Environmental fate and exposure models: advances and challenges in 21 st century chemical risk assessment

Environmental fate and exposure models are a powerful means to integrate information on chemicals, their partitioning and degradation behaviour, the environmental scenario and the emissions in order to compile a picture of chemical distribution and ﬂ uxes in the multimedia environment. A 1995 pioneering book, resulting from a series of workshops among model developers and users, reported the main advantages and identi ﬁ ed needs for research in the ﬁ eld of multimedia fate models. Considerable e ﬀ orts were devoted to their improvement in the past 25 years and many aspects were re ﬁ ned; notably the inclusion of nanomaterials among the modelled substances, the development of models at di ﬀ erent spatial and temporal scales, the estimation of chemical properties and emission data, the incorporation of additional environmental media and processes, the integration of sensitivity and uncertainty analysis in the simulations. However, some challenging issues remain and require research e ﬀ orts and attention: the need of methods to estimate partition coe ﬃ cients for polar and ionizable chemical in the environment, a better description of bioavailability in di ﬀ erent environments as well as the requirement of injecting more ecological realism in exposure predictions to account for the diversity of ecosystem structures and functions in risk assessment. Finally, to transfer new scienti ﬁ c developments into the realm of regulatory risk assessment, we propose the formation of expert groups that compare, discuss and recommend model modi ﬁ cations and updates and help develop practical tools for risk assessment. that are  t for purpose. We propose the formation of expert groups to address this task. volatile chemical calculated by Goktas & MacLeod 60 using the BETR Global multimedia transport model. Higher values of remoteness index and pTE indicate locations on the Earth that are more remote from emissions of chemicals to air that occur in proportion to the intensity of night-time light observed from space. Histograms illustrate the area distribution of di ﬀ erent values of remoteness index or pTE.


Introduction: evolution and current status of fate and exposure modeling
Assessing the environmental fate and exposure of chemicals used in commerce is a challenging but critical part of assessing the risks associated with their production, use and release.
Environmental fate and exposure models play an important role in the chemical assessment process, and when coupled with monitoring data, can provide reliable information about the exposure of humans and the environment to chemicals. 1 Over the last three decades, the development and application of environmental fate models has been facilitated by a repeating pattern of co-operation between model developers and users in ad hoc working groups. In 1994, workshops were organized by the Society of Environmental Toxicology and Chemistry (SETAC) at Leuven, Belgium (April, 1994) and later at Denver, Colorado (November, 1994). These two workshops brought together 30 model users and developers to discuss the application of environmental fate and exposure models; to evaluate their performance; to provide guidance on their use; and to make recommendations on how the models could be improved to broaden their acceptability and encourage their use. 2 Seven years later, an OECD/UNEP Workshop was organized in Ottawa, Canada (October, 2001), to assess the utility of multimedia fate models for estimating overall environmental persistence and long-range transport of chemicals in the context of assessing potential persistent organic pollutants (POPs) and/or persistent, biaccumulative and toxic substances (PBTs). The Ottawa workshop marked the beginning of a period of activity by a working group of model developers and users (the OECD Expert Group on Multimedia Modeling) that led to the publication of a model comparison study, 3 recommendations for assessing persistence and long-range transport with models 4 and a consensus model, the OECD Overall Persistence and Long-range Transport Modeling Tool (The OECD Tool 5 ). In 2013 a working group jointly organized by the three Scientic committees of the European Commission, DG SANCO (SCHER, SCHENIR, SCCS), 6 published the scientic opinion "Addressing the New Challenges for Risk Assessment" in which a number of opportunities to improve exposure models to increase their ecological realism and predictive ability for the growing number of chemical classes on the market were identied. Some of the fate and exposure assessment challenges from the scientic opinion were reported by Di Guardo and Hermens. 7 Most recently, the European Centre for Ecotoxicology and Toxicology of Chemicals (ECETOC) organized a workshop on environmental fate and exposure models in Brussels, Belgium (May, 2017). Following the format of previous workshops, the ECETOC workshop brought together users and developers of environmental fate models. At this workshop, industry scientists and regulators who conduct chemical hazard assessments using PBT criteria, and risk assessments for industrial chemicals, pharmaceuticals, biocides and agricultural chemicals, came together to review current standards of practice in exposure modeling and to identify key challenges for addressing 21 st century needs in chemical hazard and risk assessment. The authors of this review article have been active in many of the workshops summarized above, and all participated at the recent ECETOC workshop. This review article summarizes key observations from the workshop, with a particular emphasis on how advances in environmental fate models over the last 20 years have addressed the challenges identied in the 1994 workshops, and to identify the primary needs to further advance the models.
Environmental fate models provide a framework to assemble a quantitative and mechanistic description of the sources of chemicals to the environment, and the pathways they follow enroute to their ultimate degradation or irreversible sequestration. Fate models can be integrated with, or loosely-coupled to, multi-pathway exposure models that describe the transfer of chemicals from the environment into the food and water ingested by humans, and water and soil that directly contacts the human skin. 8 Models are currently heavily relied upon to assess chemical hazard, exposure and risk, and to formulate chemical management strategies. But models did not always occupy such a strong position as scientic and policy support tools.
One of the main outputs of the modeling workshops that were held in the 1990s was a report by Cowan et al. 2 that identied a number of research areas that required attention in order to strengthen the utility of models as scientic and decision-support tools. Cowan et al. 2 recognized that the models were powerful conceptual tools capable of characterizing and quantifying the sources, environmental pathways, and exposure routes for chemicals used in commerce. And, they recognized the importance of identifying and quantifying the uncertainty and variability associated with a model's description of these processes. The following ve key areas were identied as requiring attention by Cowan et al. 2 in 1995: (1) Clear classication of chemicals that are within the domain of applicability of models; chemical properties and degradation half-lives and their variability with temperature; emission data, including background environmental concentrations; (2) Model algorithms and parameterization to increase the delity of the model's descriptions of the real environment within the bulk compartments of air, water, soil and sediment that were generally included in models at the time, and to extend the models to describe additional compartments such as vegetation and ice/snow; additionally, the need to develop data sets to describe inter-media transfer coefficients in a variety of environmental scenarios; among the other areas to improve were sub-models for chemical fate in soil and sediments as well as waste-water treatment models, and bioaccumulation and food chain models.
(3) Model comparison exercises to enhance reliability and credibility, especially in model documentation; (4) Sensitivity and uncertainty analysis to provide condence intervals to support hazard and risk-based decisions and to identify key areas for further research and model development; (5) Establishment of a set of guidelines for good modeling practice in scientic and decision-support applications of models, with a particular attention to the model validation issues and the comparison of predicted to measured results.
Model developers have invested considerable resources to address these areas since the publication of Cowan et al. 2 in 1995. Moreover, they have expanded the scope and applicability of fate and exposure models in several other dimensions that go beyond these recommendations. Areas where there have been notable developments include: Development of models with a range of spatial and temporal resolutions that can be used in concert with eld data to rationalize chemical concentrations and estimate the rates of emissions or key fate processes at different spatio-temporal scales; Inclusion of transformation products in model scenarios; Scenario analysis to investigate the inuence of climate variability and climate change on exposure to environmental contaminants; Development of fate and exposure models for engineered nanoparticles and for mercury; Adaptation of the principles from chemical fate and exposure models to pharmacokinetic models and application to human biomonitoring data.
Here we review the progress made in these areas, discuss the mechanistic foundations of fate and exposure models, and identify open questions and needs for strengthening future model development and application. We rst provide a summary of progress in different areas and then highlight selected topics of particular interest.

Progress in fate and exposure model development since the mid-1990s
The areas of fate and exposure modeling where progress has been made range from chemical property data needed as model inputs to model parameterization and design, types of scenarios that can be handled by the models, model evaluation by comparison to eld data, sensitivity and uncertainty analysis, and more.

Chemical properties, specic model parameters
Chemical property data are essential as model inputs. Since the publication of the property data handbooks by Mackay et al., 9 property data for many organic chemicals have been critically evaluated, missing data have been estimated with improved chemical property estimation methods, and sets of recommended best estimates have been published for several classes of environmentally important chemicals (e.g., PCBs, 9,10 HCHs, 11 PAHs, 12 PFASs, 13 siloxanes, 14 chlorinated paraffins 15 and chlorinated dioxins and furans, 16 and complex mixtures of halogenated organics 17 ). Polyparameter linear-free energy relationships (ppLFER) 18 have been established to estimate partition coefficients for a range of environmental media for which partitioning cannot be related to the octanol-water partition coefficient (K ow ) (e.g., ice/snow, 19 mineral surfaces, 20-22 aerosols 23-25 and polymers and biopolymers 26 ). A great deal of effort has also been invested in expanding the applicability domain of models to better assess the environmental fate and bioaccumulation of polar and ionizable organic chemicals. [27][28][29][30] Lastly, model parameters describing environmental masstransfer processes have been critically reviewed and updated. 31

Emission data
Two complementary approaches have been employed to estimate the emission source strength of chemicals used in technical applications and consumer products. The rst is the bottom-up approach that was pioneered by Breivik et al. [32][33][34] to estimate the emissions of PCBs globally, and applied by Diamond and colleagues [35][36][37] for the city of Toronto. Other bottomup emission inventories have been compiled by Wang et al. 38 for PFCAs, by Glüge et al. 39 for chlorinated paraffins, Li et al. 40 for a-HCH, and by Zhang and Tao for 16 PAHs. 41 The second approach is the back-calculation of emissions from measured concentration data by inverse modeling, 14,[42][43][44][45][46][47] which is discussed in more detail below.

Expanding the scope and applicability of the models
A key contribution in this area is the denition of a variety of metrics of exposure-related environmental and human exposure hazard that can be calculated with fate and exposure models. These include metrics of persistence and long-range transport potential that reect a chemical's tendency to be widely distributed, to form long-lived reservoirs in the environment, and to lead to the formation of persistent transformation products. 5,[48][49][50][51][52][53][54][55][56][57][58] Models have also been applied to map the remoteness of locations in the global environment from sources of chemical pollution, 59,60 see Fig. 1. Metrics of environmental risk and human exposure potential for chemicals have been derived within the framework of fate and exposure models. [61][62][63][64][65] Multimedia models of the 1990s included bulk compartments for air, water, soil, and sediment. Several additional environmental compartments, sub-compartments and processes have been incorporated into fate and exposure models, including terrestrial and aquatic vegetation, 66-79 DOC in terrestrial and aquatic systems, 76,80,81 ice and snow, 82-84 different types of aerosols, 85 mountainous regions, 86,87 and the effect of intermittent rain. [88][89][90] The fundamental concept underlying all fate and exposure models is the integration of chemical transport, transfer and degradation processes into mass balance equations for the chemical investigated. This concept of setting up multi-media mass-balances is generally valid and goes beyond the most common type of models that have been developed and applied to non-ionizing, oen long-lived, organic chemicals. For example, mass-balance models have been developed for nanoparticles [91][92][93][94][95][96][97][98] and recently for microplastic. 99 A class of organic chemicals of emerging concern are the ionizable organic substances. Given that the process descriptions of most of the current multi-media environmental fate models were derived from the well-established understanding of partitioning and mass transfer of neutral organic chemicals, expanding the domain of applicability of the models to address ionizable organics represents an important development. Unit world mass balance for ionizing chemicals using activity in lieu of fugacity have been presented. 100,101 The multimedia activity model for ionizable (MAMI) organic chemicals has recently been modied in the development of a multimedia chemical fate model for China (SESAMe), designed using the nested structure of SimpleBox. 102,103 The SimpleTreat model 4.0 104 has been updated to include improved handling of ionizable organics, thus providing a tool for estimating chemical fate and behaviour in domestic sewage and activated sludge systems for down-thedrain chemicals, such as pharmaceuticals. 105 Model evaluation by comparison to eld data Fate and exposure models can be evaluated by comparing the model results to eld data. If a model can reproduce eld data (collected under conditions that correspond to the assumptions and settings in the model), this strongly increases the credibility of the model and demonstrates that the model is suitable for the chemical and environmental system considered. Results from fate and exposure models have oen been found to agree well with eld data; this includes at the global level, 63,[106][107][108][109][110][111] and at the continental level 112-115 for selected persistent organic pollutants, and also in local and regional systems, such as the cases of diuron in a small, highly dynamic tropical catchment, 116 plant protection products in agricultural catchments at different scales, [117][118][119][120][121] PBDEs in a Swiss lake, 122 and PCBs in the Adriatic Sea. 123 Sensitivity and uncertainty analysis Sensitivity and uncertainty analyses have become more and more common and are now an integral part of many fate and exposure modeling studies. MacLeod et al. 124 provide a practical guide to sensitivity and uncertainty analysis for fate and exposure modeling studies. In the OECD Pov and LRTP Tool the user can activate the built-in options for a sensitivity and uncertainty analysis of the results for overall persistence and long-range transport potential. Becker et al.  128 who set up six principles of good modeling practice in the eld of fate and exposure modeling. These principles range from a specication of the goal and scope of the assessment to specifying the model used, specifying all input and all output data, conducting a sensitivity analysis so that the most inuential model parameters can be identied, and specifying the limitations of the model results and the modeling exercise.

Improved modeling of bioavailability of contaminants
Bioavailability of chemicals to terrestrial and aquatic environments and more generally the ecological realism of scenario descriptions has been pointed out as a critical issue to evaluate the exposure and therefore the toxicity of hydrophobic chemicals in the environment. 7,129,130 Some models were recently published and have incorporated additional compartments and subcompartments (such as dissolved organic carbon, macrophytes, algae) to account for their inuence on dynamically changing bioavailable concentrations, 76,81,131,132 as well as compartment variability in scenarios. 78,132,133 3. Models for small spatial scales and dynamic scenarios In the 1990s fate and exposure models had mostly been developed to describe long-lived chemicals, such as organochlorine pesticides and PCBs. Analytical chemistry techniques for such substances required large-volume samples and intensive effort in the laboratory for analysis. Thus, the temporal and spatial resolution of available measurements was low, and, due to the persistent nature of the chemicals, long time scales were needed to observe trends in concentrations. Models developed in the 1990s correspondingly described environmental conditions using long-term average conditions for temperature, rain rate, wind speed, etc., and models were oen applied using a steady-state assumption that represents a hypothetical conditions where emissions are balanced by removal processes. 134 In the years since the 1990s it has been increasingly recognized that environmental fate models can be more exible if they implement more rapidly changing conditions. This is needed if the chemical considered has short degradation half-lives and/or if there are strong and rapid changes of the environmental conditions. Advances in analytical chemistry capabilities have provided more measurements at higher spatial and temporal scales to provide context for these model scenarios and to support model-measurement comparisons that enable a process-based understanding of chemical dynamics. This also made it possible to appreciate the role of temporal and spatial changes of specic compartments, such as a pond or a forest, in driving concentration changes in the main compartments e.g. water and air. 76,132,133,135 For example, Camenzuli et al. (2012) 116 investigated the fate of the herbicide, diuron, in a small river catchment in Queensland, Australia. The problem in this case was that appreciable amounts of the diuron used on sugarcane and other crops in Northern Queensland were ushed into the Great Barrier Reef lagoon by water runoff. In the lagoon, the herbicide affects, among other effects, the seaweed and thereby wildlife living on seaweed (dugongs, turtles). There is a strong seasonal change in the rain rates in Northern Queensland and the rain rates in the wet season are very high (with peaks above 200 mm per day). Camenzuli et al. (2012) 116 developed a model of the Tully River catchment with six compartments (agricultural soil, non-agricultural soil, seawater, sea sediment, air, water on land that is present under ooding conditions) and ran the model with very high temporal resolution (time steps on the order of seconds). This allowed them to drive the model with the highly dynamic actual rainfall data measured at a station in the Tully River catchment (Fig. 2a).
The model results of interest were the uxes of diuron into the seawater compartment and the diuron concentrations in that compartment. Comparison with diuron concentrations measured in the lagoon shows that the model results agree well with the measured data, see Fig. 2b. The concentrations follow the seasonal rainfall trend.
In this way, the model was established as a tool that was then used to explore the effect of a diuron restriction scenario (no application of diuron in the wettest months of the year, November to April). Such a restriction would lead to a decrease in diuron concentrations in seawater by a factor of two. Overall, this case study shows that fate and exposure models are not limited to cases with "slow" processes and average environmental conditions, which is sometimes seen as a serious limitation of this type of model, but can also be set up in a way that they describe highly dynamic conditions with high accuracy.
Similarly to the diuron case, an integrated compartmental model (DynaPlus) was recently developed and used to predict the surface water contamination by pesticides used in an intensive orchard cultivation in a mountain region in Italy. 121 Mountainous regions are characterized by rapid land response to precipitation, resulting in quick runoff chemical peaks and consequent potentially high peak exposure concentrations in aquatic ecosystems receiving such runoff. The model consisted of a dynamic water network model (built upon the Dyna model 136 ) and a spatially explicit terrestrial multimedia model 81,120 parametrized using specic scenario and emission data provided by the local authorities. The model incorporated an improved algorithm to calculate water runoff in steep slopes (average 30%, up to 80%) which allowed to more accurately calculate chemical runoff in these conditions. The integrated model was run for two cultivation years and results of chemical runoff from treated orchards were compared to ad hoc measured water concentrations for the organophosphorus insecticide chlorpyrifos in order to benchmark the model against eld data. The model simulations showed that concentration peaks of chlorpyrifos in river water (Fig. 3) spanned and oscillated between 3 and 4 orders of magnitude, in the range of 0.1 to 1000 ng L À1 , which is well within the range of the acute EC50 for Daphnia (around 100 ng L À1 ). 137 Given that ecotoxicity data (such as those for Daphnia) are usually determined at constant exposure, the results of these model simulations reveal the difficulty of evaluating the effect of higher concentration peaks but shorter in time (peak exposure) in protecting aquatic ecosystems.
A dynamic model for assessing pesticide residues in fruit in comparison against EU and US maximum residue levels has recently been presented by Mendez et al. 138 in relation to the cultivation of bananas in Costa Rica. In addition to water, air, soil, and sediment compartments, Mendez et al. 138 also included a banana plant and estimated the residues of diuron, ethoprofos, and epoxiconazole. Their results demonstrate the importance of dynamic models as decision-support tool that complements ongoing efforts to assess the ecological effects of

Day-night cycling of SVOCs in boundary layer air
As sampling and analytical capabilities improved for atmospheric micropollutants, measurements of persistent semivolatile pollutants with temporal resolution of several hours became more widely available in the 2000s. In 2007, MacLeod et al. 139 developed a dynamic multimedia model that included variable atmospheric boundary layer height, temperature, hydroxyl radical concentrations and windspeed. They collected measurement data from six studies that reported concentrations of semi-volatile organic chemicals in air over consecutive 24 hour cycles, and demonstrated that in all cases the model could explain the drivers of day-night variability of concentrations in terms of variable process rates. Their analysis with the dynamic mass balance model identied four controlling factors that determine the day-night variability of concentrations of persistent semi-volatile chemicals in air: (1) source characteristics, (2) temperature dependent air-surface partitioning, (3) atmospheric mixing height and stability, and (4) hydroxyl radical concentration.
The insights from the 2007 modeling study informed the design of day-night measurement campaigns that were designed specically to determine the rate of emissions of air pollutants from diffuse sources in urban areas. Several of these studies were carried out in Zurich, Switzerland to measure emission rates of PCBs, 42,140 PBDEs, 43,44 peruoro alkyl substances, 47 cyclic siloxanes, 141 and PCDD/Fs. 140 The recent quantication of ongoing PCB emissions in Zurich led the Swiss Federal Office for the Environment to re-open the question of PCB emission sources and to commission the set-up of a current and comprehensive PCB emission inventory for Switzerland. 142 Mass balance models applied in conjunction with day-night measurement campaign data have demonstrated the potential to infer process rates for hydroxyl radical degradation of PCBs, 139 for a more physically-based description of the air compartments in fugacity models, introducing hourly variable PBL (planetary boundary layer) compartment 81,143,144 and for reversible air-vegetation exchange of PCBs, PBDEs and chlordane. 42,145 Studies of this type provide the opportunity to evaluate the process descriptions provided by mass balance models in a relatively controlled situation, which helps to identify weaknesses in the models that can be improved, and ultimately improve model performance and reliability in applications at all spatio-temporal scales.

Environmental fate models as decision-support tools
Applying environmental fate models in decision-making requires that the models are understood by stakeholders, credible and that their reliability has been assessed. For multimedia models, the three key elements are: (i) the models themselves, i.e. the process descriptions and their mechanistic basis, and the model parameterization; (ii) the chemical property data used as model input; and (iii) the emission data, including their temporal and spatial resolution, used as model input.
Importantly, the three elements are independent of one another, which means that each of them may introduce a certain type and degree of error into the model results, but also that errors from these different elements can be traced separately, which makes it easier to constrain the source of error. The rst element, process descriptions and model parameterization, strongly relies on the general laws of nature such as the law of the conservation of mass, the law of mass action, and the laws of chemical kinetics and diffusion, as well as empirical data reecting the environmental conditions assumed in the model. Nevertheless, in setting up a model there is always, to some extent, room for choices to be made by the model developer. The effect of this variability in model setups can be assessed by model-model comparisons. An extensive model comparison was performed for multimedia fate models used to determine the overall persistence (Pov) and long-range transport potential (LRTP) of organic chemicals. 3 The other two elements, chemical property data and emission data, oen introduce greater error than the choice of model or details of the model setup. Chemical property data available in data bases and compilations oen show surprisingly high variability and uncertainty; they need to be selected carefully and are ideally checked for their internal consistency, e.g. the octanol-air partition coefficient (K oa ) needs to be equal to the ratio of K ow and the air-water partition coefficient, K aw . 146,147 Emission data are oen the most uncertain and incomplete element, 148 as discussed above.
Even when the property and emission data are selected carefully, there will be some remaining uncertainty that has to be accepted. For both, property and emission data, it is then advisable to determine a condence factor or uncertainty range and to propagate this range through the model, for example by means of Monte-Carlo simulation or by applying the rules of error propagation. 124 One environmental fate model that was developed explicitly for decision making purposes is the OECD Pov and LRTP Estimation Tool. 5 Before the actual tool was created, nine different models for Pov and LRTP calculations were evaluated and compared in much detail. 3 Also a set of well-characterized reference chemicals that dene the scales of Pov and LRTP was introduced. 4 Once the validity of the underlying concepts was established and the scope and purpose of the model within a decision-making context were dened, the specic setup of the tool, i.e. the format of the data input and display of results, the model algorithm, options for uncertainty analysis, and the soware environment, was dened. 5 Because the tool is based on a robust and common soware environment (MS Excel for both Windows and Macintosh computers), is easy to use (reliable and robust handling of incorrect input), offers all results in both a graphical and numerical format, and is available free of charge from a trusted organization (OECD), it was quickly adopted by its users in government, research and industry. Since then, it has been used regularly for the evaluation of possible persistent organic pollutants in the context of the Stockholm Convention.

Outlook for improving the application of models within regulatory frameworks
This paper has summarized research conducted in the last three decades that addressed the needs identied by Cowan et al. 2 and that evolved in the various workshops that have been held since then. Improvements in characterizing and quantifying physicochemical properties and emission estimates and renements in model formulation and parameterization have led to better comparability between model results and monitoring data, which in turn imparts greater condence to the use of models within regulatory chemical risk assessment. Overall, there has been considerable progress in model development, model transparency and model reliability. However, in the context of regulatory risk assessment it is important to strike a balance between increasing model complexity and diversity driven by scientic progress, on the one hand, and the need for models that are practical and relatively easy to use, on the other hand. Ongoing scientic development and the presence of open scientic questions must not reduce the availability of practical tools for regulatory risk assessment out of "paralysis by analysis". Therefore, there is a continuous need for the translation of new scientic insights into the sphere of -oen simpliedmodeling tools suitable for regulatory risk assessment.
This translation process is a task in itself and requires time and resources. Participants from the 2017 ECETOC workshop expressed frustration regarding the lack of advances in environmental fate and exposure models that are included within the tools used for regulatory purposes. For instance, as described above, substantial progress has been made in expanding the applicability domain of the models with respect to both chemical and environmental properties. However, the European Union System for the Evaluation of Substances (EUSES), which is used for assessing exposure of general chemicals and biocides regulated under REACH, has not been updated to reect advances in research, with the most recent update (EUSES 2.1.1) parameterized according to algorithms and assumptions described in the 2003 EU Technical Guidance Document. 149 There is clearly a need to improve the pace at which advances in research are implemented within regulatory tools. However the process of evaluating new modeling approaches and adopting them into regulatory tools must be conducted in an open and transparent manner that facilitates building consensus among key stakeholders from industry, NGOs, government, and academia who are actively engaged in the development and use of environmental fate and exposure models.
The development of the OECD Tool demonstrated how bringing an expert group of modelers together can result in successful development and adoption of a regulatory modeling tool. Another example of an expert group building consensus about the application of exposure modeling within a regulatory context is the Forum for Co-ordination of pesticide fate models and their Use (FOCUS). FOCUS was an initiative of the European Commission, with the objective to harmonize the predicted environmental concentrations (PECs) of active substances of plant protection products (PPP) in the framework of the EU Directive 91/414/EEC. It provided a platform for close cooperation of expert groups of scientists drawn from regulatory agencies, academia, and industry under the auspices of the Commission's DG SANTE.
We propose that the creation of expert groups that follow the approaches provided by FOCUS and the OECD Tool is a fundamental element for achieving consensus among stakeholders to update and improve models used in regulatory exposure assessment. Multi-stakeholder expert groups provide a forum for communication and education about developments and application of environmental fate and exposure models. When the groups are tasked with a clear objective, they can work in a focused manner to dene the attributes of a fate and exposure model that is "t for purpose" to confront a regulatory question. Tools used at lower tiers of assessment will likely address questions that differ from tools used at higher tiers of assessment. Consequently, separate expert groups may be needed to confront related questions, or a single expert group might be tasked to develop a tiered set of models that address the problem at several levels of complexity. Ideally, an active and collaborative expert group that represents all key stakeholders should be able to dene the tools, assumptions, and acceptable level of uncertainty associated with the various assessment tiers that might be required in a regulatory context. The work of these groups would include model comparison exercises, and critical evaluation of assumptions and limitations of models with the ultimate goal of providing recommendations for the application of existing models, or purpose-built new soware for conducting exposure assessments that reects scientic consensus among stakeholders.
We acknowledge that creating multi-stakeholder expert groups that are tasked to agree on how to best update regulatory tools used for chemical risk assessment is not likely to be a trivial exercise. It will require commitments of time and money from the stakeholders to be successful. In the specic case of EUSES, the process could build on the foundation provided by the list of high-priority corrections, improvements, and extensions that was published by van de Meent et al., 150 the key observations published in the 2013 EU scientic opinion, 6 and the recommendations of the 2017 ECETOC workshop participants. 151 A high priority for improving exposure models used in regulatory assessments is the need to expand the chemical applicability domain. The rapid development of target and nontarget analytical chemistry techniques based on LC/MS over the last decades has made measurement and monitoring of environmental contaminants that are ionisable organic chemicals, including halogenated organic acids and chemicals used in pharmaceuticals and personal care products, an important area of environmental chemistry. Attention placed on these chemicals by their detection in the environment has been a key driver for drawing regulatory attention. Particularly at lower screening and prioritization tiers of assessment, regulatory models are currently not capable of handling these chemicals in a reliable manner, and a lack of reliable property estimation methods for such chemicals further hinders screening level assessments.
Many regulatory models are also not appropriate for describing measurement data collected at high temporal resolution (e.g. hours). The 2013 Scientic Opinion 6 expressed a need for criteria and protocols for obtaining and comparing chemical monitoring data, especially for evaluating the fate of chemical mixtures, including metabolites. Such data should be collected and assembled in databases that can be used for model development, validation and benchmarking to support developing and calibrating higher temporal resolution models that could eventually be applied with condence within a regulatory context.
In the absence of reliable property data and regulatory models that are appropriate for interpreting monitoring data and placing the data in the context of the environmental fate, transport and exposure of chemicals, measurements documenting the occurrence of pollutants place pressure on regulatory authorities to infer potential risks without a quantitative framework. 152 Recent experiences with cyclic volatile methyl siloxanes, 153,154 and the public policy debate about regulation of microplastics in consumer products 155,156 illustrate the confusion this situation can create. It is therefore vital that chemical property measurement and estimation methods, and environmental fate and exposure modeling, continue to advance and evolve to support rational management decisions for the 21 st century's environmental contaminants of concern.

Conflicts of interest
There are no conicts of interest to declare.