticide use practices of the study area and should be targeted in mitigation efforts.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Environmental Release Tool outputs for pesticide application site types which introduce 90% of the applied toxicity in California where: AI (kg) represents the mass of active ingredients used; NTI the Net Toxicity Index; NTI (%) the percent of the total NTI of the application site type; Economic, Economic Health, and Health Scores reflect the numeric score of different site types; and the Economic, Economic Health, and Health Values the quantiles of scores where 0-20^th quantile is ‘Very Low’ and 80-100^th ‘Very High’.

https://doi.org/10.1371/journal.pwat.0000124.t003

3.5 Toxicity reduction targets

To prioritize targets for reducing pesticide toxicity, we used the ERT to identify a shortlist of pesticides, application site types, and watersheds responsible for 80% or more of the applied toxicity. If efforts targeted just one pesticide, or one pesticide and application site type, it would act on 50% or more of the applied toxicity for 89% of watersheds in California. However, the specific pesticide or pesticide and site type with the most applied toxicity varied by watershed.

The accuracy of the ERT for predicting applied toxicity with available information depends on the reliability of pesticide use assignment to watersheds (see Section 2.2). To evaluate the accuracy of applied mass of pesticides to watersheds with the CDPR database (resolution of 2.6 km²), ERT predictions using these data were compared to predictions that employed pesticide use data available at the field level in Kern County. The evaluation showed the ERT provides a reasonable estimate of the spatiotemporal distribution of pesticides, with an observed MdAPE of 1.7% for AIs applied in watersheds. Furthermore, all predictions of the applied mass of pesticides to watersheds were within 5% of the field-level data. An improved MdAPE of 0.01% was observed for application site predictions per watershed.