Calibrating Predicted Mixture Toxic Pressure to Observed Biodiversity Loss in Aquatic Ecosystems

ABSTRACT

Unlike practices in applied ecology, assessing the impact of chemical pollution on biodiversity depends on species sensitivity data from laboratory toxicity effect tests. There are ~12,000 chemicals with such data, enabling quantification of a metric that characterizes the magnitude of the toxic pressure of chemical mixtures on aquatic organisms. However, the calibration between this lab‐based metric and biodiversity effects in the field is lacking. To address this gap, we calibrated both. We quantified mixture toxic pressure levels from extensive water quality monitoring data across 1286 sampling sites and expressed it as multi‐substance potentially affected fraction of species (msPAF). We furthermore quantified species abundance and richness loss for those sites. Calibration of both yielded that the observed potentially disappeared fraction of species (PDF) can be quantified from msPAF as biodiversity impact metric. Species abundance and richness generally declined with increasing toxic pressure, and a near 1:1 PAF‐to‐PDF relationship was derived. Both metrics are key in regulatory chemical policies and comparative biodiversity impact assessments, with PDF also widely used for biodiversity footprinting to assess species loss. Our results imply that the lab‐based mixture toxic pressure metric can roughly be interpreted in terms of species loss under field conditions, that assumed regulatory “safe concentrations” may not fully protect exposed species assemblages, and that comparative biodiversity impact assessments can be made based on mixture toxic pressure metrics. These outcomes are highly relevant for biodiversity protection, and support the transition toward a “safe chemical economy” by enabling the design of compounds and products with lower environmental impacts.

This study bridges the gap between lab‐based chemical toxicity estimates and real‐world biodiversity loss in freshwater ecosystems. Findings show that rising toxic pressure from chemical mixtures is closely linked to declining species diversity, even at levels considered safe. The calibration improves biodiversity impact assessments and supports the shift toward a safer, more sustainable chemical economy.

1. Introduction

Chemicals are an integral part of our daily life. Their emissions reach aquatic ecosystems, impacting biodiversity (Sigmund et al. 2023; Persson et al. 2022; Kosnik et al. 2022). Species are almost always exposed to measurable levels of more than one chemical, that is, unintended mixtures (Schäfer et al. 2023). However, it has historically been challenging to characterize the impact of chemical pollution in the field. This is due to the wide variety of marketed chemicals and the resulting variation in ambient mixtures, the presence of other pressures, as well as the natural diversity of exposed ecosystems and their species compositions (Posthuma and de Zwart 2006; Leclère et al. 2020; Sylvester et al. 2023). Moreover, assessments of chemical effects on species are currently based on insights in chemical hazards derived from single‐species and single‐chemical laboratory toxicity tests. The historical divergence of applied ecology and ecotoxicology has been partly bridged because recent evidence has shown that the toxic pressure exerted by local unintended mixtures accounted for, on average, 26% of explained biodiversity deviance in Europe's rivers (Lemm et al. 2021). This evidence shows the need to improve the scientific approaches that enable the characterization of chemical pollution impacts in the field (Schneeweiss et al. 2023) and for improved practical and policy approaches to prevent or mitigate such impacts on biodiversity. The present study addresses this challenge by characterizing pollution impact patterns in nature, with the eventual aim of improving both protective regulations and environmental quality assessments as well as comparative life cycle environmental performance assessments of products and technologies.

There is currently no effective method to characterize the magnitude of the effects of chemical pollution on biodiversity (Oginah, Posthuma, Hauschild, et al. 2023). Existing predictive approaches characterize the fraction of species potentially affected to some extent, but not per se lost, upon exposure to chemical mixtures, based on a metric derived from laboratory tests on selected species. This “predicted‐risk” toxic pressure metric does not yet translate into a defined magnitude of damage to some relevant structural biodiversity metric or associated ecosystem services effects under real‐world scenarios.

The introduction of a structural biodiversity damage metric for chemical pollution under field conditions is essential for comprehensive decision support, as recently stressed in a review report (JRC 2025). A good and useful metric should be scientifically valid and support environmental quality protection, assessment, and mitigation practices. It should thereby support chemicals policies, biodiversity footprint assessments, and the comparative environmental performance assessments of products and services. Such a metric can be based on mixture toxic pressure assessment if the latter is calibrated to true damage (i.e., species loss). With approximately 12,000 chemicals having measured in vivo laboratory effect test data, quantifying the predicted impact of chemicals and their unintended mixtures as “mixture toxic pressure” (expressed as msPAF, the multi‐substance potentially affected fraction of species) has become feasible, and mixture toxic pressure levels have been mapped, for example, for all European surface water bodies (Posthuma et al. 2019, 2020; Posthuma and de Zwart 2012). Analysis of monitoring data has indicated so far that an increased mixture toxic pressure grossly relates to increased biodiversity damage (Posthuma et al. 2020), providing latitude to derive the degree of biodiversity damage as a function of the pressure metric. In short, it is deemed relevant and possible to describe the PAF‐to‐PDF calibration.

The calibration of PAF‐to‐PDF on the basis of monitoring data of chemicals and species is not straightforward. One must consider that there is a wide variation in species sensitivity to each chemical and recognize that individual species in the field thus likely respond differently to increased toxic pressure, that is, “all species are unequal,” and also that all species are subject to direct and indirect effects of pollution and other pressures (Sigmund et al. 2023; Oginah, Posthuma, Maltby, et al. 2023). These matters have been debated since the initial uses of toxic pressure modeling in the contexts of both protective policies (Hopkin 1993; van Straalen 1993) and the comparative life cycle impact assessment of products (Goedkoop and Spriensma 2001). Nonetheless, it is currently simply assumed that a low toxic pressure level (i.e., msPAF = 0.05) as used for the protective policies implies absence of damage in the field (whereby PAF‐to‐PDF is thus interpreted to imply that 5% potentially affected species equals 0% species loss), with PAF assessments based on laboratory no‐observed effect concentration (NOEC) data. Moreover, for the life cycle impact assessments, practitioners currently assume a 1:1 association between predicted mixture toxic pressure variation and species loss (Posthuma and de Zwart 2012), with PAF assessments based on laboratory EC50 data (i.e., the concentration that causes 50% effect on a life history characteristic), whereby this association would imply, that is, that 20% potentially affected species (exposed at their 50%‐effect level) equals 20% species loss. These assumptions can and need to be scrutinized because of the vast practical implications of unjustified PAF‐to‐PDF calibrations. First, it should be recognized that a loss of five sensitive species and a parallel increase of five opportunistic species may occur, but that response is not equal to a net calculated response of $5-5=0$, and thus an apparent “absence of net damage.” Second, exposure levels in the field are often (much) lower than the aforementioned 50%‐effect level of the used test endpoint. In the present study, we accounted for the diverse abundance response patterns across species and considered a field‐relevant calibration between predicted toxic pressure and observed damage in the present study (Fantke et al. 2018). This was operationalized by using chronic 10%‐effect concentrations (EC10) as the impact metric derived from laboratory toxicity tests to define the mixture toxic pressure as msPAF_EC10 (Owsianiak et al. 2023). We also considered the no‐observed effect concentration as another metric derived from laboratory toxicity tests (msPAF_NOEC), as this metric has historically been used in the context of protective policies. In the present study, we aimed to establish a robust PAF‐to‐PDF relationship at field‐relevant exposure levels by accounting for abundance effects on individual species instead of relying solely on net changes in richness metrics, whereby we covered both the contemporary protective policies (msPAF_NOEC) as well as the current comparative biodiversity impact assessment context (msPAF_EC10).

In the following, we present for the first time a field‐based, quantitative PAF‐to‐PDF calibration based on a large monitoring dataset from the Netherlands. We first harmonized and structured a field dataset on mixture toxic pressure, other abiotic variables, and species abundance and occurrence across different sampling sites in the Netherlands. We then analyzed the data using different statistical techniques to find quantitative associations between chemical pressure, species‐specific abundance patterns, and species richness patterns. Using posterior multiple comparisons tests, we finally characterized the relationship between mixture toxic pressure classes and observed species loss across sampling sites, focusing on the two important practical metrics (msPAF = 0.05 to represent the protective threshold, and msPAF = 0.2 to represent the impact assessment working point). With this approach, we provide insights into the field effects of unintended ambient chemical mixtures and derive the PAF‐to‐PDF relationship for the two practically relevant toxic pressure metrics. We were able to focus the description of results solely on the PAF_EC10 level upon demonstrating that PAF_EC10 and PAF_NOEC—which have historically different roots in designing protective policies versus assessing life cycle impacts—are highly correlated.

2. Materials and Methods

2.1. Overview of Research Steps

Related to current practices in pollution prevention and mitigation, our working hypotheses are (1) that there is no or negligible damage to biodiversity at the regulatory protective threshold concentration, (2) that the damage from exposure levels higher than that implies higher but species‐specific impacts on exposed species, and (3) that, at increasing exposures, those impacts eventually result in abundance changes and species richness loss, at a so far unknown rate. In the historical definition of chemical policies (tracing back to Van Straalen and Denneman 1989), the protective regulatory threshold is defined when PAF_NOEC equals 0.05 (where 5% of the species is potentially affected at their no‐effect concentration, and 95% of the species is thus likely completely unaffected). This exposure level is in use as the operational, regulatory‐chosen threshold between “acceptable” risk and “unacceptable” risk, assuming no damage to structural or functional characteristics of exposed ecosystems (Posthuma et al. 2002; Oldenkamp et al. 2024). In comparative assessments of potential impacts of chemical pollution, the attention is focused on characterizing actual damage from increased ambient exposure levels caused by emissions of chemicals along product life cycles. Here, the biodiversity damage is evaluated at a specific “working point.” That impact working point is quantified as a field‐realistic degree of exposure (and impacts), that is, at PAF_EC10 = 0.2, which is an exposure level where 20% of the species would be exposed at a level causing 10% impact on a life history trait, such as growth and reproduction (Owsianiak et al. 2023). Given these backgrounds, our calibration study focused on two specific mixture exposure levels, where the mixture toxic pressure equals 0.05 and 0.2, respectively. The calibration outcomes further refer to the impact magnitude that would occur at an average, randomly selected site in the study area, as that is also the interpretation of the (site‐agnostic) protective policies and the life cycle impact assessments of products.

We aimed to collate an extensive monitoring dataset for the calibration study to avoid various problems in statistical analyses, such as too small ranges of toxic pressure variation, too few data points given the number of pressure variables, and potential collinearity issues. Therefore, we focused on a large dataset on aquatic monitoring data for the whole of the Netherlands dataset, but also recognized that there may be regional differences in typical exposure and impact patterns as well as in monitoring practices.

The raw data were collected as part of a large field dataset from the Netherlands (Whole‐NL), which was compiled, harmonized, and structured to evaluate our hypotheses. This dataset includes chemical contaminants, site characteristics, and invertebrate biomonitoring data. Chemical pollutants (in total n = 703) were monitored at 5134 locations, focusing on nutrients, heavy metals, pesticides, industrial pollutants, and pharmaceuticals, primarily in the water column, with 2–4 measurements per year. A typical data characteristic is the presence of vastly diverse subsets of chemicals monitored per sample. Invertebrates (in total n = 1830 species) were monitored at 5939 locations in Dutch surface waters, sampled once per year using a standard macrofauna net and identified at the finest possible taxonomic level.

Employing various statistical techniques, we analyzed the data to unveil associations between mixture toxic pressure and species‐specific abundance data and species richness. Subsequently, we established classes of sites with different degrees of mixture toxic pressure and characterized the relationship between observed abundance change patterns and species loss across sampling sites through posterior multiple comparison tests. As the toxic pressure assessment only considers adverse responses in the ecotoxicity laboratory test data, we primarily focused the analyses of the monitoring data on species initially present under clean conditions and showing abundance decline under increased toxic pressure. Initial presence was defined when a species was present in the cleanest data bins that were distinguished. The focus on declining species aligns with the focus on negative impacts as the response type that is also used to characterize other biodiversity impacts in the comparative assessment context, such as acidification and eutrophication. In addition, we evaluated some alternative approaches as a robustness check on the PAF‐to‐PDF outcomes, considering potentially different outcomes when considering all species or invasive species.

We finally calibrated the predicted and observed chemical‐related species loss by characterizing the pressure‐response relationship with a Generalized Additive Model (GAM), with special emphasis on the regulatory‐used protective threshold (msPAF = 0.05) and the biodiversity damage assessment working point (msPAF = 0.2). The thus‐obtained results are site‐ and time‐agnostic, representing the expected species richness loss for randomly selected sites in the Netherlands, which aligns with the site‐agnostic nature of the protective standards and comparative assessments. Figure 1 summarizes these steps, with Table S1 detailing the process flow for deriving the PAF‐to‐PDF relationship.

FIGURE 1.

Overview of the workflow followed in compiling, harmonizing, and analyzing field monitoring data for the Netherlands and for various separate waterboards in order to characterize the calibration between mixture toxic pressure and structural biodiversity damage responses, that is, the multi‐substance potentially affected fraction to the observed potentially disappeared fraction of species (msPAF‐to‐PDF) calibration, with emphasis on characterizing (absence of) damage at the regulatory protective threshold exposure and at the impact working point (msPAF = x = 0.05 and x = 0.2, respectively). The msPAF to PDF relationship was derived primarily for species that are present in the cleanest sites and sensitive, but also (as robustness check) when considering all species and when considering invasive species.

2.2. Characterizing Chemical and Species Monitoring Data

Data on abiotic water quality characteristics and on the occurrence and abundance of aquatic invertebrate taxa were obtained from the Dutch monitoring authorities (Waterboards; Postma et al. 2021). The collection of invertebrate abundance data was done quantitatively by field experts (Beers et al. 2014). Although species abundances are variable in response to the sampling moment and site conditions, we used the raw, quantitative data until we summarized outputs to derive conclusions on the PAF‐to‐PDF calibration. Data from 18 of the 21 Dutch waterboards considered to be of high quality (i.e., transparency on how data were collected) were collated, covering information on chloride anions (Cl), dissolved organic carbon (DOC), Kjeldahl nitrogen (NKj), total suspended solids (TSS), temperature of the water (Tw), and pH of surface water samples. Data on chemical concentrations covered organic, inorganic, and heavy metal compounds, with variation in the composition of the compounds measured per sample.

The data were used to derive the toxic pressure associated with the concentration of each observed chemical, as described in Table S1, yielding characterizations of the predicted toxic pressure level of separate compounds, compound groups, and total ambient mixtures expressed as (multi‐substance) potentially affected fraction of exposed species (msPAF, range 0–1). These calculations were based on species sensitivity distributions (SSDs) derived from the species‐specific exposure concentration that causes an effect of 10% on a vital characteristic (such as growth or reproduction), following calculation methods described in Posthuma et al. (2019). We note that the toxic pressure depends not only on the concentration of a chemical but also on the aforementioned water quality parameters, as those determine the chemical‐specific bioavailable concentrations [following approaches described in Postma et al. (2021)]. Initially, both msPAF_NOEC and msPAF_EC10 toxic pressure levels were derived for each sample as metrics used in the contexts of characterizing (1) (expected absence of) species loss at the regulatory protective threshold and (2) the unknown species loss at the impact working point, respectively. Both metrics are rooted in the two historically different application domains related to the present study (protective standards in chemical safety‐ and environmental quality assessment, and biodiversity damage assessment in comparative product assessments, respectively) but were expected to covary closely (as both reflect an exposure level at which adverse effects in exposed test species shift from “no effect” to “small effect (10%)” in laboratory tests). This covariation was evaluated to evaluate whether both can be exchanged in the further analysis steps. The results of this evaluation are summarized in Figure S1, confirming that the metrics are not entirely identical but exchangeable for the PAF‐to‐PDF calibration. Therefore, we further present only msPAF_EC10‐analysis results. When interpreting the outcomes of the calibration studies, we acknowledge that this mixture toxic pressure metric can be under‐estimating the true toxic pressure (per sample) if compounds that are present in the water bodies are not analyzed in the chemical monitoring.

The collected data on water quality parameters, chemical concentrations, and aquatic taxa were combined by geospatial coordinates (XY) and timestamps (T). The combined data appeared partly incomplete, because data on observed taxa abundance did not always have all‐pressure and all‐chemical data for all water bodies at all timestamps. Given the focus on toxic pressure, sites with missing mixture toxic pressure information were excluded. Datasets from the years 2000 until 2014 were selected, and unique species were counted to determine site‐specific species abundance and species richness.

The collinearity of the seven abiotic parameters, including the mixture toxic pressure metric, was evaluated via variance inflation factor analysis, to screen whether the mixture toxic pressure metric covaries with one or more other abiotic parameters, as that would complicate the attribution of observed damage to chemical pollution (Schipper et al. 2014).

We focused on Whole‐NL for the Netherlands as a whole, but also performed separate calculations for subsets of data to evaluate the robustness of outcomes. Subsets of data were defined by origin, that is, waterboards, as those are the hydrologically delineated and organizationally defined units for water quality and quantity management and monitoring. Results for Whole‐NL and one waterboard (Delfland) are shown in the main text, and outcomes for other waterboards are shown in the Supporting Information S1.

2.3. Initial Analyses of Toxic Pressure and Species Responses

The toxic pressure metric (msPAF_EC10) and species abundance and species richness were plotted as raw data to evaluate whether a higher mixture toxic pressure relates to reductions in species abundance or species richness. These data were analyzed with a Generalized Additive nonlinear relationship Model (GAM) to characterize per‐species abundance change with increased toxic pressure as a downward, neutral, or upward abundance change pattern. This was needed to focus further analysis steps on initially present species that show a downward response (sensitive, similar to the laboratory‐tested species).

2.4. PAF‐to‐PDF Calibration Methods

The PAF‐to‐PDF calibration was done using various statistical methods, employing different assumptions. In these methods, we included data points at the species level and excluded species identified at higher taxonomic levels, opportunistic species (present only in more polluted water bodies), and those observed in fewer than five sites across the considered years.

With the selected data, we investigated how mixture toxic pressure levels covary with species abundance for sensitive species. Outcomes were aggregated to the species assemblage level based on three classification principles. First, we classified the available data in five toxic pressure class bins, with approximately equal numbers of XY (toxic pressure, richness) observations per class, followed by posterior multiple comparisons tests. This step was inspired by the use of five ecological status classes distinguished in the EU‐Water Framework Directive. Second, a practice‐related approach considered the regulatory threshold (msPAF = 0.05) and the biodiversity damage working point metric (msPAF = 0.2) as critical decision‐making points, examining response magnitudes for the three bins defined on this basis. Box‐ and violin plots were employed to illustrate species abundance and richness patterns, followed by posterior multiple‐test comparisons; the Kruskal–Wallis test compared species means across groups, followed by Dunn's pairwise comparison. Third, the pressure‐response relationship (PAF‐to‐PDF) was characterized by distinguishing various bin sizes (up to 50) and using the resulting outcomes to interpolate the species richness change at msPAF = 0.05 and msPAF = 0.2 with GAM. The species losses at these levels are interpreted as the species loss at a random site in the Netherlands for the given exposure levels, and the use of multiple bins served as a robustness check on the PAF‐to‐PDF calibration outcomes.

All statistical analyses were conducted using R 4.1.2 (R Core Team 2021). Figures were generated using the ggplot R package, version 3.4.1 (Wickham 2016). All datasets and R scripts used to evaluate the PAF‐to‐PDF relationship are openly available in Zenodo at https://doi.org/10.5281/zenodo.15591077.

3. Results

3.1. Overview

First, the compiled data are characterized for the Netherlands as a whole and subset(s) defined by the hydrologically defined waterboard boundaries. Second, outcomes of the collinearity evaluation are given. Finally, the various PAF‐to‐PDF analyses are described.

3.2. Data Characterization

3.2.1. Chemical Pollutants and Mixture Toxic Pressure

Figure 2 shows the variability of the chemical monitoring data for Whole‐NL and Delfland. The top panels show the proportions of identified chemical use categories (Figure 2a) and the proportion of metals and other chemicals (Figure 2b). The panels illustrate that the representation of the use categories and the metals varies across waterboards. Metals are considered separately because ambient concentrations may partly be of natural origin. Pesticides are often a dominant class of monitored synthetic chemicals, followed by pharmaceutical, industrial, and multiple‐use substances. Different waterboards are further characterized by different patterns in the dominance of the various chemical use categories and metals (Table S2).

FIGURE 2.

Characterization of the chemical monitoring data for surface water samples of the Netherlands (left) and the relatively data‐rich Delfland waterboard (right) covering the years 2000–2015 for the data included in this study across 622 unique chemical substances. (a) The top row represents the proportion of identified chemical use categories, whereby the “others” category represents chemicals with unidentified use or metals. (b) The bottom row represents the proportions of metals and non‐metal chemicals. Percentages on top of each Delfland bar represent the proportion of Delfland data points relative to the total number of data points in the entire Netherlands for each bin.

The chemical monitoring data show the typical pattern often found in large monitoring data sets (e.g., Isaacs et al. 2022). The number of monitored chemicals per site can vary from a single compound (class 1–10) to > 1000. Despite this variation, monitoring schemes are characterized by regulatory rules that ask for consistent monitoring of dominant chemicals (named “priority substances” and “specific pollutants” for the Dutch parts of the river basins in Rhine, Meuse, Scheldt, and Ems), according to the EU‐Water Framework Directive (EC 2000). Since mixture toxic effects are often dominated by a few dominant compounds (Price et al. 2012), the monitoring data likely represent true differences in the degree of mixture‐induced damage. Despite the high variability, this makes the data useful for the PAF‐to‐PDF calibration. The distinctive data patterns of the different waterboards provide a basis to check whether the calibration is found for (very) different subsets of the data.

The calculated mixture toxic pressure (msPAF_EC10) values ranged from 0 to 0.75 across 1286 unique sampling sites (XY coordinates) throughout the Netherlands, which appeared similar to the variability across the 229 sites in the Delfland regional waterboard (Figure 3). Notably, the highest observed mixture toxic pressure level (0.75) originates from Delfland. This toxic pressure variation indicates that we may expect different damage levels across both study areas, as the highest values imply that 75% of the species would exhibit a 10% performance reduction for vital traits at the most polluted sites. Given the distribution of toxic pressure levels, the data set was considered useful for the PAF‐to‐PDF calibration.

FIGURE 3.

Cumulative distributions of mixture toxic pressure variability (msPAF_EC10) across the Netherlands and the Delfland regional waterboard (Delfland data stretched over the same X‐axis range as the Whole‐NL data). The mixture toxic pressure level is shown as a “dot” per XY‐site (rank‐ordered based on the mean calculated value), and the vertical bars (error bars) indicate temporal variability of mixture toxic pressure in the dataset for those sites over the years 2000–2015.

3.2.2. Other Physico‐Chemical Parameters and Collinearity

Six additional abiotic predictors were analyzed with a total of 231,178 observations, providing an overview of the observed ranges for each water quality parameter and additional insight into spatial variation in physico‐chemical characteristics across samples and regions. The chloride concentration (Cl) ranged from 3.72 to 6.78 μg/L, DOC from 0.43 to 1.74 mg/L, total nitrogen (NKj) from 0.13 to 5.15 mg/L, pH (KCl extraction) from 4.78 to 9.38, TSS from 0.52 to 2.21 mg/L, and water temperature (T _w) from 7.51°C to 13.28°C. Characteristics of the studied environment in each waterboard are available in Table S1.

Collinearity evaluation between the abiotic parameters and the mixture toxic pressure showed that the latter varies sufficiently independently of other considered parameters in the Netherlands and Delfland. The variation inflation factors (VIFs) were consistently (far) below 5, indicating low collinearity. This also holds for various individual waterboards with sufficient data, although a high VIF was observed for mixture toxic pressure and other parameters in the Rijn en IJssel waterboard. These findings suggest that the Whole‐NL dataset can be analyzed regarding PAF‐to‐PDF patterns, similar to several but not all waterboards. We proceeded with the analysis for separate waterboards with sufficient data (see Table S1 for complete waterboard output).

3.2.3. Invertebrates Data Summary

The Whole‐NL invertebrates dataset covers a span of three decades, covering the years 1983–2014, totaling 219,544 invertebrates data points, with an overall trend of increasing number of data points over the years; the peak year was in 2010 (n = 17,710), and the lowest intensity occurred in 1991 (n = 694). For the Delfland waterboard, 35,627 invertebrate data points were collected (Figure 4). The early years (1986–1999) show a modest count, with occasional fluctuations. The highest count in the Delfland region occurred in 2005, peaking at n = 4117. The lowest count was observed in 1986, with only 15 data points. Meanwhile, 2006 (n = 603) and 2007 (n = 474) represent a notable drop compared to the number of data points of adjacent years. Among the invertebrates, the Arthropoda taxonomic phylum had the highest number of data points for the whole of the Netherlands (n = 149,515), followed by Mollusca (n = 38,011), Annelida (n = 30,074), and Others (n = 1944). Similarly, in Delfland, the Arthropoda taxonomic phylum had the highest data point counts (n = 21,897), followed by Mollusca (n = 7580), Annelida (n = 5758), and Others (n = 392).

FIGURE 4.

The annual number of sampling data points for invertebrates records for 18 waterboards from the Netherlands (left) and the Delfland regional waterboard (right); n values indicate the total number (rows of records) of available data across the monitored years. The colors represent the considered taxonomic phyla. Data from 2000 to 2014 were used for the multi‐substance potentially affected fraction to the observed potentially disappeared fraction of species calibration study.

Invertebrate data records from 2000 to 2014, corresponding to the period of the abiotic monitoring data, were selected for further PAF‐to‐PDF calibration, comprising 183,789 data records from the Netherlands region and 24,824 from Delfland, identifying 1217 and 492 species, respectively.

3.3. Patterns in the Raw Data

3.3.1. Abundance Per Species

Raw data on species‐specific abundance observations plotted against the mixture toxic pressure level are illustrated graphically for Whole‐NL Figure 5 and Delfland (Figure S2) for the nine species with most available data points for three dominant taxonomic phyla (Arthropoda, Annelida, and Mollusca). The relative dominance of these species varies widely across the sampling sites (Figure S3). The panels in Figure 5 (Whole‐NL) and Figure S2 (Delfland) consistently demonstrate a general decreasing trend of observed abundance data per sample (density‐values) with increasing mixture toxic pressure, indicating the latter as a limiting factor for the abundance of these species (Cade and Noon 2003). The high inter‐site variability (Y) at a specific mixture toxic pressure level (X) is a consequence of natural, waterbody‐type specific differences in species composition across water bodies and other pressures.

FIGURE 5.

Illustration of trends in raw species abundance data with increasing mixture toxic pressure for selected species (9 out of 1217) in the dataset for the entire Netherlands from three taxonomic phyla with the highest number of data records. The X‐axis represents the mixture toxic pressure (msPAF_EC10), and the Y‐axis represents species‐specific abundance data (as a structural biodiversity metric). Sparse or absent dots in the upper right corner indicate that mixture toxic pressure acts as a pressure that limits species abundance. The colors represent different taxonomic phyla.

3.3.2. Species Richness

Raw data on species richness (all species) plotted against the mixture toxic pressure level are illustrated graphically in Figure 6a (top row), again showing that an increased mixture toxic pressure acts as a limiting factor to species richness. Again, high inter‐site variability is shown, which is associated with natural differences between waterbodies, multiple stress impacts, and indirect effects.

FIGURE 6.

(a) Trends in absolute species richness for all observed species with increasing mixture toxic pressure for the Netherlands (left) and the Delfland regional waterboard (right). Similar to species abundance (Figure 5), high X‐high/high Y values are scarce or absent, indicating that an increased mixture toxic pressure limits species richness. Each dot indicates the number of unique species counted at a particular site across the years, and n indicates the total number of unique sites. (b) Trend analysis of relative species richness change for only sensitive species, with indications for GAM‐interpolated values of the species richness at X = 0.0 (clean sites, blue printed value, and dot) and X = 0.05 (regulatory protective threshold, orange printed value, and dot) and X = 0.2 (impact working point, red printed value and dot). The mean species richness calculated from all sites of the Netherlands (left) and Delfland (right) is taken as a reference point, defining Y = 0.

The raw species‐specific abundance data were analyzed to identify the species needed for the PAF‐to‐PDF calibration (initially present and showing a net sensitive response). This was done with species‐specific GAM modeling. The approach (illustrated in Figure S4) resulted in a distinction of sensitive, opportunistic, and (few) neutral responding species (not shown, based on reduced, increased, or neutral differences in estimated abundances at increasing mixture toxic pressure). A total of 373 species exhibited an opportunistic response pattern (see Figure S4 for examples). This suggests that these species are relatively tolerant to chemical pollution and that they have a potential advantage in chemically stressed environments. In the context of the PAF‐to‐PDF calibration, we further focused on damage by presenting results for sensitive species only.

The raw data for the selected sensitive species were used to derive an initial PAF‐to‐PDF calibration assessment, focusing on the relative species richness at the cleanest sites vis a vis the regulatory protective threshold and the impact working point (X = 0.0, 0.05, and 0.2, respectively). For this, we expressed the richness data as relative values (with the overall species richness of all data for a study area as an anchor). Then, we assumed a smooth function between mixture toxic pressure and species richness (via GAMs). The outcomes of this imply that an increase of the toxic pressure to the protective regulatory threshold (msPAF = 0.05) would imply a species loss of 10% and that a further increase of the pressure to the impact working point (msPAF = 0.2) implies a species loss of 32% for the Whole‐NL data set. These values are 3% and 14% for the Delfland data, respectively (Figure 6b). These outcomes suggest a consistent decline in the richness of sensitive species in response to toxic pressure across both data sets, assuming that the pressure‐effect relationship follows a smooth curve (which is unknown). Note that the net species richness reduction was expected to be 0% at the protective threshold; whether the observed 10% and 3% changes are significant is addressed below.

3.4. PAF‐to‐PDF Calibrations and Significance Tests

3.4.1. Evaluating Mixture Toxic Pressure and Significant Degrees of Impacts

Laboratory toxicity tests commonly show an initial range of exposure concentrations that induce no effects, which is a generally observed phenomenon that is neglected when using a GAM to analyze the available data (as above). Therefore, the data were analyzed using other methods to explore the exposure level from which species' richness changes are significant. The raw data were therefore classified into toxic pressure bins (referred to as “Groups”), and differences between bins were evaluated with posterior tests. Figure 7 illustrates the species richness (average and variation) and posterior test results based on splitting the data into five and three bins for Whole‐NL data and a selected waterboard.

FIGURE 7.

Illustration of species richness patterns in the Netherlands (top) and in the Delfland regional waterboard (bottom), using box and violin plots, after binning the sites into five (almost equal number of samples per bin; left) or three toxic pressure groups (considering the regulatory protective threshold and the impact working point, defined by X = 0.05 and X = 0.2, respectively; right). Data analyses were made after excluding opportunistic species and species occurring in fewer than five sites. The box and violin plots highlight differences in sample densities, especially species numbers (μ) across bins. p‐values denote the significance of the differences between bins in posterior tests. The sites are categorized by mixture toxic pressure levels, from minimal (Group 1) to high mixture toxic pressure sites (Group 5), with box and violin plots indicating data distributions. Subfigure (a) shows the species richness patterns for 783 species in the Netherlands. Subfigure (b) displays the same for the Delfland regional waterboard. “μ” denotes mean species count, and “n” indicates the total number of sample sites in each mixture toxic pressure group. Groups are characterized by their specified mixture toxic pressure (msPAF_EC10)‐ranges (top left in each panel).

The panels illustrate different phenomena. First, focusing on the mean species richness (μ), increased mixture toxic pressure consistently covaries with decreased mean species richness per bin for the Whole‐NL data, significantly onward from Group 3 (0.017 < msPAF < 0.037), with 20.7% average species loss in group 5 (0.087 < msPAF < 0.75). Notably, on average the species loss is 6.3% in Group 3 at mixture toxic pressure levels that are below the assumed protective regulatory threshold (msPAF = 0.05). Second, a similar pattern was found for the Delfland waterboard, with a significant species loss of 19.4% onward from Group 4 (0.056 < msPAF < 0.115) and 24.2% average species loss in Group 5 (0.115 < msPAF < 0.75). Third, the Whole‐NL and Delfland results illustrate that the outcomes of the study are consistent (increasing toxic pressure results in decreased species richness) but that various aspects differ (e.g., the mixture toxic pressure above which the species richness effect is significant, and the shapes of the violin plots). These findings illustrate that it is not straightforward to derive significant toxic pressure‐species richness trends from a low number of sample data. However, in both cases, the findings can be attributed to a likely impact of mixture exposures (VIF < 5). Fourth, the results for the bins defined based on the practical criteria illustrate that species losses are significant for the bins (0.05–0.2) and (> 0.2) compared to the (0.0–0.05) bin. Taken together, the outcomes suggest that a significant but small species richness reduction occurs at a mixture toxic pressure that is quantified with estimates below the protective regulatory threshold (msPAF < 0.05). The results should be interpreted with care because the mixture toxic pressure metrics in the data set can underestimate the true mixture exposure (when present chemicals are not monitored).

3.4.2. Evaluating Impacts at Policy‐Relevant Mixture Toxic Pressure Levels

Figure 7 does not yet estimate the PAF‐to‐PDF calibration target, which asks for an estimate of the species loss at a randomly selected water body at an exposure level equal to the protective threshold and at the impact working point. That calibration was finally made by interpolation of binned data via GAM modeling. Hence, Figure 8 shows the summary results of the binning and interpolation assessments for various numbers of toxic pressure bins (as robustness check) for the Whole‐NL and Delfland data (output for waterboards with sufficient data and acceptable VIFs are shown in Table S1).

FIGURE 8.

Deriving the multi‐substance potentially affected fraction to the observed potentially disappeared fraction of species (PAF‐to‐PDF) relationship based on the observed fraction of disappeared aquatic invertebrate species (mean Y) per mixture toxic pressure bin (X‐group) based on varying numbers of bin groups. The relative species richness values are interpolated from the data at X = 0, X = 0.05, and X = 0.2 and depicted with blue, orange, and red colors. In practical use, these estimated values allow us to characterize the species richness loss for the average, randomly selected site in a study area, given an ambient mixture toxic pressure. Grey bands represent the 95%‐confidence intervals for the Generalized Additive Model predictions. The confidence interval is not plotted beyond the mean mixture toxic pressure value of the highest bin.

The relative species richness of each bin (Figure 8) is shown as black dots, which represent the mean mixture toxic pressure and mean species richness of each of the bins, and the GAMs (and the 95%‐confidence intervals) fitted to these data are shown (truncated to the highest dot). Regardless of the number of bins, the results show some important observations and consistent trends. First, the data density is highest in the lower toxic pressure range (covering the two points of practical interest, X = 0.05 and X = 0.2). Second, at the lowest (extrapolated) exposures (Y = 0), the species richness is consistently higher than the average of the country or region, that is, 10 or 11% higher than the average of all data. Third, evaluating impacts of increasing toxic pressures, the shape of the fitted GAM is consistently downward, implying that the species initially present in reference conditions are increasingly disappearing up to X~0.2, beyond which there are some sample points associated with higher average species richness. Fourth, the relative species richness varies between −0.02 and 0.0 at X = 0.05, implying an estimated species loss of approx. 10% at the protective threshold. Note that the posterior test showed a significant decline in richness from Group 3 onward, with mixture toxic pressures in some cases somewhat lower than 0.05 (in the five‐bin case, see Figure 7). Fifth, at the impact working point (X = 0.2), the species loss is consistent—on average over the sites in each bin—and estimated as 26%–33% (this is derived as illustrated in Figure S5 and Table S1 for all bins results). The analyses reveal that the response patterns initially decrease in the data‐dense lower exposure range (irrespective of bin numbers) and may rise at some sites with a (far) higher mixture toxic pressure, possibly due to indirect factors like food abundance. Despite thorough checks, a consistent upward trend beyond X = 0.2 is observed across all bin number options in the Whole‐NL and the Delfland datasets (Figure 8).

A similar and complete evaluation of the dataset of individual waterboards appeared feasible for four waterboards, whereby the data analyses for the other were discontinued after observing high toxic pressure VIF and/or too low data numbers (Table S1). A low number of site data, a low range of mixture toxic pressures, and collinearity of the toxic pressure with other pressures appear to act as limitations to derive PAF‐to‐PDF calibration outcomes.

3.4.3. Evaluating the Role of Species Selection

The robustness of the outcomes based on selecting sensitive species was tested by considering different species selections, that is, invasive species and all species. Invasive species are often considered robust against various pressures, which may influence the PAF‐to‐PDF calibration. The patterns and main trends of the PAF‐to‐PDF calibration appeared to remain grossly the same when invasive/alien species data were excluded from Whole‐NL and Delfland datasets (Figure S6). These outcomes suggest that invasive species do not significantly alter our PAF‐to‐PDF calibration outcomes. Similarly, when comparing PAF‐to‐PDF relationships including all taxa versus excluding invasive or opportunistic species, the overall trend of biodiversity decline with increasing toxic pressure remains consistent. While the precise estimates of potentially lost species vary slightly (see Figures S9 and S10), the general damage patterns indicate that our findings are robust across different taxonomic scopes.

4. Discussion

4.1. Overall Methodology, Patterns, and Relevance

Chemical pollution is considered to be one of three planetary crises (UN Environment 2019). Specifically, chemical diversity, production, and emissions globally are still rising (Bernhardt et al. 2017), and landscape‐level evaluations showed biodiversity impacts as a net result of this trend (Lemm et al. 2021). Both crises are thus linked, but their characterizations hinge on disjunct methods, that is, laboratory toxicity tests and biomonitoring data from the field, respectively. As existing chemical pollution risk metrics (i.e., mixture toxic pressure values) are highly important in practice for preventing and mitigating chemical pollution, we aimed to develop a data analysis method that can be employed to link the chemical pressure metrics to ecological impacts. Using various methods, we derived a policy‐relevant PAF‐to‐PDF calibration after ascertaining that the effects of other environmental stressor variables do not bias this relationship.

We identified three important phenomena for the PAF‐to‐PDF calibration. First, all data analyses suggested that species richness is increasingly affected by increased mixture toxic pressure, although we were not able to discern a field‐based “no‐effect” threshold due to natural variation and differences in packages of monitored chemicals. Second, the available data showed that the impacts of chemical pollution were already significant near‐ but also somewhat below the regulatory‐assumed protective criterion (X = 0.05). This is an important finding, as the assumption of “no significant impact” at this exposure level is the historical basis of many global protective chemical policies. Third, our results suggest an effect of approximately 30% observed species loss (PDF) based on a predicted effect on 20% of the species (PAF). This is also an important finding, as this PAF‐to‐PDF relationship can be used to define the so‐called “Damage Factor” in the formulae to quantify ecotoxicity impacts in the context of comparative sustainability assessments of products (Verones et al. 2017).

Because of the importance of the findings for multiple contexts (protection as well as impact analysis), there is a need to broaden the study to other geographical areas, exposure conditions, unintended mixtures, and exposed species groups before the findings can be generalized. It is also key to consider the limitations of the present study. One limitation in the methodology may be that chemicals in the field (and indeed causing impacts) may not be measured (consistently across sites or at all) in the monitoring programs. This implies that mixture toxic pressure values (X‐values in all figures) may be underestimated to a lesser or larger degree. That is, if chemicals that are present but unmeasured would be added to the monitoring, X‐values (and estimated curves as in Figure 8) would shift (slightly) to the right. In turn, the effect of such under‐represented chemicals is that there may be less than the calculated 10% species richness loss at X = 0.05 and less than 30% loss at X = 0.2 for the Whole‐NL data. We reason that this shift may be present, but that it is unlikely a large shift or a shift that holds for many points, because the monitoring strategies aim to focus on regionally relevant chemicals. Despite this uncertainty, the data show that chemical pollution causes adverse impacts on biodiversity at exposure levels below‐ or near the protective standard and that those consistently increase with increasing toxic pressure. This observation provides the missing and important alignment between applied ecology and applied ecotoxicology.

4.2. Field Data Considerations

The methodology developed to bridge the gap between applied ecotoxicology and applied ecology depends on large sets of biomonitoring data, as illustrated by comparing the findings for Whole‐NL and data sub‐sets, where the analyses stalled for several of them. A combination of diverse, existing monitoring data sets proved necessary for obtaining the required information. However, it is essential to acknowledge the characteristics and limitations associated with using data originally collected for other purposes. Various issues require attention. (1) Samples for biological, chemical, and environmental stressors are often collected at different sampling moments and places. To address this disparity, we integrated invertebrate data, chemical concentrations, and related information using XYT data (spatial coordinates and time stamps). Thereupon, instead of defining a specific “clean” set of reference sites, we used the full range of toxic pressure gradients in the dataset. Our results summarize the effect on species abundances and richness changes for an average, typical site anywhere in the study area (Netherlands or a specific waterboard) compared to a typical, least‐polluted site derived from the data set. (2) Chemical concentrations falling below the detection limit were omitted from the analysis, but this may be one of the causes of under‐estimating mixture toxic pressure per site. It is recommended to monitor chemicals with methods that are sensitive at chronic exposure levels that can be ecotoxic. (3) We are aware of the limited availability of biomonitoring data, which remains a challenge in ecological studies (Berger et al. 2016; Üblacker et al. 2023). (4) The monitoring dataset can be biased, for example, considering the fact that the current study contained more chemical monitoring data on non‐metals (496,589 data points) than on metals (150,600 data points, but much fewer unique compounds). This indicates a greater diversity and quantity of chemical pollution data for synthetic compounds but a much lower measurement frequency per synthetic compound than for metals. Studies on large areas and specific regions are needed to address the potential sources of bias and to obtain robust insights into PAF‐to‐PDF relationships. (5) As another source of potential bias, the Arthropoda phylum was dominant in the Netherlands monitoring data (68% of the total dataset) and the data for the Delfland region (61% of the total dataset), which can be explained by their adaptability and diversity (Santos et al. 2021). The relatively high occurrence of some species, such as Asellus aquaticus , suggests that local conditions may favor some specific species (Lafuente et al. 2021). In contrast to some dominating taxa, there is a wide variety of less common taxonomic phyla in the data, such as Bryozoa, Platyhelminthes (flatworms), Cnidaria, and Porifera (sponges). (6) The approach used to measure mixture toxic pressure relied on concentration addition and response addition models, which characterize additive effects as an overall proxy for mixture‐level effects at realistic ambient concentrations (De Zwart and Posthuma 2005). In theory, mixture effects may deviate from these additive effects under particular circumstances, such as very high concentrations or specific pairs of chemicals (Cedergreen 2014), but these circumstances do not generally apply to our monitoring data. (7) Species loss is shaped by multiple interacting pressures, including habitat change, pollution, and climate change. The direction and magnitude of species responses vary with the type of pressure, taxa involved, and spatiotemporal context. While our study focuses on bridging the gap between applied ecology and ecotoxicology by exploring the association between chemical pollution (mixture toxic pressure) and observed species loss, other abiotic factors, such as pH, might influence observed patterns. These influences likely vary across sites, complicating single‐driver damage attribution. As noted by Keck et al. (2025), disentangling the combined effects of environmental stressors remains a major challenge in field‐based biodiversity research and should receive further attention in future studies.

Regardless of these kinds of specific characteristics and limitations of the often‐available, typical sets of biomonitoring data, our results allowed us to describe a PAF‐to‐PDF relationship that is relevant and robust, and that quantifies expected damage (species richness loss) for a randomly chosen water body within a considered geographical scope: it was found for Whole‐NL data and for sufficiently data‐rich waterboards, as well as with and without invasive species.

4.3. Interpreting Mixture Toxic Pressure

The mixture toxic pressure metric (msPAF) is used globally as a summary metric to quantify the predicted magnitude of impacts of unintended mixtures on exposed species assemblages. It has an implicit interpretation, as illustrated in Figure 3: at the most‐polluted site in the Netherlands, the metric would imply that 75% of the species would show an effect of 10% on a vital characteristic. It also means that a smaller fraction is exposed at the 50%‐effect level and an even smaller fraction at the 90% level or at the—unknown—species‐loss level. It also means that some indirect effects may occur when pollution affects particular species, but this is not quantified with mixture toxic pressure assessments. Although a higher mixture toxic pressure thus grossly indicates a larger impact, the metric does not provide clear insights into the net effects that occur in the field. Deriving such insights was only possible for large data (sub)sets. Although we were able to derive a PAF‐to‐PDF calibration (Figure 8), the impact prediction remains, however, of a degree, not of a kind. The predictive model does not identify which species would be affected. However, it is the magnitude of adverse responses that is helpful, in practice, to prioritize protective or mitigation responses and select products with the lowest impacts.

4.4. A Field SSD for Mixtures

The outcomes illustrated in Figure 8 show, for the first time, a set of panels that can be seen as field‐SSDs of species loss under exposure to unintended chemical mixtures. In their traditional use, SSD‐models are commonly derived from laboratory toxicity test data and used for practical purposes in environmental protection, assessment, and management. However, those lab‐based SSDs are used for decision‐making under the assumption that the sensitivity distribution of the laboratory‐tested species for the selected test endpoints (e.g., growth, reproduction) resembles the sensitivity distribution of the species in the field. This has so far been no more than a key assumption. The results in Figure 8, however, represent a field‐SSD for chemical pollution and net species richness loss. The shape of the curve implies that different species (netto, all direct and indirect effects accounted for) are differently sensitive to exposure to unintended mixtures. The results in Figure 8 can be interpreted as a field exposure‐effect relationship for species loss.

4.5. Species Loss and Overall Change

The data analyses onward from Figure 6b have focused on sensitive species in view of the PAF‐to‐PDF calibration goal. This neglects that chemical pollution also causes other impacts, such as an increased abundance of opportunistic species. This means pollution‐induced “change” is larger than currently summarized in the PAF‐to‐PDF results. We recommend considering species‐specific abundance responses to characterize impacts and species loss of sensitive species or to characterize the upward and downward abundance changes separately to characterize all changes, to avoid that opposing abundance changes are canceled out. Additionally, spatial and temporal modeling of biodiversity trends as such—without necessarily linking to chemical (or other) stressors—could provide valuable complementary insights into underlying ecological patterns and should be considered in future studies.

4.6. On Practical Consequences

4.6.1. Protective Policy Thresholds and True Impacts

It has historically been one of the first objectives of applied ecotoxicology to derive the concentration level of chemicals below which the structural and functional integrity of ecosystems would be protected. In the 1980s, this resulted in the definition of the so‐called 95%‐protection level, which was based on the fifth percentile of a NOEC‐based SSD (Van Straalen and Denneman 1989). Two critical assumptions apply: the distribution of lab‐tested species resembles the distribution of species sensitivities in the field, and no impacts would occur when exposure remains below the 95%‐protection level. More than 30 years later, our results show the surprising outcome that the targeted goal of protection is not fully reached. A statistically significant response is found from Group 3 (Figure 7), and GAM modeling suggests that increasing exposure levels between X = 0 and X~0.2 suggest increasing species loss. In practice, however, today's policy Guidance Documents on deriving protective standards suggest employing caution. That is, the originally assumed 95%‐protection criterion is used in combination with an application factor of 5 before the protective regulatory standard is derived. This lowers the protective standard, and the factor was added to handle uncertainties in the assessment. This practice means that the goals and methods of environmental protection are, on average, serving the set goal—that is, yielding sufficient protection and no or minimal species loss.

4.6.2. Ecotoxicity Characterization in Life Cycle Assessment: Current Working Point

The impact assessment phase of life cycle assessment (LCA) studies aims to characterize the potentially disappeared fraction of species associated with product or service life cycles via all impact categories that are linked to “ecosystem quality,” such as acidification and eutrophication. For practice, there is a strong need to derive a “damage factor” for ecotoxicity via PAF‐to‐PDF association studies. The present study shows that the current working point for impact assessment (X = 0.2) falls in the range or realistic ambient exposure levels. Our results, therefore, allow one to combine ecotoxic pressures via the PAF‐to‐PDF calibration with the other impact pathways that cause species loss. This allows for quantifying the impact of various product emissions as aggregated species loss. Pending verification via other studies, we submit that our PAF‐to‐PDF calibration suggests that the predicted impact at the LCA‐working point (20% of species potentially affected) implies grossly equivalent damage in terms of effects on species richness, that is, a loss of sensitive species of approx. 20%.

4.6.3. Ecotoxicity Characterization in LCA: Historical Working Point

Earlier LCA methods for ecotoxicity also employed PAF‐to‐PDF extrapolation since Goedkoop and Spriensma (2001) formulated the earliest ideas. In this vein, LCA experts have so far employed an approx. 1:1 ratio (based on studies such as Posthuma and de Zwart (2012)), but that ratio was based on the field‐unrealistic 50%‐effect impact working point, and it neglected the difference between loss of sensitive species and net change (upward and downward responses not separately considered). Upon selecting the new working point as the global standard in LCA, this thus asked for a comparison and an update of the PAF‐to‐PDF assessments: an altered outcome would have the implication that “ecotoxicity” assessments in LCA would be relatively (far) more important than other impacts than considered so far, or (far) less. This was evaluated via a full recalculation of the data using the old EC50‐based impact working point. The outcomes suggest that the relative importance of ecotoxicity in LCA remains the same. The old and new working points relate (see regression in Figure S7), and this results in a PAF‐to‐PDF assessment for the earlier used working point as in Figure S8. The species loss estimated under the past and the current working points are similar, as shown for, for example, the Netherlands 30‐bin example: outcomes are −33% (Figure S8) and −32% (Figure 8), respectively. Apparently, changes in both the impact working point and the PAF‐to‐PDF study design resulted in no obvious net change. Nonetheless, it remains key to validate the PAF‐to‐PDF calibration with other data sets (area, unintended mixtures, species groups), as this influences global comparative biodiversity impact and footprinting assessments.

4.7. Pollution Impacts in Dutch and European Surface Waters

Our study on the impacts of chemical pollution on Dutch surface waters can be compared to other research findings (Lemm et al. 2021). In that study, observed ecological impacts in European surface water systems were, on average over all study sites, attributed for 26% to chemical pollution. This average value is surprisingly in line with the present study's findings for data on Whole‐NL or separate waterboards with sufficient data (considered average and variation of impact magnitudes in Figure 8).

5. Conclusions

A method was developed and tested with an extensive biomonitoring data set to gain insights into the meaning of predicted mixture toxic pressure values as a predictor of species richness loss. The method was employed successfully. The assessments showed that an increased predicted mixture toxic pressure implies increased impacts on species abundance and richness patterns. The response pattern was summarized as field‐SSD, which in turn could be summarized for practical applications as a PAF‐to‐PDF relationship that bridges applied ecotoxicology and applied ecology. The PAF‐to‐PDF relationship describes what an increase in toxic pressure implies for a randomly chosen water body within a considered geographical scope, and this is meant to be useful information for various practical contexts (both protective policies and LCA), as those also operate site‐ and time agnostic. The PAF‐to‐PDF relationship is functional for its purposes, but it does not allow for the exact prediction of local impacts, which may deviate from the average. The findings suggest that the currently operational approaches to set protective standards serve their goal, as the data do not suggest that there can be large impacts at exposures below those standards. The findings also suggest that a pragmatic 1:1 PAF‐to‐PDF relationship may be assumed to assess ecotoxicity damage in LCA studies. A predicted change of 0.2 as a potentially affected fraction corresponds to an average fraction of approximately 0.2 of lost species. However, further evaluation of other regions, species, and mixtures is essential for a generalized view of how the PAF‐to‐PDF relationship should be employed to evaluate protective policies and to compare biodiversity footprints and chemical footprints from a life cycle perspective.

Author Contributions

Susan A. Oginah: conceptualization, data curation, formal analysis, methodology, resources, software, validation, visualization, writing – original draft, writing – review and editing. Leo Posthuma: conceptualization, methodology, writing – review and editing. Jaap Slootweg: formal analysis, writing – review and editing. Michael Hauschild: conceptualization, writing – review and editing. Peter Fantke: conceptualization, funding acquisition, methodology, writing – review and editing.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1.

Table S1: Summary of modifying factors ranges and mixture toxic pressure across waterboards, including variance inflation factor results, statistical test outcomes, and predicted biodiversity loss at regulatory and life cycle assessment benchmarks.

Data S2.

Oginah, S. A. , Posthuma L., Slootweg J., Hauschild M., and Fantke P.. 2025. “Calibrating Predicted Mixture Toxic Pressure to Observed Biodiversity Loss in Aquatic Ecosystems.” Global Change Biology 31, no. 6: e70305. 10.1111/gcb.70305.

Data Availability Statement

The data that support the findings of this study are openly available in Zenodo at https://doi.org/10.5281/zenodo.15591077. Data on abiotic water quality characteristics and the occurrence and abundance of aquatic invertebrate taxa were obtained through the Dutch National Institute for Public Health and the Environment (RIVM) as part of national monitoring programs from the relevant Dutch authorities via Zenodo at https://doi.org/10.5281/zenodo.15585796. The Species Sensitivity Distribution (SSD) models based on ecotoxicity data for invertebrate taxa were obtained from https://doi.org/10.1021/acs.est.3c04968.