Challenges with the current methodology for conducting Endangered Species Act risk assessments for pesticides in the United States

The US Environmental Protection Agency (USEPA or the Agency) is responsible for administering the Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA). The Agency is also required to assess the potential risks of pesticides undergoing registration or re‐registration to threatened and endangered (i.e., listed) species to ensure compliance with the Endangered Species Act. To assess potential risks to listed species, a screening‐level risk assessment in the form of a biological evaluation (BE) is undertaken by the Agency for each pesticide. Given the large number of registration actions handled by the USEPA annually, efficient tools for conducting BEs are desirable. However, the “Revised Method” that is the basis for the USEPA's BE process has been ineffective at filtering out listed species and critical habitats that are at de minimis risk to pesticides. In the USEPA's BEs, the Magnitude of Effect Tool (MAGtool) has been used to determine potential risks to listed species that potentially co‐occur with pesticide footprints. The MAGtool is a highly prescriptive, high‐throughput compilation of existing FIFRA screening‐level models with a geospatial interface. The tool has been a significant contributor to risk inflation and ultimately process inefficiency. The ineffectiveness of the tool stems from compounding conservatism, unrealistic and unreasonable assumptions regarding usage, limited application of species‐specific data, lack of consideration of multiple lines of evidence, and inability to integrate higher‐tier data. Here, we briefly describe the MAGtool and the critical deficiencies that impair its effectiveness, thus undermining its intention. Case studies are presented to highlight the deficiencies and solutions are recommended for improving listed species assessments in the future. Integr Environ Assess Manag 2023;19:817–829. © 2022 The Authors. Integrated Environmental Assessment and Management published by Wiley Periodicals LLC on behalf of Society of Environmental Toxicology & Chemistry (SETAC).


INTRODUCTION
In the United States, crop protection products are registered under the Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA), which is administered by the US Environmental Protection Agency (USEPA or the Agency). The registration process is rigorous and involves the application of conservative risk assessment methods and assumptions that are intended to overestimate risk (Moore et al., 2021). Registration of a pesticide is considered an authorization of an "action" under section 7(a)(2) of the US Endangered Species Act (ESA). As such, the USEPA is required to ensure that pesticide registrations are unlikely to jeopardize threatened and endangered species or adversely modify their designated critical habitats (CH). The USEPA evaluates potential risks of a pesticide to each of 1600+ federally listed threatened and endangered species in a biological evaluation (BE). A prescribed hierarchical ecological risk assessment approach is followed in each BE, as described in the Revised Method for National Level Listed Species Biological Evaluations of Conventional Pesticides ("Revised Method") (USEPA, 2020a). To facilitate the implementation of the Revised Method, the Agency developed a new tool called the Magnitude of Effect Tool (MAGtool) (USEPA, 2020b). The Agency's intent is to use the Revised Method, along with the MAGtool, for efficiency and high throughput during the reregistration process for conventional pesticides, but with appropriate scientific and/or technical changes as the scientific knowledge base improves.
Each pesticide BE prepared by the Agency makes one of three determinations for each listed species: no effect (NE), not likely to be adversely affected (NLAA), or likely to be adversely affected (LAA). Analogous determinations are also made for CH of listed species that have them. Subsequently, the US Fish and Wildlife Service (USFWS) and the National Marine Fisheries Service (NMFS) (collectively, the "Services") review the BE to determine if they concur with the BE determinations and prepare a Biological Opinion (BiOp). For those species that are LAA, the Services then determine whether they are in jeopardy and, if so, what mitigation measures are required to eliminate the jeopardy concern. Potential mitigation measures include reduced application rates and frequency, no spray buffers, no applications during crop flowering, county bans where listed species are found, and many others.
The purpose of each BE is to identify those species and critical habitats potentially at risk from use of a pesticide. Unfortunately, the BEs released to date have dramatically overstated risk to listed species. This result is due to not only the compounding conservatism of the risk assessment process but also a highly conservative protection goal, that is, species are deemed by the USEPA to be LAA if there is the potential for adverse effects to one or more individuals of a listed species. In their jeopardy determinations, the Services do not focus on the individual but rather determine whether the pesticide of interest poses a risk to the listed species population.
Overall, the USEPA's BE process has not been effective at filtering out listed species and critical habitats that are at de minimis risk, leaving the Services to assess far more listed species than necessary. Consider the insecticide malathion as an example. In 2017, the USEPA released their final BE for malathion, which found that 97% (1778 of 1835) of listed species (including candidate and proposed) were likely to be adversely affected by continued malathion use (USEPA, 2017a). However, the USFWS initially determined that only a small fraction of the species deemed LAA were potentially in jeopardy due to continued malathion use (N = 78) (USFWS, 2021). Subsequent species-specific analyses by the USFWS further reduced the number of jeopardy species and, following consultations with the registrant and other stakeholders, mandatory risk mitigations were enacted to eliminate 100% of jeopardy concerns in the final BiOp (USFWS, 2022).
In recent BEs (USEPA, 2020c(USEPA, , 2020d(USEPA, , 2020e, 2020f, 2020g, 2020h, 2021a(USEPA, , 2021b(USEPA, , 2021c(USEPA, , 2022c, application of the MAGtool (USEPA, 2020b) has been the methodological approach used and the primary reason for failing to screen out listed species that are clearly not at risk. In this paper, we briefly describe the MAGtool and the critical deficiencies that impair the effectiveness of the tool. We then describe realistic and efficient improvements for assessing listed species and their designated critical habitats that address these deficiencies. Case studies are presented to highlight the issues and methods for improving listed species risk assessments. In more recent BEs (e.g., USEPA, 2022aUSEPA, , 2022bUSEPA, , 2022c, the USEPA not only assigned effect determinations but also began to make likely jeopardy (likely J) and likely adverse modification (likely AM) calls. The case studies provided herein provide a pathway for future effect determinations and likely J and/or AM calls that are scientifically defensible, transparent, and efficient. These adjustments will provide the USEPA and the Services with sufficient information to make realistic and reasonable management decisions for species protection when registering pesticides.

THE MAGNITUDE OF EFFECT TOOL
The MAGtool and its companion methods represent a complicated software ecosystem that the USEPA relies on to develop each pesticide BE. The MAGtool v2.3.1 software architecture is based on Microsoft-Excel™ including Microsoft Visual Basic for Applications™ scripts and is supported by many other applications (e.g., Pesticide Water Calculator and Variable Volume Water Model (PWC/VVWM), Plant Assessment Tool (PAT), ESRI ArcGIS™). The MAGtool contains listed species data, generic nomograms (i.e., concentrations in different receptor groups normalized to application rate), processing functions, and generates output files that require considerable interpretation. At the core of the MAGtool (Figure 1) are the USEPA's standard screening-level models (e.g., PWC/VVWM, terrestrial exposure model [T-REX], Terrestrial Herptile model [T-HERPS]) (USEPA, 2008(USEPA, , 2012. The screening-level models have been parameterized to be specific to listed species (e.g., obligate and general dependencies, body weights, diets, habitats, methods of dispersal). For each of the USEPA's screening-level models, conservative assumptions are applied to account for uncertainties and variability in the data or to address a lack of data (e.g., default foliar half-life of 35 days). In all cases, worst-case input parameters and application scenarios are used to predict risk to one individual of a species. The MAGtool is intended to consistently implement the Revised Method, which consists of two steps.
Step 1 identifies potentially exposed species by evaluating the overlap of species ranges and critical habitat locations with pesticide use data layers (UDLs). This process is illustrated in Figure 2. The UDLs are crop grouping layers that spatially identify where use patterns are located. For those listed species that overlap with one or more UDLs, a screening-level risk assessment is conducted for both direct effects and adverse effects to prey, pollination, habitat, and dispersal (PPHD) upon which each species depends (i.e., indirect effects). In Step 2, the MAGtool implements the quantitative aspects of the Revised Method as well as qualitative information (e.g., is the species potentially extinct, extirpated, or found only in remote habitats and thus will not be exposed?). Thus, spatial evaluations (e.g., magnitude of overlap), incorporation of usage data, aquatic and terrestrial exposure modeling, derivation of effects metrics, and risk estimation are handled within the MAGtool itself ( Figure 3). The Agency also conducts a quantitative weight-of-evidence (WoE) analysis in the MAGtool by changing key assumptions (e.g., assuming average expected usage in a species range rather than worst-case usage) to be less conservative. The goal of the WoE analysis for each listed species is to determine whether the risk conclusion (e.g., NLAA, LAA) is altered by the changes. The sensitivity of the risk conclusion to changes in key assumptions is inversely related to the level of confidence that the USEPA assigns to the risk conclusion.

ISSUES WITH THE MAGNITUDE OF EFFECT TOOL
Conducting a National Endangered Species Assessment for~1600+ listed species for pesticides that typically have numerous use patterns is a complex and difficult task. The approach and methods are expected to improve over time with experience and additional scientific knowledge. Based on reviews of the BEs conducted to date, we have identified several major issues with the MAGtool to be addressed to improve the utility and efficiency of the BE process: • Compounding conservatism. The MAGtool is essentially a compilation of prescriptive screening-level models and other tools within an automated software framework. Each component has its own conservative input parameter requirements and assumptions. The aquatic exposure modeling component lumps all listed aquatic species as occurring in only a few habitats (e.g., the standard farm pond, index reservoir), which do not cover the wide diversity of habitats inhabited by listed aquatic species in the United States. The terrestrial exposure models in the MAGtool are applied in a highly conservative manner (e.g., assuming 100% consumption of the dietary item with the highest expected pesticide concentrations for each listed species, 100% of contacted pesticide is dermally absorbed by birds). To assess acute risk to listed aquatic and terrestrial species, the tool assumes that all individuals in a population have sensitivity equal to the most sensitive individual and that the species itself is highly sensitive. To account for offsite transport due to runoff and downstream transport, the USEPA assumes that a near stream assessment covers risk to listed aquatic organisms further downstream. Unfortunately, because the USEPA does not use a proper flowing water model (e.g., Soil and Water Assessment Tool or SWAT; Bieger et al., 2017), dilution, pesticide degradation, and other fate and behavior properties are not considered as the pesticide moves downstream into the species range. Clemow et al. (2018) describe limitations in the refined aquatic exposure model that was developed for malathion and used to assess risks to three listed species: California red-legged frog (Rana draytonii), delta smelt (Hypomesus transpacificus), and California tiger salamander (Ambystoma californiense). Brain et al. (2015) and Moore et al. (2021) describe many other sources of conservatism associated with the screening-level models that are part of the MAGtool. As currently constructed and applied, the MAGtool is designed to maximize Type I error (i.e., falsely identify a listed species or critical habitat that is not adversely impacted by a pesticide), thus minimizing Type II error (i.e., not identifying a listed species or critical habitat that is being adversely impacted by a pesticide). As a result, there have been inflated numbers of MA/LAA determinations in the BEs conducted to date (Table 1). Thus, the Revised Method (USEPA, 2020a) and its implementing software (i.e., MAGtool) do not allow for the meaningful identification of listed species or critical habitats that are reasonably certain to be exposed to and potentially adversely affected by one or more uses of a pesticide as legally defined by the label.
• Lack of a weight-of-evidence assessment. The USEPA has relied strictly on the MAGtool results in making their effects determinations for listed species in the BEs. Mesocosm studies, field studies, incident reports, biological surveys, and targeted monitoring studies that would support or refute the modeling line of evidence have not been seriously considered because these lines of evidence are not required inputs of the MAGtool. We will consider this topic in more detail in the case studies. • Usage data are incorrectly considered. Not all use sites of a crop or other use pattern on a pesticide label will be treated in any given year. Pesticides are often rotated among different modes of action to prevent development of pesticide resistance or may only be used for specific pests during sporadic outbreaks (Moore et al., 2021). Usage data are used to account for the reality that not all potential pesticide use sites will be treated with a specific active ingredient. However, there is uncertainty regarding which use sites have been treated in recent years because usage data are generally only available at the state level in the United States.
California is a notable exception as the state collects agricultural usage data at a much finer scale. Two usage scenarios are generally considered in the BEs and applied in the MAGtool. The first is a highly conservative scenario that assumes the maximum Percent Crop Treated (PCT) in a species range. In this scenario, the USEPA assumes that all treated acres in a state are entirely located within a species range up to the maximum of the available treated acres in the range. This assumption leads to a severe overestimate of exposure, particularly when the species range is much smaller than the size of the state. The Agency bases their NLAA and/or LAA determinations on the maximum PCT assumption. In their WoE analysis, the USEPA investigates the impact of assuming average PCT across the state, that is, usage is uniformly distributed across a state and species range. However, the results assuming average PCT, although somewhat more realistic, only affect the level of confidence that the Agency assigns to the overall effects determination and not the effects determination itself.
Consider the draft atrazine BE (USEPA, 2020e), which had a starting point of 1795 listed species, including proposed and candidate species, as an illustrative example of the above and other issues with the MAGtool. Listed species in Hawaii, Alaska, Puerto Rico, and other US territories received "no effect" determinations solely because the registrant proactively agreed to limit the use pattern for atrazine to the lower contiguous 48 states in the future. Other listed species that had no potential for exposure (e.g., extinct or extirpated species, species only found on remote islands or in deep ocean habitats) were not considered with the MAGtool (USEPA, 2020b). The remaining listed species assessed with the MAGtool in the lower contiguous United States were deemed LAA, requiring formal ESA Section 7 consultation with the Services (USEPA, 2020e). As shown in several of the case studies that follow, consideration of other lines of evidence or species-specific information clearly demonstrates that many species deemed LAA by the USEPA for atrazine are, in fact, not at risk from the pesticide. This was also comprehensively demonstrated in the ESA perspective published by Smith et al. (2021). Similar results were obtained with the carbamate BEs (USEPA, 2020c, 2020d) and neonicotinoid BEs (USEPA, 2021a(USEPA, , 2021b(USEPA, , 2021c. Clearly, the current BE approach and methods are failing to identify listed species and their critical habitats that have little potential to be impacted by a pesticide.

PROPOSED SOLUTIONS FOR FUTURE ASSESSMENTS: CASE STUDIES
The following discussion highlights several of the proposed solutions that result in more realistic estimates of potential risk for the effect determinations, likely J and/or likely AM calls, and to provide scientific rationales for potential mitigations.

Issue #1-Compounding conservatism
To evaluate the impact of compounding conservatism on MAGtool risk estimates, we determined how changes in assumed species tolerance to atrazine affected the risk conclusions for two listed plant species. The intent was to determine the inflection point for changing a risk conclusion from an LAA to an NLAA determination for the dicot San Diego thornmint (Acanthomintha ilicifolia) and the monocot Sonoma alopecurus (Alopecurus aequalis var. sonomensis). Both species received LAA determinations in the draft atrazine BE (USEPA, 2020e).
We modified the chemical input file for atrazine following the instructions provided for version 2.3.1 of the MAGtool model (USEPA, 2020b). In brief, the toxicity endpoints for terrestrial plants in the toxicity inputs file for atrazine were multiplied by several orders of magnitude (i.e., 10X to 10 000X). Thus, terrestrial plants were assumed to be orders of magnitude more tolerant than they potentially are. No other changes to the inputs assumed by the USEPA were made. The multiplication factor required to obtain an NLAA determination was 1000X for Sonoma alopecurus and 10 000X for San Diego thornmint. Essentially, plant species would have to be almost completely tolerant to a herbicide for listed terrestrial plant species not to receive an LAA effects determination, irrespective of proximity to potential use sites.
The reason why plant species must be highly tolerant to receive an NLAA designation is because the MAGtool dramatically overestimates terrestrial exposure. Exposure to atrazine for the San Diego thornmint and Sonoma alopecurus was overestimated for the following reasons: • Neither of the selected species are exposed to atrazine because they do not occur in habitats that would be in treated or near treated areas. The San Diego thornmint is   (USFWS, 2009(USFWS, , 2011. • The species ranges of the San Diego thornmint and Sonoma alopecurus are not found close to approved use patterns for atrazine. We conducted a proximity analysis to characterize the spatial relationship between the species ranges and potential atrazine use sites (i.e., use patterns included on atrazine labels in California). In brief, proximity distances from all pixels (i.e., using a 30 m × 30 m discretization) within a species range or critical habitat to the nearest potential use site (i.e., represented by crop footprints) were determined and used to create proximity probability distributions. Based on our proximity analysis (see Moore et al., 2017 for method details), the 1st percentile proximity distance to the closest potentially treated crop (i.e., hay, grass and/or turf) is 122 m for the San Diego thornmint and 288 m for the Sonoma alopecurus. This means that 99% of the species ranges of these two listed species occur at even farther distances from the closest potentially treated crops for atrazine. At the 1st percentile distances, assuming worstcase aerial applications in the USEPA's spray drift model, AgDrift, the fractions of applied material reaching 122 and 288 m are 0.025 and 0.011, respectively. Thus, our conservative exposure estimates at these distances are 40and 90-fold lower than what the USEPA assumed. Given the much-reduced atrazine exposure estimates for these species and the low atrazine usage in California for turf uses (CalDPR, 2022), potential risks of atrazine to our case study species are exceedingly low.
As specified in the ESA, the best available data and realistic and reasonable assumptions are required when evaluating risks to listed species and their critical habitat. The case study suggests that a review of the habitat and spatial data for these species would enable the USEPA to reduce the compounding conservatism of the screeninglevel models and arrive at effect determinations that more clearly reflect the potential for risk.

Issue #2-Lack of consideration of species-specific information
Because of its prescriptive nature, the MAGtool often fails to consider critical species-specific foraging behaviors, diets, and habitats, many of which are highly specialized. For example, the Alameda whipsnake (Masticophis lateralis euryxanthus) has a near obligate dependency on western fence lizards (Sceloporus occidentalis) and western skinks (Eumeces skiltonianus) for its diet, though it occasionally preys on frogs, small mammals, nesting birds, and other snakes (USFWS, 2021). In the thiamethoxam BE, a neonicotinoid insecticide, the USEPA (2021c) found that the Alameda whipsnake was likely to be adversely affected by this pesticide because of effects on its prey. The current implementation of the MAGtool, however, only considers terrestrial insects in estimating the effects of pesticides, including thiamethoxam, to the prey of the Alameda whipsnake. Terrestrial insects are infrequently consumed by this species (USFWS, 2021) and there is no evidence that reduced availability of insect prey would have any impact on the Alameda whipsnake. As with the Alameda whipsnake, the issue of focusing on an inconsequential dietary item has played out for other terrestrial wildlife species, for example, the Everglade snail kite (Rostrhamus sociabilis plumbeus) and the salt marsh harvest mouse (Reithrodontomys raviventris) as discussed below.
Everglade snail kite (ESK). The ESK is a wide-ranging raptor found in Florida, with other subspecies found throughout the Caribbean, Central America, and south to Argentina and Peru. The ESK was listed as endangered in the United States in 1967 and has maintained a small population since. It has a highly specialized diet comprised almost exclusively of apple snails (Pomacea paludosa), though the ESK has been observed feeding upon crayfish and, on one occasion, speckled perch (Cary, 1985;Reichert et al., 2020;USFWS, 1999). The principal threats to the species are habitat loss, fragmentation, and degradation due to agriculture and urban development. The most recent population estimate for the ESK was 1754 birds in 2014 (Reichert et al., 2020), showing somewhat of a plateau after a period of sharp decline in the late 1990s and the early 2000s (Martin et al., 2007). Everglade snail kites live primarily near freshwater marshes and at the edges of lakes usually with interspersed emergent vegetation and riparian trees such as willow and wax myrtle on the shoreline. They require clear and open areas to forage for apple snails, which are aquatic but climb emergent vegetation to breathe, feed, and lay eggs (Beissinger, 1988;Sykes, 1979).
In their draft BE, the USEPA (2021c) concluded that the ESK was likely to be adversely affected by use of thiamethoxam in citrus, grapes, vegetables, ground fruit, other crops, field nurseries, developed areas, etc. due to predicted effects of thiamethoxam on its prey. The EPA's Pesticide in Water Calculator (PWC) was used and the results were input into the MAGtool to estimate upper bound concentrations in water for the generic standard pond known as Bin 7 (i.e., high-volume static water body, 20 000 m 3 ) in south Florida, where the ESK is found. The ESK habitat generally comprises much larger freshwater systems (e.g., several lakes) (USFWS, 1999) and thus the upper bound Bin 7 concentrations are highly conservative. The toxicity endpoint used to estimate risk to prey of the ESK was from a species sensitivity distribution (SSD), that is, an acute HC05 of 3.58 μg a.i./L derived for clothianidin by the USEPA using toxicity data for aquatic insects only. Aquatic insects have a high sensitivity to neonicotinoids including thiamethoxam (Finnegan et al., 2017;Prosser et al., 2016;USEPA, 2021aUSEPA, , 2021bUSEPA, , 2021c. The USEPA (2021c) assumed that clothianidin is a reasonable toxicity surrogate for thiamethoxam because the former is a metabolite of thiamethoxam in soils, insects, and plants (Fan & Shi, 2017). This is a dubious assumption, given that the USEPA's aquatic exposure analysis stated that thiamethoxam represents most of the residue in water, likely due to the lack of degradation of thiamethoxam to clothianidin in water. Thus, there is no reason for the USEPA (2021c) to have used a clothianidin toxicity endpoint instead of a thiamethoxam toxicity endpoint for freshwater invertebrate prey.
The available toxicity data indicate a wide range of sensitivity for aquatic invertebrates to thiamethoxam, with acute EC50s and/or LC50s spanning over four orders of magnitude ranging from 5.5 μg a.i/L for the mayfly (Neocloeon triangulifer) to >100 000 μg a.i./L for several noninsect species ( Figure 4). Because the ESK is nearly completely dependent on apple snails for prey, it is critical to consider the availability of snail toxicity data in the assessment of effects to ESK prey. Three species of aquatic snails have been tested for sensitivity to thiamethoxam. They are highly tolerant to thiamethoxam exposure, with EC50s and/or LC50s ranging from 6195 μg a.i./L for the ramshorn snail (Planorbella pilsbryi) (Prosser et al., 2016) to >100 000 μg a.i./L for the great pond snail (Lymnaea stagnalis) and the wandering pond snail (Radix peregra) (Finnegan et al., 2017).
According to the USEPA (2021c), 1-in-15 year daily thiamethoxam concentrations in habitats assumed to be favored by the ESK and their apple snail prey (i.e., Bin 7 in southern Florida) are predicted to range up to 14.2 μg a.i./L for the field nurseries and the developed use patterns. This upper bound value for the worst-case use patterns is 429-fold below the EC50 and/or LC50 of the most sensitive aquatic snail species. Given the safety margin between the upper bound predicted concentration in ESK habitats and the most sensitive acute LC50, thiamethoxam does not pose a risk to the only prey species of importance to the Everglade snail kite. In addition, because the studies conducted by Prosser et al. (2016) and Finnegan et al. (2017) involved static, without renewal, exposures in the acute studies, any conversion of thiamethoxam into clothianidin would have been accounted for in the test results. Thus, there is no need to consider clothianidin toxicity to aquatic snails.
Compared with available monitoring data, the estimated concentrations of thiamethoxam in Bin 7 in southern Florida are overestimated, likely by orders of magnitude. Monitoring data obtained from the National Water Quality Monitoring Council's Water Quality Data Portal by the USEPA (2021c) indicate that of the 1604 samples taken from 499 sites in southern Florida from 2011 to 2021, 1426 had thiamethoxam concentrations below the detection limit. The concentrations in the remaining samples ranged from 0.0082 to 2.7 μg a.i./L. Thus, the margin of safety for apple snail prey is far greater than the nearly 429-fold safety margin estimated above when using the worst-case exposure concentration and the most sensitive EC50 and/or LC50 for aquatic snails.
Salt marsh harvest mouse. The salt marsh harvest mouse (Reithrodontomys raviventris, SMHM) is a rodent in the family Muridae (subfamily Sigmodontinae) (USFWS, 2010). The SMHM is endemic to the saline and brackish marsh habitats of San Francisco Bay and its tributaries in the central coast area of California (Bias & Morrison, 1999). The SMHM's primary habitat is pickleweed (Salicornia virginica)dominated vegetation. The value of pickleweed habitat increases with its depth, density, and degree of intermixing with fat hen (Chenopodium album) and alkali heath (Frankenia salina) (USFWS, 2021). This type of habitat is preferred because it provides year-round cover from predators, an escape from flooding, reduced competition with other small mammals, and an important food source, that is, pickleweed (Bias & Morrison, 1999;Fisler, 1965;Geissel Integr Environ Assess Manag 2023:817-829 © 2022 The Authors wileyonlinelibrary.com/journal/ieam FIGURE 4 Acute species sensitivity distribution for aquatic invertebrate species exposed to thiamethoxam. Data are from the PMRA (2021) and Miles et al. (2017). Arrows indicate unbounded EC50s and/or LC50s et al., 1988). The diet of the SMHM comprises seeds, grasses, leaves, plant stems, forbs, and insects (Brylski, 1999;USFWS, 1984). The diet varies seasonally based on the availability of vegetation. The SMHM typically consumes fresh green grasses in the winter and pickleweed and saltgrass throughout the rest of the year (Fisler, 1965).
The major threats to the SMHM are habitat loss and alteration of the vegetation zones of San Francisco Bay marshes, as the species has very specific habitat requirements (Geissel et al., 1988;Shellhammer et al., 1988;USFWS, 1984USFWS, , 2010. Approximately 80% of the SMHM's historic tidal marsh habitats have been destroyed and many of those that remain support few to no mice due to filling, diking, subsidence, changes in water salinity and vegetation, nonnative species invasions, sea-level rise associated with global climate change, and pollution (Bias & Morrison, 1999;USFWS, 2010). Pesticides are not listed as a threat to the SMHM (USFWS, 2021).
In the draft BE for thiamethoxam, the USEPA (2021c) concluded that the SMHM was likely to be adversely affected because of effects to terrestrial insect prey for six use patterns (i.e., poultry litter, open space developed, developed, other crops, field nurseries, other orchards). The MAGtool exposure estimates for these six use patterns ranged from 6.08 to 17.3 mg a.i./kg ww and were based on a nomogram for insects assuming maximum label application rates for the flowable formulation, that is, prey of the SMHM were assumed to be exposed on treated areas rather than in nearby downwind areas. The corresponding toxicity endpoint used to estimate risk to prey of the SMHM was the most sensitive 48-h LD50 of 0.032 mg a.i./kg ww for the Asiatic honey bee (Apis cerana) exposed via contact to thiamethoxam (USEPA, 2021c).
The draft thiamethoxam assessment for the SMHM was inaccurate because the assessment assumed that (1) their prey occurs exclusively on treated areas immediately after application of the flowable formulation and (2) effects to terrestrial invertebrates would ultimately adversely affect the SMHM despite plants being the primary dietary item for this species. We explore each of these issues below.
The SMHM is endemic to the emergent wetlands of San Francisco Bay and its tributaries and is generally restricted to saline or brackish marsh habitats. Such habitats are not conducive to agricultural crops, nor would they be developed. A proximity analysis was conducted using the same approach as that described for the San Diego thornmint and Sonoma alopecurus species range and atrazine crop footprints. The 1st percentile proximity distances from the SMHM species range to agricultural crop footprints are large, that is, 360 m for citrus to >2501 m for other crops. The remaining 99% of the species range would be at greater distances from agricultural use patterns. The 1st percentile proximity distances for most crops exceed the maximum distances beyond which the USEPA assumes zero spray drift for aerial (i.e., 792 m), airblast (i.e., 549 m), and ground (i.e., 305 m) applications for other grains, other crops, and cotton. At 360 m, the USEPA's spray drift model (i.e., AgDrift) estimates an upper bound fraction of applied of 0.0000474 for airblast application to citrus orchards. The corresponding mean insect concentration for the citrus use pattern is 0.00053 mg a.i./kg ww, which is nearly two orders of magnitude below the most sensitive 48-h contact LD50 of 0.032 mg a.i./kg ww for the Asiatic honey bee. The mean arthropod concentration for ground applications to all other agricultural use patterns, given the 1st percentile proximity distances, is zero. For aerial application, only one use pattern produced a mean arthropod concentration that slightly exceeded the acute effects metric for Asiatic honey bees, that is, vegetables and ground fruit (0.0402 mg a.i./kg ww). Given that insects are a minor part of the diet of the SMHM and the conservativeness of the USEPA's AgDrift spray drift model (Moore et al., 2021), the slight exceedance is not a concern for the diet of the SMHM. The above analysis was not done for nonagricultural use patterns because of data limitations. Usage of flowable thiamethoxam for nonagricultural use patterns is quite low in the United States (USEPA, 2021c).
The diet of the SMHM is nearly entirely comprised of vegetative matter (Fisler, 1965). The USEPA (2021c) found that thiamethoxam and clothianidin have very low toxicity to terrestrial plants. The only meaningful biological effect was observed for oilseed rape at an application rate of 0.26 lb a.i./A in a vegetative vigor test. Expected application rates on broadleaf foliage in SMHM habitats would be orders of magnitude below this toxicity endpoint at the 1st percentile proximity distances derived for the agricultural use patterns for which thiamethoxam may be used in California. The USEPA (2021c) found no concern for direct effects of thiamethoxam to the SMHM and thus the overall finding should have been that thiamethoxam is not likely to adversely affect this species.
Species-specific information plays a critical role in the risk characterization for listed species. The highly conservative screening-level approach applied by the USEPA neglects much of this information when developing the BE effect determinations. The USEPA has made some strides in this direction when developing likely J and/or AM calls in more recent BEs (e.g., USEPA, 2022a, 2022b, 2022c. However, further progress is needed to address the compounding conservatism that leads to poor effect determination decisions to ensure that the process is more efficient in the future.
Issue #3-Failure to consider other lines of evidence and higher-tier data In making NLAA and/or LAA calls for listed species in their BEs for pesticides, the USEPA has nearly exclusively relied on one line of evidence, that is, comparison of conservative modeled exposure values to prescribed toxicity endpoints for the most sensitive taxa from required ecotoxicity studies to support pesticide registration via the MAGtool. This is despite other lines of evidence being available including monitoring data, higher-tier effects data from mesocosm and field studies, and biological surveys. According to the USEPA (2016), ecological risk assessments should consider all available lines of evidence in a weight-of-evidence framework to (i) provide context to the modeling risk characterization, (ii) support whether an adverse effect is likely or not likely to occur, and (iii) apply all best available data as required by the US Endangered Species Act.
Consider malathion as an example. For this organophosphate insecticide, there are numerous in situ, whole medium, mesocosm, and field studies that should have been considered in the USEPA's final Biological Evaluation (USEPA, 2017a). In that BE, the USEPA found that 97% of listed species were likely to be adversely affected including threatened and endangered aquatic invertebrate, fish, bird, mammal, herptile, and terrestrial invertebrate species. For listed aquatic invertebrate species or listed species that depend on that receptor group, there are several lines of evidence that were not considered for use of malathion to control adult mosquitoes. For example, Phillips et al. (2014) determined the aquatic toxicity of freshwater samples collected following application of insecticides, including malathion, to control mosquitos in California. Some of the insecticides, particularly the naled breakdown product, dichlorvos, were acutely toxic to aquatic invertebrates in receiving systems, but malathion did not cause acute toxicity. Jensen et al. (1999) showed that ultralow volume applications of malathion used to control adult mosquitoes did not substantially affect aquatic invertebrate or fish populations in seasonal wetlands in Central California. For listed herptiles, such as the Wyoming toad (Anaxyrus baxteri), a field study showed that amphibians and reptiles were unaffected by malathion applications in a small, forested watershed (Giles, 1970). Dickerson et al. (2003) showed that malathion drifting into the habitat of the listed Wyoming toad did not reduce adult survival or prey availability, nor did it affect predator avoidance behavior.
Nine mesocosm studies are available that have determined the effects of malathion on aquatic communities (Brogan & Relyea, 2015;Ebke, 2002;Halstead et al., 2014;Hua & Relyea, 2012Nataraj & Krishnamurthy, 2012;Relyea, 2005Relyea, , 2009Shrestha et al., 1987). The studies included periphyton, phytoplankton, macrophyton, zooplankton, macroinvertebrates, fish, and aquatic-phase amphibians. The studies showed that impacts to sensitive biota of aquatic communities can occur, but generally at concentrations well above those found in the environment. Recovery of these communities was also generally rapid.
To assess potential effects to birds, Sotti and Laucht (2018) conducted a field study involving applications of malathion to citrus orchards in Italy. They found no evidence of any long-term effects such as reduced breeding success or mortality of young reared by exposed individuals or on the bird community itself in treated orchards.
The above-cited higher-tier studies are by no means exhaustive for malathion. These and other studies should have been evaluated as additional lines of evidence in the malathion BE but were not. There is also an extensive monitoring data set available for malathion, including studies targeted to areas of intensive malathion use (e.g., P. Anderson et al., 2004Anderson et al., , 2007P. D. Anderson & Dugger, 2008;Burke et al., 2006). The monitoring data set should have been used to characterize the aquatic modeling predictions for malathion and as an independent line of evidence.
Atrazine is one of the best-studied compounds on the planet. As result, over 40 microcosm and mesocosm studies have been conducted to evaluate the effects of atrazine on aquatic plant communities. The USEPA (2020e) found that atrazine was likely to adversely affect many listed aquatic invertebrate, fish, and other species because they depend on aquatic plants for food and habitat. The USEPA has reviewed these studies many times in the past, as have several Scientific Advisory Panels and others (e.g., Giddings et al., 2018;Moore et al., 2017). Yet, no mention was made of the aquatic microcosm and mesocosm studies, nor were they considered in the weight-of-evidence assessments for listed aquatic plant species or other listed plant species that depend upon aquatic plants for habitat or food. Further, the MAGtool application for atrazine did not consider the extensive, targeted monitoring data set available for atrazine in surface waters of the Midwest, where usage is highest (Perkins et al., 2021).
For terrestrial plants, Brain and Anderson (2019) showed that the USEPA's standard approach for estimating spray drift effects to downwind plant communities (i.e., comparing AgDrift predictions of exposure to the results of greenhouse tests), which is the approach used in the MAGtool for atrazine, grossly overestimated impacts measured in the realworld field study. The Brain and Anderson (2019) study was conducted under clearly worst-case conditions (i.e., wind speeds above maximum label recommendations, test plants placed on bare ground with no intercepting vegetation present, test species known to be highly sensitive to atrazine). Incorporation of the Brain and Anderson (2019) study into the draft atrazine BE would represent a significant improvement to the weight-of-evidence assessments for listed terrestrial plant species and to listed species that depend on terrestrial plants for habitat and food.
Considerable documentation is available on conducting qualitative and quantitative weight-of-evidence assessments for regulatory decision making (e.g., Hall et al., 2017;Linkov et al., 2009;Lutter et al., 2015; Society of Environmental Toxicology and Chemistry [SETAC], 2018). Risk assessments concerning listed species have also been conducted with a weight-of-evidence component and illustrate how lines of evidence, including the modeling line of evidence, are incorporated into the risk characterization to inform effect determinations (Clemow et al., 2018;Moore et al., 2016;Whitfield-Aslund et al., 2017). The application of higher-tier data in higher-tier risk assessments provides important, relevant, consequential, and contextual information to ensure that realistic and reasonable effect determination calls are rendered. We recognize that not all higher-tier studies may be entirely relevant for this purpose. Levine et al. (2019) summarized the findings and recommendations from the 2017 Workshop on Regulation and Innovation in Agriculture on how regulatory authorities can include data from higher-tier studies in pesticide ecological risk assessments and for risk management. Some of the workshop recommendations regarding open and transparent communication between regulators and the regulated, guidance for higher-tier study design, evaluating the relevance and reliability of the higher-tier data should be implemented in the USEPA's future BEs. This requires the USEPA to move beyond a screening-level risk paradigm to a refined assessment calibrated to the level of detail available, thereby facilitating a more informed, reasonable, and realistic risk evaluation.

CONCLUSIONS
The USEPA conducts hundreds of pesticide registration actions annually that require the application of efficient risk assessment tools to complete the actions in a timely and scientifically defensible manner. Although originally intended to be an efficient tool for implementing the Revised Method guidance (USEPA, 2020a), the MAGtool has failed to deliver on this intention. There are many technical issues (e.g., compounding conservatism, poor integration of lines of evidence, unrealistic exposure estimates, inability to integrate higher-tier data) that render the process ineffective. Ultimately, this places the burden squarely on the resourcechallenged Services to evaluate essentially all listed species evaluated in the BEs prepared by the USEPA. This inefficiency is a drain on the Services' resources that would otherwise be devoted to protecting habitat, establishing partnerships for species protection (e.g., Habitat Conservation Plans), evaluating candidate species for listing under the ESA, and other core functions. Thus, the adverse impact of a tool that generates unrealistic and unreasonable risk determinations to listed species programs in general is significant and must be addressed.
As the USEPA moves forward with their BEs, we believe that the recommended solutions to the previously identified issues provide a way forward to achieve an efficient, defensible, and reasonable approach to assigning BE effect determinations and likely J and/or likely AM calls to listed species.

ACKNOWLEDGMENT
The authors thank Syngenta Crop Protection LLC for financial support of this project. The authors thank Paul Whatling for sharing his experience and providing helpful comments on an early draft of the manuscript and Cynthia Cheney for formatting and editing assistance.

DISCLAIMER
The peer review for this article was managed by the Editorial Board without the involvement of D. Moore.