Qualitative analysis of the most locally relevant runoff and erosion parameters for constructing Brazilian scenarios

Estimating exposure is one of the most important steps in an environmental risk analysis of crop‐protection products to nontarget organisms. Regulatory agencies such as the US Environmental Protection Agency (USEPA), Pest Management Regulatory Agency (PMRA), and European Food Safety Authority (EFSA) all use mathematical exposure models in their regulatory assessment process. Brazil has been discussing the adoption of the Pesticide in Water Calculator (PWC) to be applied in aquatic pesticide risk assessment. Therefore, a qualitative sensitivity analysis (Morris OAT method) was performed to understand which are the most important local parameters in the PWC to estimate environmental concentrations in surface water (EECSW). In addition, an exercise made up of two corn scenarios in two Brazilian regions was developed (Uberlândia [UDI] and Arapoti [ARA]). Two herbicides with different soil‐binding properties and modes of action were selected to estimate the EECSW. The results demonstrated that the parameters of importance were different for each site, probably the result of different soil characteristics and meteorological patterns. This outcome suggests that regulatory agencies should consider developing more than one scenario to account for different agricultural regions. For Herbicide 1, the EECSW for UDI were similar to US scenarios, whereas for ARA they were lower. For Herbicide 2, the EECSW for the UDI site was higher than most of the US scenarios, whereas at the ARA site, EECs were similar to four US scenarios and lower than the other six. Local data were used as a refinement, resulting in the decrease in the EECSW for both herbicides in the UDI site. For the ARA site, Herbicide 1 displayed a similar EECSW value, whereas for Herbicide 2, it was lower after the refinement. Overall, these results demonstrated the importance of developing local scenarios to provide more realism to estimate pesticide exposure from its agricultural use and may help regulators to determine and recommend mitigations regarding the use of crop‐protection products. Integr Environ Assess Manag 2023;19:1374–1384. © 2023 The Authors. Integrated Environmental Assessment and Management published by Wiley Periodicals LLC on behalf of Society of Environmental Toxicology & Chemistry (SETAC).


INTRODUCTION
Mathematical tools can be used to predict the fate and transport and assess the exposure potentials of pesticides in the environment and under different environmental conditions (D'Andrea et al., 2020;Scorza Júnior & Da Silva, 2011). They offer an alternative to estimating pesticide concentrations in water and soil, especially where monitoring data are scarce and infrequent as in Latin America (Carriquiriborde et al., 2014). Validated regulatory models such as the Pesticide in Water Calculator (PWC), adopted by the US Environmental Protection Agency (USEPA), the Canadian Pest Management Regulatory Agency (PMRA), and the Chinese Institute for the Control of Agrochemicals, Ministry of Agriculture (ICAMA) (pesticide surface water exposure model [PSEM]; Geng et al., 2021) are used to estimate pesticide concentrations in water bodies.
Regulatory models evolved from simplistic screening to a more complex scenario-dependent tool. In the first case, screening models such as the Generic Estimated Environmental Concentration Model (GENEEC; Parker et al., 1995) were developed for screening purposes and helped regulators understand whether a given pesticide required further refinements (USEPA, 2022a). Screening models do not consider the influence of weather conditions, differences in soil topography, and product use patterns on different crops (USEPA, 2022a). On the other hand, a scenario-dependent model includes data such as crop growth stages, soil properties, weather patterns, field hydrology, and pesticide use patterns and fate (Sinnathamby et al., 2020;Young & Fry, 2019). In general, scenario-dependent models have been built for the regulatory agencies of the US (USEPA, 2022b) and the European Union (European Food Safety Authority [EFSA], 2013) in partnership with industry and academics, whereas for other countries, especially tropical regions, there is need for scientific knowledge development including potential adaptations especially regarding the development of local scenarios. The development of local scenarios is important to provide regulators with more locally relevant exposure estimates (Casallanovo et al., 2021;D'Andrea et al., 2020;De Oliveira Kaminski & Viera, 2021) after the generic screening tools are applied and additional mitigations are identified as needed.
The PWC is a model that simulates an agricultural field treated with a pesticide and the processes involved in its fate in the areas surrounding the field (Young & Fry, 2019). In fact, the PWC is composed of two other tools that were already available, the Pesticide Root Zone Model (PRZM) and Variable Volume Water Body Model (VVWM). The first model simulates the movement of the pesticide in the root zone of the plant, and the second simulates the transport and fate of chemicals in a water body (such as a pond or reservoir) receiving the PRZM-predicted potential off-target mass load of pesticide residues from the field (Young & Fry, 2019. Currently, the USEPA has 77 standard scenarios for the most relevant crops in the territory, considering soil, climate, and hydrological characteristics. The PWC allows simulations based on these already established scenarios and from data parameterized by the user considering a local condition (USEPA, 2022b). All these features allow this model to be used to identify the most influential drivers in developing local scenarios through a qualitative analysis. This analysis allows the user to determine the relative importance of each of the evaluated parameters, allowing prioritization of factors to be entered (Iooss & Lemaître, 2015).
Based on the considerations stated previously, the objective of this work was to perform a qualitative sensitivity analysis using the Morris method (Morris, 1991) to identify the most relevant runoff and erosion parameters in the calculation of estimated environmental concentrations in surface water (EEC SW ) from the PWC model. Furthermore, corn scenarios for two regions in Brazil were developed and compared with USEPA standard corn scenarios. Also, this study investigates the use of local parameters as a refinement in the calculation of EEC SW .

Crop-protection products
Two herbicides with different soil-binding properties and modes of action were selected, henceforth referred to as Herbicide 1 and Herbicide 2. Herbicide 1 had a high K d (strong binding to soil) and high water solubility, whereas Herbicide 2 had a low K d (weak binding to soil) and low water solubility. Additional details can be found in the Supporting Information, including the input parameters needed for the scenario parameterization of the PWC.

Site selection
Two local PWC scenarios in Brazil were developed (Figure 1), one located in Uberlândia (Minas Gerais, GPS coordinates −48.160; −18.897) and the second in Arapoti (Paraná,, henceforth referred to as UDI and ARA, respectively. Uberlândia is representative of the Brazilian Cerrado and ARA is representative of southern Brazil. Both regions are relevant to Brazilian agricultural production (Instituto Brasileiro de Geografia e Estatística [IBGE], 2022), especially for soybean, corn, and sugarcane. Arapoti is also representative of winter cereals. In addition, both sites had previously been used in Syngenta regulatory studies, which allowed the re-collection of soil and weather data collected in situ. Maps of sites can be found in the Supporting Information and were taken from the Prona-Solos, a repository of Brazilian soil data (PronaSolos, 2022).

Qualitative sensitivity analysis
The qualitative sensitivity analysis of the PWC parameters was performed in accordance with the Morris method (D' Andrea et al., 2020). This method was used to determine the type of effect that the parameters may have on the final output (Saltelli, 2004). They were classified in three types of effects: (1) negligible, (2) linear and additive, (3) nonlinear or related to interaction with other parameters.
Usually, the factors considered in a model follow a very asymmetric distribution of importance with few factors accounting for most of the output variability and most factors playing little to no role in the output result. Therefore, a definition of importance was necessary, as the ordering of factors by importance may be an issue of great significance when the model is used, especially in risk analysis or decision-making (Saltelli et al., 2019). Whenever a computer model is used to predict an important outcome, one of the many outputs produced must be selected by the given model and identified as the output of interest, which is the top-most information that the model is supposed to provide. In this case, the output of interest will be the EEC SW . However, the sensitivity analysis does not focus on the model output per se; it aims to indicate and rank the relative importance and contribution of each tested parameter that, once determined, leads to the greatest reduction in the variance of the output of interest. Hereafter, one can define the second most important factor until all the factors are ranked in order of importance. Thus, the concept of importance is more precise, linking it to a reduction in the variance of the target function (Saltelli, 2004).
The process of analysis proposed by Morris (1991) uses experiments in which one parameter is modified at each test and randomized (one-factor-at-a-time [OAT] method). The impact generated at the final output by changing each chosen parameter can be assessed at each iteration. Each evaluated parameter can assume a specific value that is randomly chosen within the range defined for this parameter. Two sensitivity measures are assessed by this method: (1) µ, which estimates the total effect of the parameter to the final output and can be obtained by computation of a variable number r of incremental values in the parameter space (e.g., x(1), …, x(r) for each parameter).
The number r can also be defined as the sample size.
(2) σ, which determines the group of second or high order effects in which the parameter is involved (including curvature effects and iterations; Saltelli, 2004).
The relative importance measure for each parameter (μ) is obtained by computing an average of incremental ratios (r) at different points, x(1), …, x(r) of the input space. The number r of selected points is known as the sample size of the experiment. A revised version μ* of the Morris μ was proposed by Campolongo et al. (2007), who demonstrated that this new measure μ* is more successful in ranking factors in order of importance and performs capably when the setting is selected for fixing factors.
In the revised version μ*, Campolongo et al. (2007) proposed that the distribution of the absolute values of the elementary effects should be considered. The examination of these distributions provides useful information about the influence of the several input factors on the output. In this approach, the most informative sensitivity measures of μ * were taken, which is defined as the mean of the distribution FIGURE 1 Herbicide 1-Comparative qualitative sensitivity analysis on Pesticide in Water Calculator (PWC) parameters, peak estimated environmental concentration (EEC). Data from two different sites: (A) full data set, (B) scale expansion on both axes. To facilitate the visualization, data were normalized by log transformation and σ as the standard deviation. Therefore, μ* was used to detect input factors with an important overall influence on the output, and σ was used to detect factors involved in interaction with other factors or whose effect is nonlinear. Morris (1991) defined that the two sensitivity measures also include the mean (μ) and the standard deviation (σ). However, according to Morris's original work, if the distribution has negative elements, which occurs when the model is nonmonotonic, some effects may cancel each other out. In this situation, the measure μ on its own is not reliable for ranking factors in order of importance. It was necessary to consider at the same time the values of μ and σ, as a factor with elementary effects of different signs (that could cancel each other out), which generate a low value of μ but a considerable value of σ to avoid underestimating the importance of these factors.
For sensitivity analysis, R Studio was used along with the sensitivity package (https://CRAN.R-project.org/package= sensitivity) to perform the calculations. A description of the code for each molecule (Herbicides 1 and 2) is available in the Supporting Information.

Development of local scenarios DataBase
Physical-chemical data. Data entered in the PWC chemical tab for both pesticides were taken from pesticide properties DataBase (PPDB; Lewis et al., 2016) and can be found in Supporting Information: Table S1. As for the number of applications and application timing, these parameters were taken from the product label and inserted in the application tab.
Parametrization of applications tab. A survey for all labels approved by the Brazilian Ministry of Agriculture, Livestock and Supply (MAPA) was performed through its online platform System of Phytosanitary Agrochemicals (AGROFIT) to define the rate and agricultural practices regarding the use of Herbicides 1 and 2 formulated products. The worst-case scenario of agricultural practice in Brazil for corn was selected and the highest practicable use rate identified was used in the modeling (AGROFIT, 2019).
For the application efficiency parameters (eff.) and drift, required by the modeling, standard values were considered according to each type of application as constant in the PRZM5 manual (Young & Fry, 2020). The summary of the information used in the application tab for Herbicides 1 and 2 are shown in Table 1.
Parametrization of land tab. The weather files for UDI and ARA were built according to the PRZM5 manual (Young & Fry, 2020). The meteorological data for these two regions (precipitation, average temperature, windspeed, and solar radiation) were obtained through the Syngenta weather station at the site and cover the period January 2017-December 2018 (see Supporting Information).
The local soil data used as a reference in obtaining the required parameter data were taken from local trials (UDI and ARA) from field studies sponsored by Syngenta and can be found in Table 2, as well as each respective GPS coordinate. According to the USDA soil texture classification, the top 0-10 cm soil horizon was classified as sandy clay (UDI) and sandy clay loam (ARA). The UDI site had a higher clay content (43%) than the ARA site (23%). As for the other soil horizons (10-20 and 20-30 cm), the texture classification profile was the same as in the topsoil for both sites. Regarding their taxonomy, both soils are classified as oxisols according to the USDA classification and as latosols according to the Brazilian Taxonomy Classification system (Revista Brasileira de Ciência do Solo [RBCS], 2022). Latosols are the most representative and predominant soil type in Brazil, comprising approximately 39% of the national soils and therefore are some of the most relevant soils in the agricultural landscape, being found in crop growing areas, such as for soybean, corn, and rice (PronaSolos, 2022). According to the PronaSolos database (2022), the soils in the experimental areas are predominantly red dystrophic latosols.
The weather data (windspeed, rainfall, atmospheric temperature) were sampled continuously for a period of two years by in-site weather stations. The parameters for hydrological factors and the soil layers used in the land tab are shown in Supporting Information:   Table 3. Although corn has two seasons in Brazil and can be planted at different times (Empresa Brasileira de Pesquisa Agropecuária [EMBRAPA], 2015), the crop cycle was defined as one per year and planted during the main season in each region. It coincides with the rainy season in south and southeast Brazil and therefore represents the worst-case scenario regarding the potential off-target transport of pesticides to water bodies. The planting period is between August and September in southern Brazil and between October and November in the southeast (EMBRAPA, 2015). Therefore, the planting was set as September for the ARA and October at the UDI. The corn variety under consideration had a season cycle of 120-130 days, where the flowering period occurs 65 days after planting and the maturity occurs 50-60 days after flowering (Magalhães & Durães, 2006). Therefore, the maturity day was set as 125 days after planting and the removal day as 5 days after maturity to complete a life cycle of 130 days. It is important to note that these factors are dynamic, and practices vary with time, although they were considered valid for this exercise.
Regarding the root depth, the length was limited to maximum allowed value by the PWC, which is 0.5 cm smaller than the soil column depth (Young & Fry, 2019). In this exercise, data for the soil profile were 30 cm deep; therefore, the maximum root depth was set at 29.5 cm. The remaining parameters, including canopy cover, canopy height, and canopy holdup, were not changed, so USEPA corn scenario data were used in the PWC.
Runoff tab. The parameters were maintained fixed throughout the modeling exercise according to the instructions in the PRZM5 manual (Young & Fry, 2020). They can also be found in Supporting Information: Table S3. The number of time-varying factors was set at 24. All the other parameters related to the Soil Universal Loss Equation (USLE) varied based on the sensitivity analysis (see Supporting Information).

Qualitative sensitivity analysis
The output of the qualitative sensitivity analysis according to the Morris method is shown in Figures 1 and 2 Table S4. The sensitivity of each parameter is given by μ values for which a number greater than 0.1 indicates that the estimated final EEC SW is sensitive to the evaluated input parameter. The Morris method also indicates secondary influences, that is, whether an input parameter may influence another one. This influence is given by the s value.
For Herbicide 1, in the UDI site (Figure 1), the results indicated that eight of the 10 evaluated parameters had an influence on the final output (1-d EEC SW ), displaying m values greater than 0.1. A different outcome was observed for the ARA site where only two parameters were shown to influence the final output. Parameters, such as the curve number (CN) and the application date regarding the number of days before the emergence of the crop, were revealed to have an influence on both sites (Figure 3). The CN is an empirical parameter developed by the USDA Natural Resources Conservation Service used in hydrology for predicting direct runoff or infiltration from rainfall excess (US Department of Agriculture [USDA], 1986). As for the influence on the pesticide application date, Boithias et al. (2014) also observed that the timing of application can affect the pesticide EEC SW , albeit in a differential manner because of the pesticide soil-binding properties. Lewan et al. (2009) concluded that precipitation up to five days before the application may influence the pesticide flow to water bodies of surface water, and therefore it may also explain why PWC input parameters affected by rainfall were observed to be sensitive. By comparing the meteorological data from the two sites (Supporting Information: Figure 1S), one can observe the significant rainfall events that occurred on the UDI site at the time of pesticide application (1 October) whereas at the ARA site, the last significant rainfall event occurred around 10 days before the set application date (1 September), after which the region endured a period of drought until April. Therefore, anticipating the application date on both sites might have increased the probability of runoff events and the final EEC SW .
The co-occurrence of significant rainfall events around the application time may partially explain why additional PWC input parameters at the UDI site, like the USLE-LS, USLE-C,   a According to PWC model setting, root depth must be 0.5 cm smaller than the soil horizon length, which in this case is 30 cm (see Table 2). and USLE-P, were all sensitive according to the qualitative analysis. The available rainfall data for the UDI site reveal that precipitation was concentrated between September and March, corresponding to the rainy season in the region. Between April and August, prolonged periods with no precipitation or an insignificant amount of rainfall were observed. Sajikumar and Remya (2015) have indicated that there is a reduction in base flows during dry periods due to the reduction in percolation. One of the evaluated PWC input parameters that is related to the flow from an area to water bodies is the Manning's roughness coefficient (N), which is used in the Manning's equation to calculate the flow rate from a given area and is used for erosion estimates (Suárez, 2005). Therefore, as Herbicide 1 has a high K OC (see Supporting Information), it was expected that the N parameter would contribute to the EEC SW , which was not observed in the results.
Regarding USLE-C and USLE-LS, which are related to soil loss, the modeling result indicated that they were sensitive, which is potentially the result of the concentration of the rainfall in a specific period of the year. Cassol et al. (2018) concluded that years with uniform rainfall distribution result in lower soil loss than years where rainfall is concentrated in a given period of the year. As for the USLE-P factor, Zhao et al. (2014) indicated that agricultural practices (e.g., bare soil, no-till, reduced tillage) may also influence the transfer processes to water bodies, affecting both soil erosion and pesticide runoff. Sajikumar and Remya (2015) noted that land cover influences the drainage basin and, consequently, the availability of surface water in the evaluated area.  The qualitative analysis also reveals that the organic matter content (OC%) is important for the final output in the UDI site. Herbicide 1 is highly soluble in water but is also tightly bound to soil particles and therefore classified as nonmobile. The variation in the organic matter content can affect its mobility and its partitioning between soil particles and water. Consequently, it can affect its propensity to runoff and soil erosion. The increase in the OC% may lead to a partitioning toward the solid phase and render the molecule less available to runoff. In addition, as indicated by Tiryaki and Temur (2010), soil moisture, timing, and amount of rainfall can also affect pesticide runoff. In addition, soil texture is also an important factor as it essentially affects the distribution of rainwater between infiltration and runoff. Taken together, the differences between the two sites regarding the soil texture and rainfall events may contribute to the distinct outcome of the analysis.
As for Herbicide 2, which has a low K OC and therefore is more mobile than Herbicide 1, a different pattern was observed, where the same input parameters were equally important for both sites: the CN, OC%, and the application date (Figures 2 and 3). Except for CN, for the UDI site, the qualitative sensitivity analysis indicated the great importance of the OC% and the application date (Figure 3), which may be the result of the difference in soil textures between both sites and the meteorological data. Molecules with a low K OC (<500 mL/g) are more mobile (Lewis et al., 2016) and therefore are more easily transported to water bodies (Gavrilescu, 2005). In addition, Boithias et al. (2014) have demonstrated that molecules with a low K OC are affected by the first rainfall event after the application. At the UDI site, as stated previously, applications coincided with significant rainfall events, which may explain higher m values (Figure 3), and likewise at the ARA site, anticipating the application date brought it closer to greater rainfall volumes. Regarding the OC content, Luo et al. (2011) demonstrated that the organic matter content may affect the mobility of the pesticides in the soil, with mobility decreasing with higher organic content. In the simulation, the OC% varied from 0.98% to 2.14%, which may explain its significance in the estimation of the EEC SW .
In addition to one-day concentrations, PWC also estimates exposures averaged more than 21-60 days, and typically these estimated EEC SW can be used for chronic risk assessments. Therefore, the importance of each input parameter with time was evaluated (Figure 3). For Herbicide 1, a significant difference between both sites was observed in the μ values, whereas for Herbicide 2, in general, it can be observed in Figure 3 that the contribution of each parameter was similar at both sites, except for the USLE-P and IREG, which for the UDI site displayed a larger contribution at 60 days.
Overall, the modeling output reveals the importance of using local parameters. The CN, the application timing, and the %OC have been shown as the most sensitive parameters Integr Environ Assess Manag 2023:1374-1384 © 2023 The Authors wileyonlinelibrary.com/journal/ieam FIGURE 3 Comparative qualitative sensitivity analysis (m) indicating the relative importance of Pesticide in Water Calculator (PWC) input parameters within time for each herbicide in Arapoti and Uberlândia to be considered on EEC SW calculations. Its relative importance differs depending on the physical-chemical characteristics of each pesticide and soil type of each site, corroborating the findings of D' Andrea et al. (2020).

Comparison between Brazilian and USEPA standard scenarios
In this work, typical application rates for corn in Brazil were evaluated for both herbicides. In the absence of local scenarios, regulators may resort to using foreign-developed scenarios with the local (national) rates to estimate the EEC in water bodies. For corn, the USEPA has 12 standard corn scenarios that are readily available. Typically, to estimate the EEC SW , one would use the recommendations from the Brazilian labels and select the regions more similarly related to Brazil. Nonetheless, it is important to assess whether there might be an under-as well as an overestimate of the EEC SW . De Oliveira Kaminski and Viera (2021) have evaluated this hypothesis by comparing two sugarcane standard scenarios (Louisiana and Florida) against the results of a local Brazilian scenario (Brotas, SP). The output was similar between the local Brazilian and the US Florida scenarios, whereas the EEC SW was four times higher in the Louisiana scenario than the Brazilian scenario, suggesting that care must be taken when using scenarios from other regions to avoid an overestimation of the exposure.
In an approach similar to that of De Oliveira Kaminski and Viera (2021), the estimated EECs were compared for Herbicides 1 and 2. In this study, all 12 corn scenarios were tested and compared with the results using local scenarios (UDI and ARA). The highest rate surveyed for corn in Brazilian labels was used, and the parameters which display a sensitivity index higher than 0.1 at peak EEC SW were considered important (Figure 3). The most critical value for each herbicide and for each region yielded four scenarios. Because USLE-P was sensitive in some scenarios, the highest slope value was also used associated with previous modeling exercises. As for the other input parameters, default PWC recommendations were used (Young & Fry, 2020). The Guidance of the Ontario Ministry of Agriculture was used to estimate the Universal Soil Loss Equation parameters. As stated previously in the Materials and Methods section, the runoff tab was maintained fixed, and local soil and weather parameters were used. The parameters inputted in the PWC are indicated in Supporting Information: Tables S3 and S4.
The output of this comparison is displayed in Table 4. The EEC SW for the ARA scenario was always smaller or slightly smaller than the pesticide concentrations estimated for the 12 USEPA corn scenarios, especially for Herbicide 1. As for the UDI scenario, pesticide concentrations were like those of the USEPA, with Herbicide 1 located at the lower range and Herbicide 2 at the upper range. This difference may be related to the different K OC values for both herbicides and, consequently, the propensity for runoff. The PWC indicates the transport mechanism for a given pesticide to surface water. According to the modeling output (Figure 4), runoff is the predominant mechanism for Herbicide 2 in both Brazilian sites, which is expected considering this molecule's high mobility in the soil. Soil erosion and spray drift were not relevant. For Herbicide 1, soil erosion is the predominant transport mechanism for the UDI scenario, and drift is the main mechanism for the ARA site. The PWC has also indicated that the fraction of the applied pesticide that goes to the water body is larger for the UDI site, which may also be linked to rainfall regime and the soil texture.
Brazilian scenarios-refinement of the EEC SW implementing local parameters into the PWC The influence of using local parameters as a refinement was investigated. A search for local data which make a significant contribution to the EEC SW calculation was performed (CN and USLE parameters). Curve numbers were taken from the Brazilian Water Agency (Agência Nacional de Águas [ANA], 2022), and the input data were inserted according to the PRZM Manual (see Supporting Information:  Tables S5 and S6). For the soil erodibility factor (factor K), data were extracted from an erodibility map for Brazil published by Godoi et al. (2021) according to the soil type. Regarding other USLE parameters, local data were extrapolated from neighboring areas. The output of PWC modeling with local data is shown in Table 5 and compared with the data from Table 4. A decrease was observed in EEC SW refined with local data for both herbicides.
For Herbicide 1, the use of local data resulted in a fivefold decrease in the EEC SW compared with the UDI site. For Herbicide 2, a 10-fold decrease in the EEC SW was observed at the UDI site. For the ARA site, a 1.05-and 6.2-fold decrease in EEC SW was observed for Herbicides 1 and 2, respectively.
The CN had a major influence on EEC SW calculation, especially on Herbicide 2, which has a low K OC . This finding agrees with other publications that also consider the use

CONCLUSION
The development of local scenarios, and parametrization of environmental fate, crop, and land parameters has proved to be a feasible task, allowing local regulators to identify, through a qualitative sensitivity analysis, which parameters are the most relevant to each relevant scenario. Ideally, depending on the country size and the crop area distribution, more than one local scenario should be developed. The EEC SW may differ depending on the physicalchemical characteristics of a pesticide, as demonstrated by Brodeur et al. (2022), who developed and modeled  a Exposure refined with local data for each region (e.g., CN and USLE parameters).

FIGURE 4
Relative importance (%) of each pesticide transport mechanism to water bodies and the fraction of the total applied pesticide that is estimated to reach the water body. (A) Herbicide 1 and (B) Herbicide 2 PWC-based scenarios and have used the pesticide's dissociation constant (K d ) to identify the most sensitive areas in the Pampa region in Argentina. The use of data modeling to estimate the exposure of environmental concentrations is part of the regulatory framework in countries with a robust regulatory system. In Latin America, aside from the modeling exercise from Brodeur et al. (2022) andD'Andrea et al. (2020) in Argentina, efforts to develop modeling tools or adapt existing ones are occurring in the Andean countries and Brazil. Ritter and Patiño (2021) gathered local data to develop scenarios for Colombia and Peru, whereas in Brazil, there is an initiative to create local scenarios for the PWC (DOU, 2019). Both initiatives should be supported because of the absence of local scenarios; regulators may have no choice but to use scenarios developed by other countries expecting that they may be able to accommodate local conditions. This study demonstrates that local label rates with foreign scenarios can result in overestimated environmental concentrations, affecting the outcome of a risk assessment for aquatic organisms. Using local data brings more realism to the risk assessment scheme and provides regulators more appropriate tools to evaluate environmental risks to nontarget organisms and mitigation actions in their own regions.