Optimizing model selection across global countries for managing pesticide emission and surface freshwater quality: a hierarchical screening approach

Abstract

Pesticides in surface freshwater primarily originate from their emissions in agricultural lands, potentially leading to violations of surface freshwater quality standards. To aid global regulatory agencies in effectively managing surface freshwater quality by estimating and controlling pesticide emission rates, this study proposes a hierarchical screening approach for countries and regions worldwide to select appropriate modeling tools. Hierarchical indicators are introduced to classify countries globally, considering their spatial distribution areas, pesticide emission conditions, and legislative systems. Consequently, different categories of countries are matched with suitable model groups, such as the standard model group for regulatory scenarios, the general model group for continental scenarios, and the advanced model group with high spatial resolution. Results indicated that a total of 193 countries worldwide were categorized into six country groups, of which 153, 34, and 6 countries were found to fit the standard, general, and advanced model groups, respectively, based on the model assignments for these country groups. Furthermore, 12 commonly used pesticides were selected to demonstrate the back-calculation process, which estimates the pesticide emission rate (input) by pesticide surface freshwater quality standards (output) by standard and general model groups. The Advanced model group was not applied in this process due to its intensive computation. An approximate approach was developed to simplify the calculation of the emission rate factor of pesticides using the PWC and TOXSWA selected in the standard model group as well as SWAT in the general model group, serving as a demonstration. This approach can be applied to control pesticide emission rates from surface freshwater quality standards across countries that fit in the standard and general model groups. The results highlight that pesticide fate models selected through the hierarchical screening approach, can assist global countries in establishing a quantitative relationship between pesticide emission rates and surface freshwater quality standards, which can help global agencies manage pesticide emissions and freshwater quality from a legal perspective. There is a need to update and simplify suitable advanced model for calculation demonstration in future studies to aid in pesticide management. Further research is needed to thoroughly investigate pesticide emissions and freshwater residue concentrations under varying conditions.

Introduction

Pesticide emissions pose a global issue as these substances, intentionally released into the environment, are designed to eradicate pests, insects, weeds, parasites, and other threats to our agricultural and economic crops [1, 2]. However, if not effectively managed and regulated, the residues of these pesticides can present significant risks to both human health and ecological systems [3,4,5,6]. Among the various risks associated with pesticide emissions, freshwater pollution stands out as a major concern [7,8,9].

To mitigate ecological and human health risks, countries worldwide have implemented pesticide regulations and established environmental quality standards for surface freshwater [10]. These regulations and standards serve as legal obligations for responsible parties, requiring them to engage in pre-pollution management, including emission control and adherence to regulations, as well as post-pollution remediation, such as water treatment. To ensure proactive water quality management, it is crucial to regulate and control pesticide emissions, thereby limiting the source of pollution. Prevention is always preferable to remediation when it comes to pollution, as it helps to avoid the need for costly damage control. The United States Environmental Protection Agency (USEPA) has developed Total Maximum Daily Loads (TMDLs), which can be derived using mass balance calculations or water quality models. TMDLs calculate and regulate the maximum amount of a pollutant entering waterbodies based on the water quality standards of pollutants [11]. The European Union (EU) requires member states to implement pollutant discharge permit limits, which integrate water quality management with pollutant emissions. These limits are calculated by existing technology or water quality standards inversion methods [12, 13]. From this perspective, developing models to estimate pesticide emission rates based on existing environmental quality standards proves to be a proactive and cost-effective approach, which can ensure that pesticide residue levels remain within safe thresholds, thereby preventing potential pollution events [14]. This not only reduces the need for regulatory responses concerning ecological and human toxicity but also helps in avoiding additional expenses associated with remediation and treatment.

However, accurately modeling pesticide concentrations in surface freshwater based on emissions is a complex task due to various influencing factors. These factors encompass a range of emission scenarios, meteorological conditions, geological features, chemical properties, water statuses, agricultural practices and more. To address these challenges, researchers have developed different modeling tools that are specifically designed for different scenarios and conditions. These modeling tools span from regulatory scenarios, which assume standardized pond conditions, to real scenarios that incorporate geographical features. For instance, USEPA has introduced tier-based exposure scenarios and developed multiple models such as Generic Estimated Environmental Concentration (GENEEC2), Pesticide in Water Calculator (PWC), and Surface Water Concentration Calculator (SWCC), specifically designed for regulatory scenarios with conservative assumptions. These models are widely utilized for risk assessment of aquatic life at vulnerable sites [15, 16]. The regulatory-scenario models require fewer input variables and provide conservative estimates of chemical concentrations, making them practical for regulatory purposes. European countries have developed the Pesticide Root Zone Model (PRZM), model of water flow and solute transport in microporous soil (MACRO), and Toxic substances in Surface Waters (TOXSWA) to predict pesticide residuals in surface water. In addition, these models were selected to develop and implement the Surface Water Scenario Help (SWASH) software tool, which is routinely employed to assess pesticide exposure in specified environmental scenarios [17].

To address real-world situations, researchers have developed various models for simulating pesticide water concentrations and estimating residue levels in surface water. For unsteady-state conditions, a fugacity model was introduced [18], which takes into account pesticide application on croplands and surface water runoff. At the catchment-scale watershed level, models like System for the Evaluation of Pesticide Transport to Waters (SEPTWA), Drainage Runoff Input of Pesticides in Surface Water (DRIPS), Soil and Water Assessment Tool (SWAT), and Agricultural Non-Point Source Model (AGNPS) have been utilized to estimate water residue levels from multiple emission sources of chemical pollutants, allowing for one-time or long-term modeling [19,20,21,22,23]. For rivers and streams, catchment-scale river models such as Quality Analysis Simulation Program No 2 Enhanced (QUAL2E), principal component analysis modeling, MIKE11, and River Water Quality Model No 1 (RWQM1) have been employed to describe the transport and fate of pesticides, ranging from static steady-state estimation to dynamic spatial–temporal simulation [23,24,25,26,27]. Moreover, the EU also developed the Forum for Co-ordination of pesticide fate models and their Use (FOCUS) model for assessing pesticide dynamics in river catchments [28]. As environmental pollution becomes a global concern, there is a need for models capable of estimating pesticide distributions and human exposures on a large geographic scale. Pangea, for example, integrates multiple computation tools and enables the modeling of chemical fate and transport at high spatial resolutions [29,30,31]. Furthermore, other modeling, prediction, and analytical methods (e.g., multimedia chemical fate model, chemical footprint, USEtox, PestLCI) have contributed to pesticide transport and fate assessment in surface water, facilitating pesticide emission management and aquatic risk assessment [32,33,34,35]. These models are extensively applied, calibrated, validated, modified, improved, and upgraded to address environmental challenges under various conditions and for diverse requirements [29, 36,37,38,39,40].

Many countries worldwide rely on international assessment rather than local conditions in the pesticide registration process [41]. Meanwhile, there is a lack of an appropriate model in the post-registration to estimate and control pesticide emissions preventing excessive pesticide residues in water. To the best of our knowledge, there are no studies available on the model selection for different countries. Given the vast diversity in pesticide application patterns, surface freshwater resources, spatial distributions, and regulatory management systems across nearly 200 countries and regions worldwide, it becomes apparent that a one-size-fits-all approach for model selection is inadequate. Some countries have extensive spatial coverage, requiring more sophisticated models beyond standardized conditions. Conversely, smaller regions with minimal pesticide emissions may not warrant the use of resource-intensive models with high spatial resolutions. Therefore, the objective of this study is to propose a hierarchical screening approach that assists global countries in selecting appropriate models and then controlling pesticide emissions by linking emission rates and surface quality standards. The specific objectives of this approach are to: (i) categorize global countries and their current modeling approaches based on pesticide-management-related features, (ii) match global countries with suitable models based on their specific characteristics, and (iii) calculate the acceptable pesticide emission rate based on existing global water quality standards.

Materials and methods

Hierarchical screening approach

To assist global countries in selecting appropriate fate models for the regulatory management of pesticide emission and surface freshwater quality, we proposed a hierarchical screening framework (Fig. 1). Theoretically, inputting pesticide emissions into the fate model can output water residue estimate (i.e., bottom-up modeling). Therefore, pesticide emissions can be back-calculated and managed using a selected model based on surface water standards (i.e., top-down management), hopefully achieving a balanced relationship between pesticide residues and standards (i.e., water residue estimates $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}$ surface water quality standards). In terms of model selection, this framework consists of three levels that are classified based on the combination of three indicators. Initially, countries worldwide are divided into two primary model pools (i.e., yellow and green) by the first indicator. Subsequently, they are subdivided into three specific model pools (red, purple, and the intersection of yellow and green) by two secondary indicators. In detail, we defined three specific groups of models: (i) the standard model group for the regulatory scenario; (ii) the general model group on a continental scale; and (iii) the advanced model group with high spatial resolutions. The indicators were selected hierarchically, taking into account the pesticide-management-related features of each country, including spatial distribution (${\text{S}}^{{1{\text{st}}}}$, first hierarchy), pesticide emission (${\text{E}}^{{2{\text{nd}}}}$, secondary hierarchy), and legislative system (${\text{L}}^{{2{\text{nd}}}}$, secondary hierarchy). The first hierarchy indicator, ${\text{S}}^{{1{\text{st}}}}$, determines whether a country exhibits significant variation in meteorological and hydrogeological conditions or not. If a country has a small spatial distribution area, it is assumed to have relatively low spatial heterogeneity, and the advanced model with high spatial resolutions may not be necessary. On the other hand, a country with a wide spatial distribution is considered to have high spatial heterogeneity, indicating that the standard model designed for a simplified and conservative scenario may not adequately capture the real geographical conditions. Therefore, the first hierarchy indicator, ${\text{S}}^{{1{\text{st}}}}$, is applied to initially divide the fate models into two groups:

$$ \left\{ {{\text{Standard}} \Leftrightarrow {\text{General}} \Leftrightarrow {\text{Advanced}}} \right\}{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}}{\text{S}}^{{1{\text{st}}}} = \left\{ {\begin{array}{*{20}c} {\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{S}}}^{{1{\text{st}}}} = \left\{ {{\text{Standard}} \Leftrightarrow {\text{General}}} \right\}} \\ {\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{L}}}^{{1{\text{st}}}} = \left\{ {{\text{General}} \Leftrightarrow {\text{Advanced}}} \right\}} \\ \end{array} } \right. $$

(1)

where $\left\{ {{\text{Standard}} \Leftrightarrow {\text{General}} \Leftrightarrow {\text{Advanced}}} \right\}$ represents a model pool consisting of three groups of models that are equally interchangeable (⇔), meaning that all model categories within this set can be applied by a country without any difference. The symbol "⇲" signifies the hierarchical operation, and “$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}^{{1{\text{st}}}}$" indicates the hierarchical operation with the first hierarchy indicator, ${\text{S}}^{{1{\text{st}}}}$. It is important to note that "⇲" does not follow the commutative law due to its hierarchical operation order, i.e., $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}^{{2{\text{nd}}}} \ne \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}^{{2{\text{nd}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}^{{1{\text{st}}}}$. Since ${\text{S}}^{{1{\text{st}}}}$ comprises two sub-indicators, namely small area (${\text{S}}_{{\text{S}}}^{{1{\text{st}}}}$) and large area (${\text{S}}_{{\text{L}}}^{{1{\text{st}}}}$), the hierarchical operation by "$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned} {\text{S}}^{{1{\text{st}}}}$" results in two outcomes: $\left\{ {{\text{Standard}} \Leftrightarrow {\text{General}}} \right\}$ and $\left\{ {{\text{General}} \Leftrightarrow {\text{Advanced}}} \right\}$. This implies that a country with a smaller area can choose between two equally interchangeable model candidates (i.e., a simple model or a general model), while a country with a larger area will opt for either a general or advanced model.

Fig. 1

Hierarchical screening structure for model selection to connect pesticide emissions with water residue estimates and surface freshwater water quality standards for countries worldwide. (Text and arrow next to the triangle illustrate the relationship of “Pesticide emissions”, “Water residue estimates” and “Surface Water quality”)

After undergoing hierarchical operations based on the first hierarchy indicator, Eq. (1) can be further hierarchically operated using the secondary hierarchy indicators ${\text{E}}^{{2{\text{nd}}}}$ (i.e., light emissions ${\text{E}}_{{\text{L}}}^{{2{\text{nd}}}}$ and heavy emissions ${\text{E}}_{{\text{H}}}^{{2{\text{nd}}}}$) or ${\text{L}}^{{2{\text{nd}}}}$ (i.e., unitary constitution ${\text{L}}_{{\text{U}}}^{{2{\text{nd}}}}$ and federal constitution ${\text{L}}_{{\text{F}}}^{{2{\text{nd}}}}$), resulting in Eqs. (2) and (3) respectively:

$$ \left\{ {{\text{Simple}} \Leftrightarrow {\text{General}}} \right\} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned} {\text{E}}^{2nd} = \left\{ {\begin{array}{*{20}c} { \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned} E_{L}^{2nd} = \left\{ {Standard} \right\}} \\ {\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}E_{H}^{2nd} = \left\{ {General} \right\}} \\ \end{array} } \right. $$

(2)

$$ \left\{ {{\text{General}} \Leftrightarrow {\text{Advanced}}} \right\}\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned} {\text{L}}^{{2{\text{nd}}}} = \left\{ {\begin{array}{*{20}c} {\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}L_{U}^{{2{\text{nd}}}} = \left\{ {{\text{General}}} \right\}} \\ {\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}L_{F}^{{2{\text{nd}}}} = \left\{ {{\text{Advanced}}} \right\}} \\ \end{array} } \right. $$

(3)

where we determine that the secondary emission indicator ${\text{E}}^{{2{\text{nd}}}}$ should be operated after the hierarchical operation "$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{S}}}^{{1{\text{st}}}}$". This decision is based on the understanding that a country with a larger area (i.e., the hierarchical operation of “${\text{S}}_{{\text{L}}}^{{1{\text{st}}}}$”) requires more comprehensive background information than a standard model can provide. In the case where a small-area country (i.e., "$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{S}}}^{{1{\text{st}}}}$") has light emissions of pesticides (i.e., “$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}_{{\text{L}}}^{{2{\text{nd}}}}$"), the outcome is the generation of a standard model, as depicted in Eq. (2). This indicates that the pesticide emissions in this country do not pose significant concerns to water quality from a regulatory perspective. Therefore, employing a standard model to predict the upper level of pesticide concentrations would not lead to significant overestimation, simplifying the regulatory management process. However, if a small-area country intensively uses pesticides, resulting in a significant increase in pesticide concentrations in surface water, the standard model may overestimate the actual concentration (e.g., monitoring data) due to the inputs used. Consequently, the back calculation of the acceptable pesticide emission rate based on the standard model would be excessively conservative. To address this, we use a general model (i.e., "$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{S}}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}_{{\text{H}}}^{{2{\text{nd}}}}$") to estimate pesticide concentrations for small-area countries with intensive pesticide use on a continental scale. The quantification of light or heavy emissions is further discussed in the next section.

On the other hand, when considering a large-area country (i.e., "$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{L}}}^{{1{\text{st}}}}$"), it becomes apparent that a standard model without spatial resolutions may not be suitable. In this case, we need to select a model candidate from the $\left\{ {{\text{General}} \Leftrightarrow {\text{Advanced}}} \right\}$ pool. To make this selection, we also take into account a country's legislative system as a secondary indicator, as it significantly influences the regulatory management process. If a large-area country operates under a federal system, where states have the authority to develop their own regulations, an advanced model with high spatial resolutions (i.e., "$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{L}}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{L}}_{{\text{F}}}^{{2{\text{nd}}}}$") is necessary. This enables the country to facilitate its states in developing regional management strategies tailored to their specific needs. On the other hand, if a large-area country has a unitary constitution, a general model would be more helpful in estimating pesticide concentrations on a continental scale. Typically, regional governments in such countries adopt national standards or develop their authority based on national regulations. Therefore, a general model is sufficient for estimating pesticide concentrations in this context. Table 1 summarizes the hierarchical screening process for model selections, taking into account spatial distribution and legislative systems.

Table 1 Hierarchical screening model selections by first and secondary hierarchies

Hierarchical indicators quantification

Spatial distribution area

We used the first hierarchy indicator “${\text{S}}^{{1{\text{st}}}}$” to categorize countries based on their spatial distribution areas, which we assume will positively correlate with spatial heterogeneity, including factors such as climates and hydrogeological conditions. To quantify the selection of ${\text{S}}^{{1{\text{st}}}}$, we employed the country's area as a representation of its spatial distribution area, as it generally correlates positively with spatial heterogeneity. This can further be related to the latitudinal temperature gradients observed across a country. It is known that latitudinal temperature changes by approximately 1 °C for every 150 km ($\frac{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{L}}_{{\text{N}}} }}{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{T}}}}$) in the Northern Hemisphere and 197 km ($\frac{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{L}}_{{\text{S}}} }}{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{T}}}}$) in the Southern Hemisphere [42]. Taking into account a temperature gradient of 10 °C across a country latitudinally, we identify this as a critical point for determining whether the spatial distribution is wide or not. To quantify the selection of ${\text{S}}^{{1{\text{st}}}}$, we use a simple square estimation for the critical area as follows:

$$ {\text{CV}}\left( {S^{{1{\text{st}}}} } \right) = \frac{A}{{\left( {10^{o} C \times \frac{1}{2} \times \left( {\frac{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}L_{N} }}{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}T} + \frac{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}L_{S} }}{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}T}} \right)} \right)^{2} }} \Rightarrow \left\{ {\begin{array}{*{20}c} { {\text{CV}}\left( {S^{{1{\text{st}}}} } \right) < 1;\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned} S_{S}^{{1{\text{st}}}} } \\ {{\text{CV}}\left( {S^{{1{\text{st}}}} } \right) \ge 1;\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}S_{L}^{{1{\text{st}}}} } \\ \end{array} } \right. $$

(4)

where ${\text{CV}}\left( {{\text{S}}^{{1{\text{st}}}} } \right)$ is the critical value for selecting ${\text{S}}^{{1{\text{st}}}}$. A (km²) is the area of a country. $\frac{1}{2}\left( {\frac{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{L}}_{{\text{N}}} }}{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{T}}}} + \frac{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{L}}_{{\text{S}}} }}{{\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{T}}}}} \right)$ denotes the average length latitudinally per temperature gradient for the Northern and Southern Hemispheres. Then, if the estimated value of ${\text{CV}}\left( {{\text{S}}^{{1{\text{st}}}} } \right)$ is greater than or equal to 1, indicating the area of a country larger than or equal to 3,010,225 km² (301,022.5 kha), a country is assumed to have a large spatial distribution.

Pesticide emission conditions

For countries with small spatial distributions (i.e., $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{S}}}^{{1{\text{st}}}}$), we used the secondary indicator “${\text{E}}^{{2{\text{nd}}}}$” to determine a suitable model from the standard and general model groups. The purpose of using ${\text{E}}^{{2{\text{nd}}}}$ is to quantify whether the pesticide emission condition raises significant concerns from a regulatory perspective. If the emission condition is a cause for concern, employing a standard model could introduce additional uncertainty in estimating water residue levels. This is because the scenarios in models like the PWC assume immobile and inland waters, such as farm ponds and reservoirs [40]. However, if the pesticide emission condition does not pose significant concerns, the overestimation of water residue levels by a standard model would not be problematic due to the variance on a small scale. To assess whether the pesticide emission condition is a major concern, we compared the estimated pesticide concentrations in surface waters to the national standard values of each country under the standard assumption. If the estimated concentration is lower than the environmental quality standard for a pesticide, it indicates that the pesticide emission condition in the country is not a significant concern. To quantify this assessment, we define ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)$ as the critical value for ${\text{E}}^{{2{\text{nd}}}}$, which represents the ratio of estimated concentrations by a standard model to the water quality standard:

$$ {\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\text{i}}} = \frac{{{\text{f}}_{{{\text{STD}}}} \left( {{\text{ER}}_{{\text{i}}} } \right)}}{{{\text{RS}}_{{\text{i}}} }} $$

(5)

where ${\text{ER}}_{{\text{i}}}$ (kg ha⁻¹ d⁻¹) is the emission rate of pesticide i, and ${\text{f}}_{{{\text{STD}}}} \left( \cdot \right)$ denotes the concentration simulation function by a standard model. ${\text{RS}}_{{\text{i}}}$ (mg L⁻¹) is the environmental quality standard of pesticide i in surface freshwaters promulgated by national authorities to protect the human or ecological health. If the simulated ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)$ for pesticide i is higher than 1.0, then we consider the emission as a great concern and suggest a general model for the pesticide i, i.e.,$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{S}}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}_{{\text{H}}}^{{2{\text{nd}}}} = \left\{ {{\text{General}}} \right\}$. Otherwise, a light emission is assumed, and the country can apply a standard model, i.e., $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{S}}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}_{{\text{L}}}^{{2{\text{nd}}}} = \left\{ {{\text{General}}} \right\}$.

It is important to note that the critical value for ${\text{E}}^{{2{\text{nd}}}}$ is derived based on current regulations, which primarily focus on legal obligations related to pesticide applications and management, rather than population health risks. For instance, the EU has stipulated Directive No 2009/128/EC and Regulation 2021/2115/EC to regulate the sustainable use of pesticides and achieve the pesticide reduction target [43]. Environmental quality standards act as a legal framework for pesticide practices, while human and ecological risk assessments serve as technical support for establishing these standards. Although risk assessments play a crucial role in protecting human and ecological health, they mostly do not directly restrict or impact pesticide emissions due to the lack of legal enforcement. In cases where a country has multiple sets of surface freshwater quality standards, we consider the minimum value among these standards as the guideline. If a country does not define any water quality standards for a specific pesticide, we set ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\text{i}}}$ to zero. This indicates a "zero" regulatory concern for using that particular pesticide, implying that there are no limitations in the water quality standard unless restricted by physical possibilities, such as a legally ceiling value. This phenomenon of regulatory unrestricted use can also be observed in other fields. For instance, some freeways in Germany have no speed limits [44], but drivers are still bound by other safety-related traffic rules. Similarly, in the context of pesticide emissions, the absence of specific water quality standards implies an assumption of "no limitations," unless restricted by physical constraints. The definition of ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)$ for the emission of all pesticides in a country can be summarized as follows:

$$ {\text{CV}}_{{{\text{RS}}}} \left( {E^{{2{\text{nd}}}} } \right) = {\text{Max}}\left\{ {{\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{i} } \right\} \Rightarrow \left\{ {\begin{array}{*{20}c} { {\text{CV}}_{{{\text{RS}}}} \left( {E^{{2{\text{nd}}}} } \right) \le 1;\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}E_{42L}^{{2{\text{nd}}}} } \\ { {\text{CV}}_{{{\text{RS}}}} \left( {E^{{2{\text{nd}}}} } \right) > 1;\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}E_{H}^{{2{\text{nd}}}} } \\ \end{array} } \right. $$

(6)

where ${\text{Max}}\left\{ \cdot \right\}$ denotes as the maximum value function. If ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right) > 1$, indicating that at least the emission of one pesticide is considered as a great regulatory concern (i.e., heavy emissions), then the model pool for the country should be operated by “$\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}_{{\text{H}}}^{{2{\text{nd}}}}$” to select a general model. If the maximum value is less or equal than 1 (${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right) \le 1$), it indicates that no emissions of pesticides will cause the concern from the regulatory perspective. In this study, we focused on 12 currently used pesticides that are widely and commonly employed by countries across the globe. The selection of these pesticides was based on a comprehensive statistical analysis of their usage, production, and trade information [45], which included 2,4-D (CAS No. 94-75-7), aldicarb (CAS No. 116-06-3), atrazine (CAS No. 1912-24-9), chlorpyriphos (CAS No. 2921-88-2), diazinon CAS No. 333-41-5), dicamba (CAS No. 1918-00-9), diuron (CAS No. 330-54-1), glyphosate (CAS No. 1071-83-6), malathion (CAS No. 121-75-5), MCPA (CAS No.94-74-6), metolachlor (CAS No. 51218-45-2), and trifluralin (CAS No. 1582-09-8). It should be noted that the use of aldicarb in the US was banned by the USEPA in 2010 [46], and the new product AgLogic 15GG, of which the essential formulation is the same with aldicarb, was approved to register and use conditionally by the USEPA in 2016 [47]. The use of atrazine was banned by the EU in 2004 but it remains widely used in other countries such as Australia and the US [48]. Persistent organic pollutants (POPs) and legacy pesticides were not considered because of a global ban and severe restrictions [49]. Therefore, the ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)$ in Eq. (6) is the result of the maximum simulated value from 12 current widely and commonly used pesticides, i.e.,${\text{Max}}\left\{ {{\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{1} , \ldots ,{\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{12} } \right\}$.

Fast-filter approximation approach

To determine which secondary sub-indicators of ${\text{E}}^{{2{\text{nd}}}}$ to use, we proposed a fast-filter approach using the PWC model in a standard scenario to showcase the modeling process (Detailed in the Supplementary file). The PWC model is flexible and countries can select or create their own standard scenario when adopting this approach. We selected corn to represent the pesticide reception crop, as it is one of the most commonly grown agricultural commodities worldwide. In general, this method provides a conservative estimation of pesticide concentration in surface water compared to the water quality standard. Since emission rates of individual pesticides are often unavailable for many countries, we employ the emission rate for a pesticide group ($\sum\nolimits_{{i\, = \,1}}^{m} {{\text{ER}}_{i} }$) based on the pesticide's classification (Supplementary database), which generates a more conservative concentration value. If this conservative value exceeds the water quality standard, then $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}_{{\text{H}}}^{{2{\text{nd}}}}$ must be implemented. This fast-filter approximation approach can be summarized as follows:

$$ {\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{i \in {\varvec{J}}_{{\varvec{i}}} }} = \frac{{f_{{{\text{STD}}}} \left( {\mathop \sum \nolimits_{i = 1}^{{\text{m}}} {\text{ER}}_{i} } \right)}}{{{\text{RS}}_{i} }} \ge \frac{{f_{{{\text{STD}}}} \left( {{\text{ER}}_{i} } \right)}}{{{\text{RS}}_{i} }} = {\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{i} $$

(7)

where $\sum\nolimits_{{i\, = 1}}^{m} {{\text{ER}}_{i} }$ denotes as the total emission rate of the group J_i of m pesticides that have the same classification with pesticide i. Since commonly used pesticides make up a significant portion of the total emission rates for their respective pesticide groups, it is likely that the emission rates for these individual pesticides are quite substantial. Consequently, if a country lacks specific emission information for individual pesticides, Eq. (6) can be modified as follows:

$$ {\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\varvec{J}}} = {\text{Max}}\left\{ {{\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{1 \in {\varvec{J}}_{1} }} , \ldots ,CV_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{12 \in {\varvec{J}}_{12} }} } \right\} $$

(8)

where ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ denotes that the critical value for ${\text{E}}^{{2{\text{nd}}}}$ is derived from the emission rate of a group of pesticides due to the lack of specific information for individual pesticides. If a country defines the pesticide surface freshwater quality standard in both fixed (e.g., maximum values) and morphing forms (e.g., annual averages) [10], we use the PWC model to generate peak values and 365-day averages to simulate ${\text{CV}}_{{{\text{RS}}\_{\text{Max}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ and ${\text{CV}}_{{{\text{RS}}\_{\text{Avg}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ respectively. If both ${\text{CV}}_{{{\text{RS}}\_{\text{Max}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ and ${\text{CV}}_{{{\text{RS}}\_{\text{Avg}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ values are less than or equal to 1.0, we considered the light emission situation for this specific country.

Legislative system

For countries with extensive spatial distributions (i.e., $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{L}}}^{{1{\text{st}}}}$), we used the secondary indicator ${\text{L}}^{{2{\text{nd}}}}$ to determine whether a general or advanced model should be employed. ${\text{L}}^{{2{\text{nd}}}}$ is directly linked to a country's legislative system. In the case of a country with a federal constitutional system, such as the United States, where individual states have the authority to establish their own environmental quality standards, we recommend the use of an advanced model with high spatial resolutions to enhance regional pesticide management. On the other hand, for large-sized countries like China, which follows a unitary political system where provinces typically adhere to national standards [50], a general model is more suitable for estimating water residue levels on a continental scale. It is important to note that a country's legislative system can be complex, going beyond a simple categorization of "unitary" or "federal" constitution due to historical, political, and economic factors. Some unitary governmental countries have developed multilevel legislative systems, while some federal countries have adopted unified regulations. In addition, certain countries operate under quasi-federal systems, where they are officially federal states but function as unitary governments. In our classification based on ${\text{L}}^{{2{\text{nd}}}}$, we consider countries in terms of their legislative types. However, regulatory agencies can adjust the model selection process according to their unique circumstances. The critical value for ${\text{L}}^{{2{\text{nd}}}}$ can be defined as follows:

$${\text{CV}}\left( {{\text{L}}^{{2{\text{nd}}}} } \right) = \left\{ {\begin{array}{*{20}c} { 1;{\text{if}}\, {\text{unitary system}}} \\ {0;{\text{if}}\, {\text{federal system}}} \\ \end{array} } \right. \Rightarrow \left\{ {\begin{array}{*{20}c} {1;\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}L_{{\text{U}}}^{{2{\text{nd}}}} } \\ {0;\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}L_{{\text{F}}}^{{2{\text{nd}}}} } \\ \end{array} } \right. $$

(9)

where ${\text{CV}}\left( {{\text{L}}^{{2{\text{nd}}}} } \right)$ is defined to 1 and 0 for the unitary and federal constitutional system, respectively. We determine that the secondary indicator ${\text{L}}^{{2{\text{nd}}}}$ should only be used for large-area countries, specifically those falling under the category of $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{L}}}^{{1{\text{st}}}}$. This decision is based on the fact that in a country with low spatial heterogeneity, it is more practical to estimate nationwide water residue using a simple or general model, even if individual states have their own environmental quality standards.

Model groups and calculation demonstration

To accommodate the diverse geological, agricultural, and regulatory conditions encountered in various countries and regions, we proposed a model pool, which encompasses different model groups, ranging from standardized-condition models operating at a farmland scale to real-geo-referenced models operating at a global scale. To illustrate the calculation procedure, we employed the standard and general model group (i.e., the PWC and SWAT model) for the modeling exercise. Notably, considering that the TOXSWA model is widely applied in the European countries, we also included it in the standard model group and conduct the calculation demonstration.

The PWC model [51] was selected as a representation of the standard model group. This model was specifically developed to predict estimated environmental concentrations (EECs) in aquatic environments. The PWC model offers a convenient approach by employing a standardized scenario and limiting the number of input variables, thereby simplifying its application. Within the PWC model, we selected a regulatory scenario based on the USEPA's Tier II exposure. This scenario allows for a certain level of uncertainty and variability, with more input data compared to the Tier I scenario where pesticides are directly released into water bodies [36, 52]. The pond scenario of the PWC model assumes that pesticides, carried by runoff water from a 10-ha farmland due to heavy rainfall, are transported into a static water pond with a depth of 2.00 m and a volume of 20,000 m³. It is worth noting that the outputs of the PWC model typically yield higher concentrations than the upper levels of measured concentrations. This is because the model regulatory inputs incorporate conservative assumptions to safeguard regulatory scenarios for vulnerable sites [53]. Many studies have employed the PWC model in conjunction with earlier USEPA models to estimate conservative aquatic exposure concentrations of chemicals [36, 54, 55]. Therefore, the PWC model proves particularly useful for countries or regions with a limited spatial distribution and low pesticide emission (i.e., $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{S}}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}_{{\text{L}}}^{{2{\text{nd}}}}$). In addition, we applied the PWC model as a fast-filter tool to assess whether pesticide emissions in a country raise concerns from a regulatory perspective. Due to the conservative assumptions of model inputs, a simulated concentration of a pesticide in a conservative scenario lower than the corresponding water quality standard indicates a condition of light emission.

If a region has a relatively wide spatial distribution or under a heavy pesticide emission condition, use of the standard model that is designed for a regulatory scenario will lower the spatial resolution or too much overestimate the water pesticide concentrations. Then, the general model group can be used to specify the outputs for different geographic regions, which is known for its ability to incorporate multiple levels of spatial resolution and can be applied on a relatively large scale. For instance, the SWAT model is capable of estimating chemical concentrations across different scales, ranging from small watersheds to entire river basins [56]. SWAT is extensively used to assess the environmental impact caused by land use and climate change. As a result, this model group is well-suited for larger countries where a general estimate is sufficient for proposing national regulations (i.e., $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{L}}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{L}}_{{\text{U}}}^{{2{\text{nd}}}}$), as well as countries with small areas but high pesticide emissions (i.e., $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{L}}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{E}}_{{\text{H}}}^{{2{\text{nd}}}}$).

If a region exhibits a significant level of spatial heterogeneity and requires multilevel regulation and management of pesticide emissions, it is necessary to implement an advanced model group with high spatial resolutions. It is noteworthy that current models with high spatial resolutions require intensive computation and are continuously under development, and therefore are unsuitable for regulatory assessment in this study. Meanwhile, the fate models used for pesticide regulatory around the world tend to be more simplified and user-friendly [57]. Therefore, the advanced model group now lacks of specific suitable model to manage pesticide emissions (i.e., $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{S}}_{{\text{L}}}^{{1{\text{st}}}} \begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}{\text{L}}_{{\text{F}}}^{{2{\text{nd}}}}$).

Emission rate factor to calculate pesticide emission rates

To determine the pesticide emission rate necessary to maintain theoretical water pesticide concentrations below the water quality standard, we employed a fitted-inverse-function approach using data collected from the model simulation, which can be expressed as follows:

$$ ER_{i} = a_{i,0} + a_{i,1} \left[ {f_{MODEL} \left( {ER_{i} |{\varvec{V}}_{i} ,{\varvec{M}}} \right)} \right] + \ldots + a_{i,n} \left[ {f_{MODEL} \left( {ER_{i} |{\varvec{V}}_{i} ,{\varvec{M}},{\varvec{A}}} \right)} \right]^{n - 1} $$

(10)

where ${\text{f}}_{{{\text{MODEL}}}} \left( {{\text{ER}}_{{\text{i}}} |{\mathbf{V}}_{{\text{i}}} ,{\mathbf{M}},{\mathbf{A}}} \right)$ is the simulated concentration of pesticide i by the selected model according to its emission rate ${\text{ER}}_{{\text{i}}}$, given the physicochemical variables for pesticide i (in the set ${\mathbf{V}}_{{\text{i}}}$), meteorological condition variables (in the set ${\mathbf{M}}$ ), and application pattern variables (in the set ${\mathbf{A}}$).${\text{a}}_{{{\text{i}},0}}$,…,${\text{a}}_{{{\text{i}},{\text{n}}}}$ are the equation fit coefficients for pesticide i. Based on the various ER_i inputs and the simulated ${\text{f}}_{{{\text{MODEL}}}} \left( {{\text{ER}}_{{\text{i}}} |{\mathbf{V}}_{{\text{i}}} ,{\mathbf{M}},{\mathbf{A}}} \right)$ outputs, we fit the polynomial equation in Eq. (10). We then input the environmental quality standard for pesticide i to replace ${\text{f}}_{{{\text{MODEL}}}} \left( {{\text{ER}}_{{\text{i}}} |{\mathbf{V}}_{{\text{i}}} ,{\mathbf{M}},{\mathbf{A}}} \right)$ to calculate the reference emission rate ${\text{ER}}_{{{\text{i}},{\text{S}}}}^{0}$ (kg ha⁻¹ year⁻¹) using the standard model for pesticide i that does not cause any regulatory violations in surface freshwater. This approach can be performed in the PWC, TOXSWA and SWAT models. The ${\text{ER}}_{{\text{i}}}^{0}$ can also be estimated by the inverse modeling method [58] that continues to change the input values in the model until the simulated output values are very close to the environmental quality standards. It is worth noting that in the both PWC and TOXSWA model, Eq. (10) can be applied to derive the ${\text{ER}}^{0}$ values to avoid exceeding the maximum and annual average standards, because the these models generate both peak and 365-day average values, wherein the $f_{MODEL} \left( {ER_{i,MODEL} |{\varvec{V}}_{i} ,{\varvec{M}},{\varvec{A}}} \right)$ has a linear relationship with ${\text{ER}}_{{{\text{i}},{\text{model}}}}$ passing through the origin. Thus, simulating the peak and annual average pesticide concentrations in surface freshwater using the two standard models can be expressed as follows:

$$ ER_{i,PWC} = \left\{ {\begin{array}{*{20}c} {ERF_{i,PWC}^{Peak} f_{PWC}^{Peak} \left( {ER_{i,PWC} |{\varvec{V}}_{i} ,{\varvec{M}},{\varvec{A}}} \right)} \\ {ERF_{i,PWC}^{Ave} f_{PWC}^{Ave} \left( {ER_{i,PWC} |{\varvec{V}}_{i} ,{\varvec{M}},{\varvec{A}}} \right)} \\ \end{array} } \right. $$

(11-1)

$$ ER_{i,TOX} = \left\{ {\begin{array}{*{20}c} {ERF_{i,TOX}^{Peak} f_{TOX}^{Peak} \left( {ER_{i,TOX} |{\varvec{V}}_{i} ,{\varvec{M}},{\varvec{A}}} \right)} \\ {ERF_{i,TOX}^{Ave} f_{TOX}^{Ave} \left( {ER_{i,TOX} |{\varvec{V}}_{i} ,{\varvec{M}},{\varvec{A}}} \right)} \\ \end{array} } \right. $$

(12-2)

where ${\text{ERF}}_{{{\text{i}},{\text{PWC}}}}^{{{\text{Peak}}}}$ (kg ha⁻¹ year⁻¹ per mg L⁻¹) and ${\text{ERF}}_{{{\text{i}},{\text{PWC}}}}^{{{\text{Ave}}}}$ (kg ha⁻¹ year⁻¹ per mg L⁻¹) are the emission rate factors of pesticide i using the PWC model for the estimation of the peak and annual average concentrations, respectively. ${\text{ERF}}_{{{\text{i}},{\text{TOX}}}}^{{{\text{Peak}}}}$ and ${\text{ERF}}_{{{\text{i}},{\text{TOX}}}}^{{{\text{Ave}}}}$ means the emission rate factors of pesticide i using the TOXSWA model for the estimation of the peak and annual average concentrations, respectively.

Results and discussion

Matching global countries with suitable model groups

Firstly, we used the indicator “${\text{S}}^{{1{\text{st}}}}$” to categorize countries worldwide based on their spatial distributions. The country area data were sourced from the FAO database [59]. Out of the total, seven countries exhibited a ${\text{CV}}\left( {{\text{S}}^{{1{\text{st}}}} } \right)$ value greater than 1.0. For these countries, we then employed the first hierarchical indicator “${\text{S}}_{{\text{L}}}^{{1{\text{st}}}}$” to exclude them as candidates for the standard model group. Next, we employed the secondary hierarchical indicator “${\text{L}}_{{\text{F}}}^{{2{\text{nd}}}}$” to assess large-area countries with federal constitution systems, namely Russia, Canada, the US, Brazil, Australia, and India. This allowed us to propose the advanced model group with high spatial resolutions specifically tailored for these countries to model water pesticide concentrations in different regions within these countries. The advanced model assists in implementing state- or provincial-level pesticide management strategies. Considering China's unitary political structure and the widespread adoption of national environmental quality standards by local governments, we applied the indicator “${\text{L}}_{{\text{U}}}^{{2{\text{nd}}}}$” to select the general model candidate for China. For this purpose, we recommend SWAT, a general model capable of predicting chemical concentrations in environments on a continental scale, which aligns well with China's coast-to-coast application of national standards. After applying the indicator “${\text{S}}_{{\text{S}}}^{{1{\text{st}}}}$” to the small-area countries, we employed the PWC model as a rapid filtering tool to determine whether pesticide emissions raised significant concerns from a regulatory perspective. To calculate ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ for each small-area country, we utilized pesticide surface freshwater quality standards [10]. Out of the total, 49 small-area countries had regulations in place for pesticides in surface freshwater. Consequently, for the remaining small-area countries without such regulations, the ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ values were set to zero as per Eqs. (6) and (8). This implies that pesticide emissions in these countries would not trigger any regulatory responses in surface water. Therefore, for countries without any pesticide surface water regulations, we assume that pesticide emissions are minimal, and we recommend employing the standard model group (e.g., the PWC model).

Among the 49 small-area countries, there were 9 countries that did not have any regulated quality standards for the 12 common pesticides. These countries include Andorra, Cambodia, Japan, Laos, Palau, Philippines, Saudi Arabia, Thailand, and Vietnam. It is likely that these countries primarily focused on managing legacy pesticides in surface water, possibly due to infrequently updated regulatory jurisdictions [10]. Consequently, the ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ values for these 9 countries were found to be zero. For the remaining 41 small-area countries that provided standards for at least one of the common pesticides, we employed the PWC model to conduct an initial filtering study. We categorized their surface freshwater quality standards based on the 12 selected common pesticides, as well as estimated pesticide emission rates. The emission rates for individual pesticides were estimated by calculating the ratio of the annual amount of application for specific pesticide classifications to the country's cropland area in 2017 [59], assuming that the primary use is for agricultural purposes. To perform the calculations, we used cropland area instead of agricultural land because cropland is commonly defined as arable land used for crop and fiber production, where pesticides are extensively utilized. On the other hand, agricultural land may also include grassland and grazing land used for livestock production, where different types of insecticides are employed for insect control [60]. In addition, crops in cropland receive a significant volume of pesticides, particularly the commonly used ones [61, 62]. Although the PWC model was originally designed for regulatory scenarios, it incorporated additional application information such as crop types and weather conditions. To simplify the initial filtering process, we applied the standard scenario of the PWC model (see Supplementary file) [51]. The physicochemical properties and application pattern information for the common pesticides were sourced from a published study [40], which compiled a comprehensive set of model input information from current literature, technical reports, and pesticide manufacturer labels. For all pesticides, the “Photosynthesis Ref Latitude” variable was assumed to be 40° [40] and the reference temperatures for water, benthic, and soil were set as 25 °C.

Table 2 provides a summary of the current countries worldwide along with the corresponding matched model groups based on indicators. As of now, there are 193 countries recognized as member states of the United Nations [63]. Out of these, seven large-area countries were identified with a ${\text{CV}}\left( {{\text{S}}^{{1{\text{st}}}} } \right)$ value greater than 1, and except for China, which operates under a unitary political structure, the remaining six countries with federal legislative systems were assigned the advanced model group. Among the 186 small-area countries, 49 countries have regulations in place for pesticides in surface freshwater. However, nine of these countries have not implemented regulations for any of the common pesticides defined in this study. Therefore, currently, 146 small-area countries have minimal concerns regarding pesticide emissions in surface freshwater from a regulatory perspective, and for these countries, the standard model group is recommended. Next, we conducted a fast-filtering process for the 40 small-area countries that have regulations for at least one of the common pesticides in surface water. The results indicated that 14 countries had simulated pesticide concentrations in surface freshwater exceeding the corresponding water quality standards. Hence, we recommend employing the general model group for these 14 small-area countries. Five countries, namely Estonia, Iceland, Latvia, Spain, and the UK, had ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ values below 1.0, suggesting that the pesticide emissions in these countries are considered light in model selection. Therefore, a standard model would be suitable for addressing their specific needs. In addition, 21 small-area countries lacked emission data for specific pesticide groups. In light of conservative considerations, we recommend using a general model for these countries. The ${\text{CV}}_{{{\text{RS}}}} \left( {{\text{E}}^{{2{\text{nd}}}} } \right)_{{\mathbf{J}}}$ values for the small-area countries are further summarized in the Supporting Information.

Table 2 Classification of global countries and the recommended model groups

Back calculation of pesticide emission rates using the specific model

To compare the back estimates of pesticide emissions using the proposed model, a total of 12 widely used pesticides were selected to illustrate the simulation process, including 2,4-D, aldicarb, atrazine, chlorpyriphos, diazinon, dicamba, diuron, malathion, mancozeb, MCPA, metolachlor, and trifluralin. These pesticides were selected due to their top usage in agricultural lands and available input information for the standard model (i.e., the PWC model) [40, 45]. Notably, the advanced model group is now unable to demonstrate the back calculation due to the lack of specific suitable model. Therefore, we only provide modeling demonstration using the standard model and general models (i.e., the PWC, TOXSWA, and SWAT models) to show the calculation procedure. Table 3 summarizes the simulated emission rate factors using the standard model. For the PWC model, the peak and annual average modes for the pond scenario are used to simulate ${\text{ERF}}_{{{\text{i}},{\text{PWC}}}}^{{{\text{Peak}}}}$ and ${\text{ERF}}_{{{\text{i}},{\text{PWC}}}}^{{{\text{Ave}}}}$ values, and the results indicate that in general ${\text{ERF}}_{{{\text{i}},{\text{PWC}}}}^{{{\text{Ave}}}}$ is higher than ${\text{ERF}}_{{{\text{i}},{\text{PWC}}}}^{{{\text{Peak}}}}$ by at least one order of magnitude. This is because the PWC model simulates the pesticide concentration in nearby surface freshwater using discrete emission patterns (i.e., twice a year), which results in the simulated peak concentrations much higher than the annual average concentrations and thereby the lower simulated ${\text{ERF}}_{{{\text{i}},{\text{PWC}}}}^{{{\text{Peak}}}}$ values for pesticides. Thus, the PWC model can be also used to regulate pesticide emissions in agricultural fields according to surface freshwater regulations if the country provides the environmental quality standards of pesticides for both peak and annual average situations. It is observed that the simulated logarithms of ${\text{ERF}}_{{{\text{i}},{\text{MODEL}}}}^{{{\text{Ave}}}}$ and ${\text{ERF}}_{{{\text{i}},{\text{MODEL}}}}^{{{\text{Peak}}}}$ values for the selected pesticides using the PWC and TOXSWA models all present a linear relationship, shown in Eqs. (12–13) and Figs. 2, 3:

$$ \log ERF_{i,PWC}^{Peak} = 0.56\log ERF_{i,PWC}^{Ave} + 0.07\,\,\, n = 12\,R^{2} = 0.92 $$

(12)

$$ \log ERF_{i,TOX}^{Peak} = 1.40\log ERF_{i,TOX}^{Ave} + 0.87\,\,\, n = 12\,R^{2} = 0.99 $$

(13)

Table 3 Simulated emission rate factors (kg ha⁻¹ year⁻¹ per mg L⁻¹) using the standard models

Fig. 2

Fitted linear relationship between the logarithm (10 base) of the emission rate factors (log ERF) of 12 selected pesticides: peak mode vs annual average mode using the PWC model

Fig. 3

Fitted linear relationship between the logarithm (10 base) of the emission rate factors (log ERF) of 12 selected pesticides: peak mode vs annual average mode using the TOXSWA model

These simulated values can be directly used to determine the pesticide emission rate by employing the pesticide surface freshwater quality standards, which include both peak and annual average standards. From a regulatory standpoint, one of the strategies for pesticide management involves controlling emissions based on environmental quality standards. China has established a surface freshwater quality standard for malathion at 0.05 mg L⁻¹ [10]. Using China as a modeling demonstration, the results showed that if a provincial or regional regulatory agency applies China's national standard to regulate local pesticide emissions, then according to the PWC model (annual average), the discrete emission rate should not exceed approximately 10,000 kg ha⁻¹ year⁻¹ in the region (assuming application once a year). Furthermore, it is important to note that China has not specified whether the surface water standards are set as annual averages or peak values. If the peak value of 0.05 mg L⁻¹ for malathion is applied, then the emission rate using the PWC peak mode in the region should not exceed approximately 33.3 kg ha⁻¹ year⁻¹.

Moreover, we further performed this back calculation on the general model (i.e., the SWAT model). Considering the complexity of spatial data in different countries, pesticide simulation was carried out using only the example data accompanying the model. After watershed definition described in the Supplementary File, annual pesticide application (assuming application once a year) can be set in several orders of magnitude to obtain the corresponding pesticide levels in surface water. For example, input emission rates of aldicarb as 1, 10, 100, 1000, 10,000, 100,000 kg ha⁻¹ year⁻¹ (assuming application once a year) to explore the linear relationship between emission rate (ECR kg ha⁻¹ year⁻¹) and the output annual concentration (AC, mg L⁻¹), the result is shown in Eq. (13) and Fig. 4:

$$ \log AC_{SWAT}^{Ave} = 1.36\log ECR_{SWAT}^{Ave} - 5.77\,\, n = 6\,\, R^{2} = 0.97 $$

(14)

Fig. 4

Linear relationship between the logarithm (10 base) of the emission rate factors (log ERF) of aldicarb and annual concentration of soluble pesticide (log C) $\begin{aligned} \searrow\kern-6.5pt\raisebox{-3pt}{$\lrcorner$}\end{aligned}$

That means, if the region in this example has surface freshwater quality standard for aldicarb at 0.01 mg L⁻¹, the emission rate should not exceed approximately 591.6 kg ha⁻¹ per year. It is worth noting that at the same pesticide application rate, the SWAT outputs will vary considerably across countries, due to the differences in topography, land use, soils, climate and planting structures. Countries in the general model group are advised to conduct this simulation based on their own accurate spatial data.

Regulatory implementations and recommendations

Refining model selections for regulatory assessment

Our proposed approach for modeling selection focuses on managing the quality of surface freshwater. However, as mentioned in the previous section, the advanced model group is now lacks of specific suitable model and thus not suitable for back calculation. We notice that one possible model that can be utilized for the advanced model group in future studies is the Pangea model, which was developed by integrating a Python-based ArcGIS geoprocessor with a MATLAB-based computation engine [29,30,31]. This integration enables Pangea to accurately predict water pesticide concentrations in multimedia and track chemical transports through multi-physics processes. Furthermore, Pangea incorporates geo-referenced information and multiple compartmental systems, allowing for spatially modeling chemical fate and transport across multiple scales. However, this model now is still computation-intensive and it is recommended future research to continue updating this model group and to apply the simplified version in advanced model group. In addition, it's important to note that pesticide emissions to agricultural lands are also subject to other environmental quality standards, such as those for soil and groundwater [64]. Pesticides can be carried off agricultural areas through wind drift and can infiltrate groundwater through rainfalls or irrigations. Certain countries have implemented strict environmental quality standards for soil and groundwater [65], which can further limit the amount of pesticides that can be emitted, in addition to the standards for surface freshwater quality. After pesticides are emitted, the soil compartment, including agricultural and nearby residential soils, becomes the primary recipient of pesticide residues. This means that the fraction of pesticides remaining in the soil may be higher than the fraction present in surface freshwater. Therefore, it is crucial to conduct comprehensive studies to estimate and control the rate of pesticide emissions, taking into account the environmental quality standards. This approach can help effectively manage pesticide emissions and establish regulatory connections between different environmental compartments. In detail, the SWAT and Pangea models can conduct concentration simulations in both soil and water (i.e., surface water and groundwater), but the PWC model cannot. Therefore, in future studies, the standard model group can focus on more comprehensive fate models, such as the PRZM model. In addition, it is important to consider the implementation of good agricultural practices (GAP) for pesticides to ensure a comprehensive approach to pest control and environmental quality standards. Otherwise, if we solely focus on limiting pesticide emission rates to meet water quality standards, it could potentially lead to a reduction in crop production, especially in countries with extremely low environmental quality standards.

Refining model selections for risk assessment

In this study, we utilized water quality standards to calculate acceptable pesticide emission rates. However, it is crucial to adopt a comprehensive approach to managing freshwater quality through various levels of risk assessment. While water quality standards are derived from related risk assessments, it is important to recognize that freshwater bodies serve multiple functions for ecosystems and human society. Therefore, the management of pesticides in freshwater should consider the specific roles that freshwater plays in these contexts. For instance, some freshwater bodies serve as sources of drinking water for humans, while others are intended for aesthetic purposes, such as providing a scenic view in parks. Depending on the specific role of a freshwater body, different levels of risk assessments for pesticides must be conducted to determine the acceptable pesticide concentrations in surface freshwater. In the back calculation process of pesticide emission rate, utilizing the pesticide fate model coupled with the risk assessment model (e.g., cumulative and aggregate risk evaluation system, species sensitivity distributions), can derive a wider variety of values based on the different uses of surface water. This requires the use of appropriate fate models to derive corresponding safe pesticide emission rates.

Refining model selections for environmental agencies

We classified the fate models into three groups based on hierarchical indicators. However, in real-world scenarios, the fate and transport of pesticides within different environmental compartments are influenced by numerous complex factors, including environmental conditions, weather patterns, geological features, pesticide characteristics, agricultural practices, and more. Therefore, there cannot be a rigid rule for environmental agencies to select a specific model group. For instance, in this study, we assigned the general model group for recommending back calculations of acceptable pesticide emission rates in China. While the provinces and regions in China adhere to national environmental quality standards to mitigate pesticide-related environmental pollution, there are significant variations in environmental factors across these regions. In addition, as mentioned earlier, surface freshwater bodies serve multiple roles in society, necessitating risk assessments at different levels. Hence, while China is recommended to use the general model group to define acceptable pesticide emission rates based on national water quality standards, different provinces and regions can employ other model groups for localized environmental practices and management. For example, at a screening level, a standard model can be utilized to assess the impact of pesticide emissions on small water bodies near agricultural lands, such as ponds and rivers. However, for higher-level risk assessments involving multiple purposes like ecological environment and drinking water protection, the advanced model group may be a more suitable choice for local environmental agencies to determine the corresponding pesticide emission rates. EU has developed suitable fate models based on the local environment and agricultural practices. Consequently, when applying the hierarchical screening approach, the PEARL model is also representative in the standard and general model groups, respectively. In summary, the suggested model groups should be used flexibly, taking into account the diverse roles of surface freshwater bodies in ecosystems and human society.

Conclusions

In this study, we propose a hierarchical screening approach for countries worldwide to choose appropriate modeling tools for estimating pesticide emission rates. The approach establishes connections between pesticide emissions in agricultural lands and surface freshwater quality standards. We consider factors such as spatial distribution, pesticide emission conditions, and legislative systems to create hierarchical indicators that classify countries into three main categories. Each category is then provided with corresponding standard, general, or advanced model groups. In addition, we develop an approximate method for standard and general model groups to regulate pesticide emission rates from surface freshwater quality standards based on 12 commonly used pesticides. These hierarchical indicators and approximate method can assist regulatory agencies in selecting appropriate models and then managing pesticide emissions in accordance with surface freshwater quality. However, further research is needed to select a simplified model for advanced model groups and comprehensively investigate pesticide emissions and other environmental aspects.