Challenges and future prospects for developing Ca and Mg water quality guidelines: a meta-analysis

Abstract

Anthropogenic activities have the potential to increase water hardness (Ca + Mg) in receiving waters to toxic concentrations, and thus, water quality guidelines (WQG) for Ca and Mg are warranted. However, Ca can modify Mg toxicity in Ca-poor water and additional interactions with other major ions (Na⁺, K⁺, HCO₃⁻/CO₃²⁻, SO₄²⁻ and Cl⁻) may occur, potentially obscuring the water hardness–effect relationship. In a meta-analysis of toxicological studies, we: (i) evaluate the performance of three WQG derivation methods, and (ii) determine the influence of several variables (acute/chronic data, anions, Ca:Mg ratios, non-geographically relevant species) on the models. We find that the most sensitive species- or species sensitivity distribution (SSD)-based WQG derivation methods greatly overestimate water hardness toxicity, particularly if non-resident species are included. Broad-scale implementation of most sensitive species- or SSD-based WQG is impractical because water hardness varies beyond and within the regional scale. Anion type does not affect water hardness toxicity across species, but the Ca : Mg ratio is toxicologically relevant, underscoring the importance of considering ion ratios when developing major ion WQG. Although data supporting formal water hardness WQG are unavailable, we suggest using a two-component background condition approach that supports simultaneous management of water hardness and Ca : Mg ratio, and WQG that are applicable beyond the regional scale.

This article is part of the theme issue ‘Salt in freshwaters: causes, ecological consequences and future prospects’.

1. Background

Changes in both the salinity (total ionic content) and ionic balance of water can be toxic to freshwater organisms [1–3]. With the salinization of freshwaters increasing on a global scale, a call has been made to develop ion-specific water quality guidelines (WQG) for all major ions in water (Ca²⁺, Mg²⁺, Na⁺, K⁺, HCO₃⁻/CO₃²⁻, SO₄²⁻ and Cl⁻) [3–6]. Yet, few major ion WQG currently exists for the protection of freshwater aquatic life (i.e. only SO₄²⁻ and Cl⁻) [7,8]. Relative to other major ions, the toxicity of Ca and Mg has largely been overlooked, despite these ions being present in relatively large quantities in wastewater or run-off owing to a variety of anthropogenic activities [9–14]. Instead, Ca and Mg have predominantly been assessed for their capacity to ameliorate trace metal toxicity [15–18]. While WQG exists for the protection of agricultural water uses, e.g. for livestock watering in Canada [19] at 1000 mg l⁻¹ Ca, no WQG for the protection of aquatic life exists for Ca and Mg. However, the current lack of protection from Ca and Mg does not indicate a lack of importance from a biological or environmental management perspective: the health and population dynamics of freshwater zooplankton in aquatic ecosystems, including the keystone species, Daphnia spp., has been linked to Ca availability [20,21] and potentially Ca excess [22,23]. The survival of salmonid fish early life stages is negatively affected by increased Ca concentrations [24], which could put affected fish populations at risk. The biological roles of Ca and Mg are connected—Ca homeostasis is linked to Mg supply [25] and Mg functions as a neuromuscular relaxant, opposing the action of Ca [26]. The toxicity of Ca and Mg are also interlinked, because biota in waters of low ionic content can be extremely sensitive to small increases in Mg without additional Ca [27]. Moreover, changes in Ca and Mg concentrations in freshwater may cause indirect effects on aquatic ecosystems. For example, the precipitation of Ca, as calcite (CaCO₃), drives nutrient (P) cycling in headwater streams [28]. Thus, it is important to understand the limits to which Ca and/or Mg concentrations can be altered in aquatic ecosystems before having deleterious effects on biota and ecosystem function and to develop appropriate WQG to protect freshwater aquatic life.

While both Ca and Mg are essential elements, in excess quantities they can be more toxic to freshwater species than other major ions co-present at high concentrations. Relative to water containing 34 mg l⁻¹ Ca, rainbow trout (Oncorhynchus mykiss) embryo survival to the eyed-stage of development was decreased by 71–91% if eggs were exposed to water containing ≥520 mg l⁻¹ Ca, immediately post-fertilization [24]. Toxicity owing to an elevated SO₄ concentration in the test water was definitively ruled out [24]. Rainbow trout hatchlings exposed for 25 days to 486, 848 and 1696 mg l⁻¹ Ca (as CaCl₂) exhibited a significant decrease in length (5%, 8% and 11%, respectively), and in mass (16%, 27% and 31%, respectively), relative to control fish [9]. While inhibitory effects owing to Cl⁻ exposure could not be completely ruled out, NaCl only showed an inhibitory effect at the highest (Cl⁻ equivalent) concentration tested, and MgCl₂ showed no effect at all [9]. By contrast, the 96 h 10% inhibitory concentration (IC₁₀) for Lemna aequinoctialis growth rate was 1.9 mg l⁻¹ Mg and the 96 h IC₁₀ for Hydra viridissima population growth rate was 1.8 mg l⁻¹ Mg [27]. Magnesium chloride and MgSO₄ were 18 and 43 times more toxic than Na₂SO₄ to H. viridissima (96 h IC₅₀, population growth rate), respectively, and 49 and 75 times more toxic than Na₂SO₄ to L. aequinoctialis (96 h IC₅₀ for growth rate), respectively [27]. Interestingly, Ca and Mg toxicity are related. In Ca-poor water, Ca decreased Mg toxicity but at higher concentrations, Ca and Mg were additively toxic [1,2,27,29]. Therefore, from a toxicological, and hence a management perspective, it is important to consider both the combined concentrations of Ca and Mg in water (i.e. water hardness) and the balance of Ca and Mg (i.e. Ca : Mg ratio).

Differences in water hardness and the Ca : Mg ratio between aquatic ecosystems is linked to natural variations in geological composition and ease of weathering (as reviewed by Wetzel [30]). Common, easily weathered, geological sources of Ca²⁺ and Mg²⁺ include limestone (CaCO₃), dolostone (CaMg(CO₃)₂), gypsum (CaSO₄·2H₂O) and magnesite (MgCO₃), within which the relative abundance of Ca, Mg and other elements vary [31–33]. However, resource extraction and use can greatly increase the rate of weathering of these materials and their subsequent deposition into freshwater ecosystems, modifying the water hardness. As such, anthropogenic activities, including mining [10], oil and gas extraction [11,12], agriculture [13] and road salt use [9,14], have the potential to profoundly increase water hardness in aquatic ecosystems and beyond the toxicity threshold of freshwater species.

However, owing to the different natural background concentrations of ions present (from underlying geology), it is likely that a mixture of ions resulting from anthropogenic activities will be released into the environment rather than Ca and Mg alone, presenting an interesting challenge for aquatic ecotoxicologists and policy makers alike—i.e. how does one manage Ca and Mg within a mixture of ions when multiple ion interactions may occur, which can affect toxicity [1,2,34–36]), and the ion mixture composition changes widely within and beyond the regional scale (reviewed in Mount [1]). One potential solution is to use conductivity as a proxy for toxicity of the mixture of interest, as has been successfully implemented via the development of a field-based water quality benchmark by the United States Environmental Protection Agency (USEPA) [37] for a mixture of Ca²⁺, Mg²⁺, SO₄²⁻ and HCO₃⁻. However, this conductivity-based benchmark is regionally specific, meaning that it has limited flexibility in its current form to be applied to different or variable ion mixtures [37]. For example, the benchmark criterion is below background condition in other jurisdictions, including Western Canada [10]. Moreover, using conductivity (a measure of total salinity) as a proxy for the toxicity of ion mixtures would permit changes in the balance of ions. While a conductivity benchmark may be appropriate for application within a specific region and the mixture from which it was developed, its application as a universal management approach in all jurisdictions, or to manage water hardness across a broad scale is impractical.

Thus, it is critical to understand whether the effects of select major ions (here, Ca²⁺ and Mg²⁺) present in a major ion mixture can be isolated. That is, whether the water hardness–effect relationship remains clear on a broad (>regional) scale, despite the aforementioned interactions and potential limitations, or whether alternate approaches to policy development are required. For example, site-specific modelling approaches could increase the accuracy and precision of the estimate of background condition [38,39], but the extremely large variation in measured water hardness in some management regions (such as in British Columbia (BC), Canada, shown herein) could make province-wide WQG difficult to derive. Therefore, in a meta-analysis of toxicological studies on water hardness, we applied WQG derivation/analysis techniques to better understand the challenges and future prospects for deriving a WQG for water hardness. Specifically, we: (i) evaluated the performance of three WQG derivation methods (most sensitive species, SSD and a novel, two-component, per cent change from the background condition approach), and (ii) determined the effects of several factors (acute/chronic data, anions, Ca : Mg ratio and non-geographically relevant species) on the models. It is expected that the ideas presented herein for Ca and Mg will be applicable to other major ions and our results will be of particular interest to ecotoxicologists and policy-makers alike.

2. Material and methods

(a). Water hardness and Ca : Mg ratio—definitions, units used

In this study, water hardness was defined according to the APHA [40] definition:

2.1

where Ca and Mg represent their respective dissolved concentrations, typically measured by either titrimetric or spectroscopic methods [40,41], and water hardness is expressed in mg equivalents of calcium carbonate. Since water hardness is a combination of both Ca and Mg concentrations, the ratio of Ca : Mg can differ even though water hardness remains the same. All Ca : Mg ratios presented herein were calculated based on mass (mg l⁻¹).

(b). Background condition in the exemplar region

For this study, we used the province of BC, Canada, as an exemplar region of interest. Underlying geology in the province offers a mixture of igneous, metamorphic, and sedimentary rocks and minerals [31,33]. These materials contain various amounts of Ca, Mg and other elements of economic importance to resource-related activities, occurring throughout BC [31,33]. From an agricultural perspective, BC accounts for 24% of all irrigated farm land in Canada, with most irrigation occurring in semi-arid mountain valleys [19]. In winter, road salts and de-icers are applied at a rate of approximately 140 000 tonnes annually to provincial roads [42]. However, with BC having less than 2% of the total share of the Canadian application rates of salts and de-icers containing Ca and/or Mg [42], BC applies less than 2800 tonnes per year of Ca- and Mg-based salts and de-icers to its roads. Water hardness and the Ca : Mg ratio of unaffected surface waters varies at the provincial, regional and local levels (figure 1; electronic supplementary material, table S1). Data from 546 provincial monitoring stations (1996–2017), reporting both dissolved Ca and dissolved Mg concentrations, were used to calculate background water conditions in BC (figure 1a; only samples with detected concentrations of Ca and Mg were used and the dataset was visually inspected to remove samples that were considered affected by human activities). The resultant mean water hardness in BC was calculated as 70.1 mg l⁻¹ as CaCO₃ (range 2–781 mg l⁻¹ as CaCO₃; figure 1b) and the mean Ca : Mg ratio as 5.9 (range 0.1–36.8; figure 1c). Broad ranges in water hardness, geology and the influence of anthropogenic activities in BC make it potentially challenging to implement WQG for water hardness, making it an excellent exemplar region for use in this study.

Figure 1.

Water hardness of unpolluted surface waters across BC, by site (a) and by region (b), and related Ca : Mg ratios (c) (n = 546 stations). Grey lines subdividing the province in (a) indicate water management regions. Inset Canada map is a modification of an image within the public domain (https://commons.wikimedia.org/wiki/File:British_Columbia,_Canada.svg).

(c). Literature collection

Peer-reviewed research articles and government reports (collectively, articles) were collected using keyword searches in Web of Science, ECOTOX, Google Scholar and several individual publishing houses (Science Direct, Wiley, PubMed). Owing to the availability of material, Research Gate was also used to acquire articles. The PRISMA checklist, a set of standardized reporting requirements for meta-analyses, was followed where possible [43]. Keywords used were: calcium, Ca, magnesium, Mg, Ca : Mg, ratio, water hardness, salinization, freshwater, salt, concentration, CaCl₂, MgCl₂, CaSO₄, MgSO₄, calcite, CaCO₃, lake whiting, road de-icer, major ion, cation, anion, ion, toxicity, conductivity, reproduction, growth, algae, plant, macrophyte, zooplankton, mayfly, caddisfly, chironomid, scud, snail, gastropod, clam, mussel, crayfish, amphibian, toad, frog, fish, egg, fathead minnow, perch, rainbow trout, brook trout, brown trout, salmonid, Diptera, Ephemeropera, Plecoptera, Trichoptera, Baetidae, Gammaridae, Gammarus, Hyalella, Daphnia, O. mykiss and combinations thereof. Reference lists of collected articles were also mined for additional articles of interest. Articles of interest inaccessible online were requested through interlibrary loans or by contacting researchers via email or Research Gate. To ensure operator bias was limited during searching, searches were conducted by three individuals and non-duplicated results were combined. No restriction on publication date was used but only articles written in English were considered. Literature acquisition was completed in early 2017.

(d). Data screening, eligibility and inclusion

Collected articles (approx. 300, listed in the electronic supplementary material, list S1) were preliminarily screened to determine if they contained WQG-relevant information. To be retained, articles had to test aqueous exposures of Ca and/or Mg on freshwater species only, could not report water hardness as a toxicity modifier of, e.g. metal toxicity, and tests had to be completed under temperature and pH conditions not considered extreme relative to the organism studied (to limit effects of covariates/extreme conditions). Suitable effect endpoints [44] included: no-observed-effect concentration (NOEC), no-observed-effect level (NOEL), no-observed-adverse-effect level (NOAEL), no-observed-adverse-effect concentration (NOAEC), lowest-observed-effect concentration (LOEC), lowest-observed-effect level (LOEL), lowest-observed-adverse-effect level (LOAEL), lowest-observed-adverse-effect concentration (LOAEC), effective concentration (ECxx), inhibitory concentration (ICxx), maximum acceptable toxicant concentration (MATC), lethal concentration (LCxx) and median tolerance limit (TLm). Only 48 collected articles (16% of total) were considered for further evaluation because only these articles explicitly tested water hardness toxicity while containing WQG-relevant endpoints.

Articles retained after preliminary screening were evaluated against provincial criteria used for water quality guideline derivation, which classified articles as either of primary or secondary quality (acceptable for inclusion), or unusable [44]. We enlisted the following modifications/clarifications to the classification criteria for use in this study: chemical purity of the substance tested had to be a minimum of American Chemical Society reagent grade. Pseudoreplicated studies were always rejected. A minimum of ≥3 replicates was required for primary status or secondary status. Very infrequently, an exception was made for n = 2 replicates for an article to maintain secondary status, if an extremely large number of solutions were tested, and it was considered impractical to test with more replication (e.g. approx. 1500 treatments across 149 tests [1]), providing the remainder of the eligibility criteria were met. Owing to the extreme limitation in the number of potentially acceptable studies available, and limited complete reporting of water quality variables, a compromise was considered. While, to achieve primary status, articles were required to report all water quality variables (alkalinity, pH, dissolved oxygen, temperature, conductivity and water hardness (or Ca and Mg ion concentrations)), articles were classified as secondary if a minimum of 4 out of 6 water quality variables were reported, including water hardness. Articles not meeting this standard were rejected. Since reporting of all major ion concentrations was rare, doing so was considered of interest but not required. Many articles did not report the concentration of both Ca and Mg in test waters because typically only one related salt was tested. Still, studies were considered secondary sources if the untested hardness cation was measured in dilution water or the Ca and Mg concentrations could be back-calculated from the salts used. Classification criteria that were used without modification from the classification protocol and a list of additional abiotic variables required for an article to achieve secondary status are not repeated here but rather are listed in the electronic supplementary material, list S1.

Post-classification, no articles were identified as primary sources and seven articles (2% of total) were identified as secondary (see the electronic supplementary material). These secondary sources often had several data points that were unusable and such points were discarded. The predominant reasons for rejecting articles, or data points, during classification were a lack of water quality data, no analytical confirmation of the treatments tested, poor statistical replication and the effect being outside of the tested concentration range. In the end, four of seven classified articles did not report conductivity; of those, one study did not report alkalinity but reported bicarbonate concentration instead, and all other articles reported all water quality variables. A total of 147 effect endpoints were identified during classification with only 112 being appropriate for WQG development. That is, 35 endpoints were discarded because they were either in the form of a median lethal time (LT50), which was not directly comparable to concentration-based endpoints (e.g. LCxx), or the concentration of the toxicant was altered during the test, which is unacceptable for guideline derivation [44]. However, these 35 discarded data points were still useful for investigating the potential for environmental effects owing to pulse exposures of Mg and due to rapid, extensive calcite precipitation [29,34].

Data were extracted manually from retained literature and input into an Excel worksheet (Microsoft, USA). Desired data items were: taxonomic group (algae/aquatic plant, invert, amphibian and fish), common and scientific names of species tested, test length, exposure type (acute or chronic; where acute was generally defined as ≤96 h but this definition was adjusted based on the lifespan of the animal, as guided by standardized test methods and guideline derivation procedures [45,46]), effect type (lethal and sublethal), effect level (e.g. ICxx), endpoint type (mortality, growth and reproduction), contaminant name (e.g. MgSO₄·7H₂O), cation tested (Ca, Mg or both), anion(s) of salts used to modify hardness (e.g. SO₄²⁻ and Cl⁻), the water hardness effect concentration (in mg l⁻¹ as CaCO₃), Ca and Mg effect concentrations (in mg l⁻¹), Ca : Mg or Mg : Ca ratio, alkalinity, pH, temperature, conductivity, conductivity at the water hardness effect concentration, concentrations of Na⁺, K⁺, HCO₃⁻, SO₄²⁻ and Cl⁻, background/control concentrations of Ca, Mg, Ca:Mg ratio and water hardness, whether time was a covariate of effect and the article reference details. The dataset was reviewed for errors by two individuals (all data are in the electronic supplementary material, data S1).

(e). Calculations and statistical analyses

Several unit conversions were completed to provide a standardized reference point from which to assess toxicity and data relationships. All reported effect concentrations based on, e.g. salts or ions, were converted to their respective water hardness effect concentrations in mg l⁻¹ as CaCO₃. Likewise, if possible, the concentration of cations and anions was calculated using the amount and type of salts used in tests. Where possible, Ca and Mg concentrations were used to calculate the Ca : Mg ratio. Data conversions were conducted in Excel (Microsoft, 2016).

Statistical analyses were conducted in R [47] v.3.3.2 using RStudio [48] v.1.0.136. Complex graphs were plotted using the R visualization package ggplot2 [49]. Data were assessed for normality using boxplots or qqnorm plots and for homogeneity of variance using the Bartlett's test. Where required, a log₁₀ transformation of the data was used to rectify deviance from the normal distribution. A Welch's two-sample t-test was used to determine if the mean effect between acute and chronic exposures significantly differed. A value of α of 0.05 was used in all tests.

Statistical evaluation of potential toxicity modifiers (including major ions) was not supported because their concentrations/measurements were often reported only in control/dilution water. In result, for example, anion concentrations were often the same across several data points, obscuring potential relationships. Hence, a visual assessment of potential anion influence and the Ca : Mg ratio was made instead. First, the Ca and Mg effect concentrations were converted from mg l⁻¹ to mM concentrations. These mM concentrations of Ca and Mg were plotted against one another in a 1 : 1 plot and the resultant position of the effect data points in the plot represents the relative contribution of Ca and Mg atoms to water hardness toxicity. Data points were then coloured to represent the anions of the salts used to modify water hardness during toxicity testing, and a visual inspection for clustering of data points by anion type was completed.

(f). Water quality guideline approaches

In this paper, we only applied WQG derivation/analysis techniques as a form of meta-analysis to better understand the challenges and future prospects for deriving a WQG for water hardness. Owing to the limitations described throughout, we did not generate formal WQG.

(i). Most sensitive species approach

The most sensitive species approach to WQG derivation is used in BC [44]. Using the classified dataset of toxicological effect endpoints, this approach selects the species datum exhibiting the lowest effect concentration and uses it to define the WQG [44]. Depending on the level of confidence in the data, the effect concentration is commonly divided by an uncertainty factor of 2–10 prior to being used as the WQG [44]. Herein, the most sensitive species was determined while both using the entire dataset and, to determine the effect of non-geographically relevant species, after constraining the dataset to BC-resident species. If acute endpoints were selected, an additional evaluation was made by constraining the dataset by chronic exposure length to determine if exposure duration affected the outcome. No uncertainty factors were considered.

(ii). Species sensitivity distribution approach

As applied to the current WQG work, the species sensitivity distribution (SSD) is a statistical model that describes the variation in the sensitivity of different species to a contaminant of concern (here, water hardness toxicity) [46,50,51]. In short, a dataset of relevant effect endpoints is assembled from toxicity tests on several (preferably keystone or otherwise ecologically relevant) species. The most sensitive effect endpoints for each species are selected for inclusion in the model, according to a ranked selection procedure. The selected endpoints are plotted as the proportion of species affected versus the log concentration of the contaminant. A log-probit (or other model) is then fitted to the selected points. The hazardous concentration for five per cent of species (HC5) is calculated using the model and this value is used to define the WQG. As with the most sensitive species approach, the resultant HC5 value can be divided by an uncertainty factor to improve the level of protection offered by the resultant WQG, but this procedure is not conducted by all jurisdictions [46,51].

To generate each SSD, a log-probit model SSD calculator was used (SSD Generator v.1) [50]. Additional model fits were not attempted because the dataset was not of sufficient quality or quantity to derive formal WQG. That is, the chronic dataset did not fulfil the minimum requirements for creating an SSD-based WQG (i.e. type A WQG [46]), owing to a lack of comprehensive species data. To enable SSD creation, we modified the data ranking order used to derive type A WQG to permit inclusion of LC50 data (i.e. acute data), thereby permitting a sufficient number of data points to permit SSD creation. Moreover, there was no significant difference in effect between acute and chronic data groups (see Results and Discussion, §3a(i)). The data ranking order used to include data into the SSD was as follows, in the order of most to least preferred endpoint [46]: the most appropriate ECx/ICx representing a low-effects threshold > EC_15–25/IC_15–25 > LOEC > MATC > EC_26–49/IC_26–49 > nonlethalEC₅₀/IC₅₀ > LC₅₀. Three SSDs were then constructed, using water hardness as the stressor, after combining acute and chronic datasets. One SSD considered all data, the second was constrained to BC-resident species only (to determine the effect of species on the first result) and the third was constrained to BC-resident species endpoints where the Ca : Mg ratio was greater than 1 (to simulate BC freshwaters). As an additional QA/QC check, a fourth SSD was generated that contained no acute data points to see if results in the first SSD were altered by the presence of acute-based data. No SSDs were created with a Ca : Mg ratio as the stressor because the resultant SSDs for water hardness were not useful. Similar endpoints were combined, where appropriate, by calculating the geometric means of data points [46], but only within studies to protect against uncertainty in Ca and Mg toxicity by influence of changes in background water chemistry between tests [1]. For each SSD, the HC5 and its 95% confidence interval were calculated. No uncertainty factors were considered.

(iii). Two-component, per cent change from the background condition approach

The two-component, per cent change from the background condition approach to WQG development represents an idea for future policy writers and environmental managers to consider. This approach calculates WQG for two components—water hardness and the Ca : Mg ratio—to permit management of both the total concentration of water hardness as well as the cation balance. Briefly, the per cent change from background condition is first calculated for both components. The smallest resultant per cent change causing an effect is identified and this threshold value is applied to the site-specific background concentration, via the equations below, to define the WQG. These WQG values describe the maximum amount of change permitted for water hardness and the Ca : Mg ratio. If mortality-based effect data are used to calculate WQG values, it is expected that the resultant WQG will describe at what per cent change in water hardness/Ca : Mg ratio negative effects will or are likely to occur.

Specific derivation of WQG using the two-component approach was as follows, while considering all classified data: first, the per cent change from background was calculated for each effect endpoint for both water hardness (in mg l⁻¹ as CaCO₃) and the (mass-based) Ca : Mg ratio as follows:

2.2

where background condition meant control/dilution water in laboratory tests (as the laboratory equivalent to reference condition in the field). Next, the most sensitive species datum was identified for: (i) an increase in water hardness, (ii) an increase in the Ca : Mg ratio, and (iii) a decrease in the Ca : Mg ratio (because an increase in water hardness can result in either an increase or decrease in the Ca : Mg ratio). The calculated per cent change for each selected datum indicated the maximum per cent change permitted for water hardness and the Ca : Mg ratio (i.e. a threshold for effects used as the per cent change WQG). These per cent change WQG can then be applied to site-specific, background water hardness and Ca : Mg ratio to calculate local WQG values as follows:

2.3

and

2.4

where Hss is the mean site-specific water hardness in mg l⁻¹ as CaCO₃ and Rss is the mean site-specific Ca : Mg ratio (mass-based). Note that the absolute value of the per cent change WQG is used in the maximum decrease equation. We calculated these ‘site’-specific WQG values using the mean provincial and mean regional water hardness and Ca : Mg ratios, from data presented in figure 1b,c (electronic supplementary material, table S1). Means were used instead of maximum values to determine the ‘site’-specific WQG because the mean is more representative of typical conditions. That is, unlike a maximum value, a mean is a measure of central tendency meaning that it is affected by all measurements of water hardness or Ca : Mg ratio within a management boundary, resulting in increased protection of biota in ecosystems with naturally low water hardness or Ca : Mg ratio. The resultant WQG values were compared to the range in background conditions in BC, both at the regional and provincial levels, to determine if the WQG were inside or outside of this range. An example of how to calculate local WQG values using the two-component approach can be found in the electronic supplementary material, Calcs S1.

3. Results and discussion

(a). Data evaluations and limitations

(i). Acute versus chronic toxicity—different questions but similar results

Within the screened dataset, there was no significant difference in mean effect between acute and chronic exposure groups (t_37.9 = 1.57, p = 0.126; figure 2). In addition, several critical data gaps were identified including a lack of plant and cold water fish data in the acute exposure group and a lack of any fish data in the chronic exposure group (figure 2). No acceptable aquatic insect or amphibian data were found, although these are not required, but highly desirable, for WQG development both provincially [44] and federally [46]. Based on these results, separate acute and chronic WQG were not supported [44], and even if they were supported, no formal WQG could be developed owing to critical gaps in the current dataset. Since there was no significant difference between exposure groups, acute and chronic data were combined for all subsequent comparisons, which enabled a broader diversity of taxa to be evaluated. Nevertheless, additional acute versus chronic data evaluations were conducted for the SSD approach and the most sensitive species approach to provide additional support for merging the exposure groups (see §§3b(i) and (ii)).

Figure 2.

Log-transformed, water hardness effect concentration plotted by acute and chronic exposure groups (boxplots, p = 0.126) and by taxonomic group (point colour). Effects are from different endpoints in acute and chronic groups (see the electronic supplementary material). Note the lack of cold water fish data.

As more data become available, there may be a statistically significant difference in effect across acute and chronic exposures. Currently, there was only one species represented in both exposure groups (Daphnia magna), meaning that there was a nearly complete difference in species assortments within the acute and chronic exposure groups. It is possible that these different species assortments currently have overlapping, groupwise sensitivities to water hardness toxicity simply because within-species differences between groups are not yet represented. For D. magna, for example, there was a difference between acute and chronic effects (acute mean 1952 ± 95% confidence interval (CI) range of 1423–2482 (n = 5) versus chronic mean 538 ± 95% CI range of 388–688 (n = 8)). On the other hand, it is also plausible that the overlap in sensitivities across exposure groups will remain even after adding more data because the concentration range over which water hardness toxicity occurs is extremely broad (water hardness range in acute exposures: 50–17 800 mg l⁻¹ as CaCO₃; chronic exposures: 8–26 400 mg l⁻¹ as CaCO₃), potentially obscuring any within-species influences on effect.

(ii). Anions and the Ca : Mg ratio as potential toxicity modifiers

A visual assessment of the effect of anion type and Ca : Mg ratio on toxicity (figure 3) suggested that the Ca : Mg ratio may be more toxicologically relevant than anion type. That is, more data points plotted on the Mg side of the 1 : 1 line and data points did not cluster by anion type. Future studies of the toxicity of water hardness should report—in all treatments and controls—the concentrations of all major ions (Ca²⁺, Mg²⁺, Na⁺, K⁺, HCO₃⁻/CO₃²⁻, SO₄²⁻ and Cl⁻), as well as dissolved organic carbon, pH, alkalinity and conductivity such that additional evaluation of potential toxicity modifiers can be conducted.

Figure 3.

Relative contribution of Mg and Ca to water hardness toxicity in mM (line defines equal contribution; see the text for details and the electronic supplementary material for data). Note points do not cluster by anion type (colour) or taxonomic group (shape), and most points lie on the Mg side of the 1 : 1 line.

Although it appears the Ca : Mg ratio is important (figure 3), it could also be that more data points plotted on the Mg side of the equal contribution line simply because Mg toxicity was evaluated more often than Ca toxicity. Of the 112 effect endpoints used, 89 were attributed to toxicity tests that studied Mg toxicity when compared with 22 for Ca (one additional data point described the combined toxicity of Ca and Mg). Nevertheless, several other studies have reported the Ca : Mg ratio as toxicologically relevant [2,27], suggesting that our results simply confirm at the broad scale what has been previously reported for individual species and local species groups. More broadly, recent research has underscored the importance of considering ionic balance rather than just total salinity from toxicological and environmental management perspectives [1,2,4,9,27,34]. For these reasons, to evaluate whether water hardness can be used to describe toxic effect, we considered the total contribution of Ca and Mg to toxicity (as water hardness) as well as their balance (as the Ca : Mg ratio).

(b). Describing environmental effect: options, issues and future prospects

(i). Most sensitive species approach

The most sensitive species approach resulted in WQG that were not applicable to BC. For water hardness, the most sensitive species in the entire dataset was H. viridissima represented by a 96 h IC₁₀ for population growth of 8 mg l⁻¹ as CaCO₃ water hardness [27], even before considering an uncertainty factor. This result falsely suggests that nearly all surface waters in BC (figure 1) are unsafe for aquatic organisms. When only BC-resident species were considered, Ceriodaphnia dubia was the most sensitive to water hardness (48 h LC₅₀ of 83 mg l⁻¹ as CaCO₃, an acute effect [1]). If data were constrained to BC-resident species while considering chronic effects only, Chlorella sp. was the most sensitive to water hardness (72 h IC₁₀ for growth, 178 mg l⁻¹ as CaCO₃ hardness [27]); however, this effect was still within background condition in BC (figure 1). The large background variation in water hardness across BC (figure 1a) would probably not permit the most sensitive species approach to be implemented, even if more data were available. The most sensitive species assessment was not completed using the Ca : Mg ratio as the stressor because water hardness was not a good descriptor of toxic effects when the most sensitive species approach was used.

(ii). Species sensitivity distribution approach

Similar to the most sensitive species approach, the SSD approach could not be used to describe toxicity owing to water hardness because the resultant HC5 values were all below background condition in BC (figure 4). When all data were considered, the HC5 for water hardness was 4.4. mg l⁻¹ as CaCO₃ (95% CI of 2.0–9.8 mg l⁻¹ as CaCO₃; figure 4a). Censoring the data to consider only BC-resident species increased the HC5 estimate; however, the censored model was less reliable, as indicated by the increased breadth of the 95% CI (figure 4c relative to 4a). Similarly, if the SSD model only included data on BC-resident species that had a Ca : Mg ratio greater than 1, the HC5 and model instability both further increased (figure 4d). This increased model instability was owing to a decrease in the number of endpoints represented in the SSD model as the dataset was further constrained, i.e. from 10 points down to 3. Both phenomena—increasing the relevance of the SSD model by using geographically relevant species and decreasing model stability as the number of data points are reduced—are known [51]. In addition to ruling out the SSD approach to generate WQG for water hardness, we have determined that BC-resident species (or North American species) are largely under-represented across the entire dataset. Most species represented in the dataset are residents of Australia [27,29]. Since water hardness was not a good descriptor of toxic effect when the SSD approach was used, SSDs were not constructed for the Ca : Mg ratio as the stressor. All data points used in all SSDs are presented in the electronic supplementary material, tables S2 and table S3.

Figure 4.

Species sensitivity distributions for water hardness using (a) all classified data, (b) only chronic exposure data, (c) all BC-resident species data, and (d) all BC-resident species data with a Ca : Mg ratio greater than 1. Hazard concentrations for 5% of species (HC5, in mg l⁻¹ as CaCO₃) and their 95% confidence intervals (95% CI, in mg l⁻¹ as CaCO₃) are also reported. Legends in (a) apply to all graphs. (Online version in colour.)

It is uncommon to merge acute and chronic data together for SSD analyses [51]. Nevertheless, the HC5 in figure 4b, where the data were constrained to include only chronic exposure data (HC5 2.2 mg l⁻¹ as CaCO₃), is similar to that in figure 4a, where the entire dataset was considered (HC5 4.4 mg l⁻¹ as CaCO₃). In comparing the 95% CIs, the combined model containing both acute and chronic data had a tighter model fit (acute + chronic SSD 95% CI 2.0–9.8 mg l⁻¹ as CaCO₃ water hardness versus chronic-only SSD 95% CI 0.1–48 mg l⁻¹ as CaCO₃ hardness).

(iii). Two-component, per cent change from the background condition approach

A two-component, per cent background condition approach to manage water hardness and the Ca : Mg ratio was attempted and WQG values were both calculable and could provide a preliminary basis, approach-wise, with which to derive formal WQG for water hardness once data become available. The equivalent range in water hardness effect concentrations associated with the calculated per cent change values was 8–26 408 mg l⁻¹ as CaCO₃. The median per cent change occurred at 2029 mg l⁻¹ as CaCO₃ water hardness. The smallest per cent change in water hardness causing a negative effect was a 78% increase over background (H. viridissima 96 h IC₁₀ for population growth) [27]. However, when the data were constrained to consider BC-resident species only, the most sensitive species was C. dubia for which the 48 h LC₅₀ was reached after a 316% increase in water hardness [1]. Both endpoints were used to calculate water hardness-based WQG for the provincial and regional water quality in BC (table 1).

Table 1.

Resultant WQG for water hardness and Ca : Mg ratio, relative to background condition at provincial and select regional scales (see the text for details; full dataset in the electronic supplementary material, table S4). (WQG were inside (I), outside (O) or partially overlapped (O–I) the range in background condition.)

				site WQG and evaluation results
	background condition			all data		BC-resident spp.
site	n	mean	range	WQGa	I/O	WQGb	I/O
water hardness (mg l−1 as CaCO3)
regional
Cariboo	99	97.6	5.2–569	174	I	406	I
Kootenay	40	85.7	7.3–305	153	I	357	O
Peace	4	65.9	34.8–95.3	117	O	274	O
Skeena	142	42.4	2.0–217	75	I	176	I
provincial	546	70.1	2.0–781	125	I	292	I
Ca : Mg ratio
regional
Cariboo	99	4.7	0.1–15.6	2.1–3.8	I	1.7–3.8	I
Kootenay	40	5.8	2.6–17.5	2.6–4.6	I	2.1–4.6	O–I
Peace	4	3.5	2.0–5.0	1.5–2.8	O–I	1.3–2.8	O–I
Skeena	142	6.2	1.2–31.7	2.7–5.0	I	2.2–5.0	I
provincial	546	5.9	0.1–36.8	2.6–4.7	I	2.1–4.7	I

^aWQG was a 78% increase in hardness; WQG for Ca : Mg ratio was a 56% decrease and a 20% increase.

^bWQG was a 316% increase in hardness; WQG for Ca : Mg ratio was a 64% decrease and a 20% increase.

When calculating the per cent change from background condition for the Ca : Mg ratio, it was noted that toxicity was more often owing to a decrease in the Ca : Mg ratio from background condition rather than an increase (median Ca : Mg ratio of 1.2 at background versus 0.07 at experimental effect, suggesting that Mg was more toxic than Ca). This result is in alignment with that of figure 3, where the Ca : Mg ratio appears toxicologically relevant, and is supported by other researchers [1,2,27,29].

The effect-based Ca : Mg ratio equivalent to the per cent change in the dataset ranged from 0.0002 to 414 across the entire dataset. If all data were considered, the smallest decrease in the Ca : Mg ratio causing detrimental effects was 56% (H. viridissima, 96 h IC₁₀ for population growth [27]) and the smallest increase causing detrimental effects was 20% (C. dubia, 48 h LC₅₀; also a BC-resident species) [52]. When only species present in BC were evaluated, the smallest decrease in the Ca : Mg ratio causing detrimental effects was 64% (Chlorella sp., 72 h IC₅₀ for growth [29]). These three endpoints were used to calculate Ca : Mg ratio-based WQG for the provincial and regional water quality in BC (table 1).

Only WQG results for representative regions are displayed in table 1 to show calculated WQG values within the context of the ranges and extremes in water hardness and Ca : Mg ratio occurring in BC, the relative predominance of the WQG being inside, outside or partially overlapping the related ranges in background conditions and to show a diversity in the number of stations (n) used to calculate background conditions (for ease of comparison, all regions are shown in the electronic supplementary material, table S4). It appears that our per cent change from background condition approach is at least a good starting point with which to guide future WQG derivation for water hardness and the Ca : Mg ratio. That is, in comparison with the other approaches, this approach was the best descriptor of where toxic effects may occur because it did not produce WQG values well below background conditions, especially if BC-resident species were used (table 1). However, additional methods of WQG derivation may exist that were not considered in this review. Owing to the current lack of appropriate ecological data, it is not yet possible to interpret the ecological significance of our derived WQG values. Still, the potential applicability of our per cent change from background condition approach on a larger than regional basis is a distinct advantage over other regional and site-specific approaches [27,29,37].

The interpretation of regional WQG as being within or outside the range in background conditions was probably affected by an inadequate number of monitoring stations being used (lowest n = 4) to describe the background water hardness of the region. More data must be used to estimate background conditions when using the per cent change from the background condition approach to ensure reference conditions are appropriately represented (either by measurement, as used here, or by modelling [38]). Based on this assessment, we caution that our approach requires more rigorous evaluation as more data become available because it would seem that a fault with this approach is that it is sensitive to limitations in data quantity. That is, with more comprehensive background datasets that report Ca and Mg ion concentrations, the WQG may possibly be further inside the background condition range than observed herein.

It would appear that, on a preliminary basis, our per cent change from background condition approach should be investigated further. For example, the WQG require validation within field sites, that is, the WQG should be compared to the natural variation in water hardness and the Ca : Mg ratio on a site-specific basis. Based on the current level of success of our proposed method, we suggest that future policy writers and environmental managers consider the above described, two-component, per cent change from the background condition approach to derive a two-component WQG for water hardness and the Ca : Mg ratio.

4. Conclusion

We find that the traditional, most sensitive species- or SSD-based WQG derivation methods greatly overestimated toxic effect when water hardness was used as the contaminant of concern. The presence of non-geographically relevant species amplified the overestimation, as did the inclusion of data with extremely low Ca : Mg ratios that are not present within the exemplar region. Anion type did not affect water hardness toxicity across species but the Ca : Mg ratio remained toxicologically relevant, underscoring the importance of considering major ion ratios when developing WQG. Interestingly, there was no significant difference in mean effect between acute and chronic exposure groups. In fact, data could be used interchangeably in SSDs with little to no effect on HC5 estimates. Still, this conclusion may also be owing to the limited amount of data available to build the models. Nevertheless, broad-scale implementation of most sensitive species- and SSD-based WQG was impractical because water hardness varied greatly beyond and within the regional scale.

It is possible that the best approach to WQG for major ions is a model that accounts for all major ion concentrations, ion balance and multiple toxicity interactions instead of using a single-ion approach. However, to produce such a model will require much better understanding of major ion toxicity from mechanistic, physiological and ecological perspectives. Indeed, a general lack of appropriate data appears to be the largest hurdle that researchers and policy-makers will need to overcome. Particularly evident data gaps were a lack of cold water fish data and data on species resident in the exemplar region of interest (and North America at large). It is critical to understand the mechanisms of toxicity of individual ions and their mixtures to identify what proportion of salinity toxicity is owing to which ion(s) to support improved environmental management decisions.

At a minimum, a future WQG for water hardness must consider the toxicity of Ca and Mg both independently and in combination to account for their interactive toxicity and contribution to total salinity. Improved reporting of basic water quality, and all major ions, such that they can be better evaluated as potential toxicity modifiers is required. Although data supporting formal WQG for water hardness were unavailable, we suggest using a two-component background condition approach, which controls for both water hardness and the Ca : Mg ratio. Doing so resulted in WQG that were applicable beyond the regional scale. However, additional laboratory and/or field research are required to determine the ecological significance of these values and support validation of the proposed approach.

Data accessibility

The data used herein have been uploaded as part of the electronic supplementary material supporting the article. Statistical software and data classification protocols are publicly available online, as are described in the Material and methods section. R code can be accessed from the Pyle Aquatic Health Lab Data Archive at the University of Lethbridge (AB, Canada) upon written request to G.G.P. (gregory.pyle@uleth.ca) [53].

Authors' contributions

S.J.B., A.A. and G.G.P. co-designed the study. S.J.B. conducted literature searches, critically evaluated studies, extracted the data, analysed the data and wrote the first draft of the manuscript. A.A. co-contributed to data interpretation and drafting of the manuscript. G.G.P. assisted substantially with data collection, data analyses and manuscript assembly. All authors contributed significantly to manuscript revisions and agreed upon the version submitted for publication.

Competing interests

The authors declare no competing interests.

Funding

This research was supported in part by the British Columbia Ministry of Environment and Climate Change Strategy and a Government of Alberta, Campus Alberta Innovation Program (CAIP) Chair to G.G.P.