A water quality index for the removal requirement and purification treatment effort of micropollutants

The European Water Framework Directive (WFD) states that measures should be taken to improve the quality of water bodies to prevent further required extension of current (drinking) water treatment. Hence, for water managers it is of key importance to evaluate and report on the quality of water and the level of purification treatment that is required. For this purpose a novel framework of indices is defined, and their definition allows the inclusion of new, emerging substances. The indices can be calculated based on micropollutant characteristics alone and do not require any knowledge of specific purification treatment installations. Applying this framework of indices to water bodies provides an objective and reproducible way of evaluating the required purification treatment level. The indices were calculated for water quality data for up to 600 micropollutants from five sampling locations along the river Rhine in the Netherlands. This revealed differences between the sampling sites (index values ranged from 145 to 273) and showed that for the river Rhine the required purification treatment level, as well as the underlying removal requirement and purification treatment effort, have not improved over the years, despite the introduction of the WFD in 2000.


GRAPHICAL ABSTRACT INTRODUCTION
Water utilities and water managers strive for, and have to meet, strict requirements regarding the quality of (sources for) drinking water. In addition to extensive monitoring and early warning systems, water utilities invest in combinations of basic and advanced treatment technologies, including membrane processes (e.g. Tul Muntha et al. In Europe, the Water Framework Directive (// EC) (WFD) has been the most comprehensive instrument of European Union water policy since its introduction in 2000. The main objective of the WFD is to protect and improve the quality of freshwater bodies, with the aim of achieving good ecological and chemical status of European waters. In the WFD, preamble 24 states 'Good water quality contributes to securing the drinking water supply of the population.' Furthermore, Article 7 contains statements related to water used for drinking water: • WFD Article 7.1 requires member states to designate water bodies for the production of drinking water • WFD Article 7.2 states that water quality objectives must be achieved in these water bodies • WFD Article 7.3 states 'Member States shall ensure the necessary protection for the bodies of water identified with the aim of avoiding deterioration in their quality in order to reduce the level of purification treatment required in the production of drinking water.' Looking back over the years since the introduction of the WFD, the question arises to what extent deterioration of water quality has been prevented since the WFD was established. Article 7.3 does not mention a quantitative measure that aids water utility managers. Therefore, the main goal of this study is to develop a quantitative measure to assess water quality in light of the level of required purification treatment using available monitoring data of chemical parameters.
The challenge is how the impact and effort of treatment on water quality can be quantitatively defined such that the definition will give insight into this required level.
Water quality relates to many aspects, so there is a need to aggregate quality indicators into one water quality index (WQI). In the WFD, the ecological status of water bodies is classified based on observations of different biological quality elements, i.e. for phytoplankton, aquatic flora, benthic invertebrates and fish (e.g. Birk et al. ). In this paper, the specific focus is on chemical status and its relation to the purification treatment of drinking water in particular.
Indices relating to chemical parameters have long been used to communicate water quality in a single aggregated score that is representative of quality impairments (Horton ; Hurley et al. ; Borges Garcia et al. ). In general, any index is obtained by applying the following procedure.
Firstly, parameters are selected that represent water quality.
Secondly, for each parameter, a sub-index is established, in which these parameters are weighted. Finally, the sub-indices are aggregated into one index according to some function (Tyagi et al. ; Borges Garcia et al. ).
Horton () was one of the first who introduced a WQI. This consisted of a weighted sum of a maximum of 10 sub-indices, divided by the sum of the weights. This number was then multiplied by two coefficients related to the temperature and pollution of a water source. Since then, some of the more well-known variations on this are the WQI-NSF (Brown et al. )  DWB) (2018) as their standard and adding a score for each element's removability based on Gibbs free energy. Van den Doel et al. () use the same concept as in our approach, estimating removal effort based on parameter characteristics. However, their calculation differs by being a more indirect and complex calculation of the contribution of parameters to the index based on frequency and number of parameters that pass the quality threshold (see the description of the WQI-CCME above) and it uses a fixed number of parameters.
In current times it is important to have a measure that includes new and emerging contaminants because these are a potential threat to water quality (van Wezel et al.

)
. This is also feasible, or will be in the near future, as water boards and utilities at present turn to risk-based monitoring to flexibly monitor significant emerging substances (Brunner et al. ). In this paper, we define and describe a novel framework of indices that relate to micropollutant quantities and the removal effort that is required to meet water quality regulations. In this defined framework these two aspects can be assessed separately as well as being integrated. In this way, two questions can be answered: does the source water have micropollutants that are, on average, difficult to remove, and how extensive is the pollution. The required purification treatment level is calculated based on the characteristics of the micropollutants and is expressed in percentages, which naturally relate to purification efficiency. If for any source water the treatment effort is low, and/or removal requirement (RR) is low, or it is declining through time, it can be argued that less advanced techniques or less capacity is needed, resulting in lower investment and operational costs for drinking water purification treatment.
The index can serve as a relatively simple instrument to evaluate the need to reduce required purification treatment level, as specifically mentioned in the EU WFD. This will enable water managers or drinking water companies to establish if these goals are met, or if additional measures need to be taken.
As a case study, we apply the water quality indices to evaluate the water quality of the Rhine at several drinking water intakes. Based on these indices we assess how the quality of the Rhine and the required purification treatment have developed since the introduction of the WFD in 2000.

METHODOLOGY
The WQI for required purification treatment level is composed of two elements: the WQI RR and the WQI PTE.

Calculation of the WQI RR
The DWB () contains quality standards for both organic and inorganic substances and chemical organoleptic/aesthetic or signalling indicators. For good quality drinking water, all of these quality standards have to be met by drinking water companies. Table S1 in the Supplementary   Information lists the water quality standards mentioned in the DWB ().
To calculate the removal requirement to meet these standards, the ratio of the maximum measured concentration in a given period and the corresponding DWB standard (see Table S1, Supplementary where C DWB is the DWB standard for a micropollutant, C MAX is the maximum measured concentration of a micropollutant in a period, i is a micropollutant and n is the total number of micropollutants. Every micropollutant that exceeds its standard will at most add between 0 and 100% to the index. The WQI RR indicates source water quality in terms of the removal requirement and this is independent of the type of treatment.

Calculation of the WQI PTE
For this index we assign a weighting factor to each substance that has to be removed; its value depends on the estimated removal effort.
The estimated removal effort is based on two substance properties: the octanol-water partition coefficient K ow and a biodegradation constant (for more information see Text S4, Supplementary Information). Log K ow is a measure of hydrophobicity. In general, the more hydrophobic a substance is (high log K ow ), the easier it is to remove from water because of a tendency to adsorb to matter. The higher the biodegradability of a substance, the easier the removal by biological processes. We choose to use log K ow and biodegradation rate constants for calculating the removal effort because both adsorption and biodegradation processes occur during conventional, 'simple' drinking water treatment. We do not explicitly define purification steps in any particular treatment, as this will differ per treatment installation ( All possible values of the two properties log K ow and biodegradation are stored in four bins with a weighting factor (w1 and w2) per property; see Table 1. This approach is adopted from the work of Fischer et al. (). Biodegradation weights are set at only half of that of log K ow .
Biodegradation is an important process, especially in wastewater treatment plants (WWTPs). However, WWTP processes differ from drinking water treatment processes.
The water quality (presence of nutrients) is different, and typically, the residence time of water in a WWTP is significantly longer (1-3 days) than the contact time in a (rapid) sand or activated carbon filter (typically 20-40 min). Therefore, if not downscaled, the application of biodegradability constants as indicators may overestimate the contribution of biodegradation in drinking water treatment plants.
These scaled weighting factors w1 and w2 are used to obtain an estimate of the effort to remove the substance in question. As drinking water treatment consists of several sequential processes, in which each process results in a certain removal, multiplication is considered more realistic than a simple averaging of the removal properties. This is the PTE (see Equation (3)).
where PTE is the removal estimate for micropollutant i, w1 and w2 are weights, as in Table 1. As an example, a PTE value of 0 means that it is expected that the substance will be fully removed. In contrast, a PTE value of 100 means that the removal of the substance is expected to be negligible.
The actual estimated removal in percentages can be calculated for a substance as 100-PTE.  (3)) has a realistic removal in the validation by expert judgement for a 'conventional' treatment setup where sorption and biodegradation processes constitute the main removal mechanisms (see Table S2, Supplementary Information). In the calculation of the WQI PTE summed micropollutants are omitted, because removal can differ between members of a group that are summed.
The WQI PTE is calculated as the average of the removal indication PTE (see Equation (3)) for all substances i that exceed their standard in the DWB () (see Equation (4)).

WQI PTE
where WQI PTE is calculated for all substances n that exceed the standard in the DWB (). The WQI PTE yields a value between 0 and 100, representing full removal and no removal respectively.
Note that EPIsuite is not able to calculate the property constants t1 and t2 (see Table 1) for inorganic substances, since these constants fall outside its modeling domain. For inorganic micropollutants, the PTE was set at a low value to ensure that the removal effort index value corresponds to an easy to remove indicator value (17) as for most of these compounds this will generally be the case in practice.
Exceptions to this assumption are compounds such as chloride, fluoride, sulfate, nitrite, ammonium, and nitrate, as these are known to be difficult to remove. These micropollutants are given a high PTE (90) If the WQI RR_PTE is high, this means the combination of removal requirement and PTE is high for the combination of different micropollutants reported.

Data characterization
To gain insight into the extent of the measuring program at different Rhine locations and how many of these measured micropollutants exceed their DWB () value, we illustrate them in Figures 3 and 4. Figure 3 shows the number of measured and reported micropollutants with a DWB standard at the five locations. In Figure 3 it can be seen that the number of micropollutants measured increases over the years by a factor of two to 16, depending on the location. This is because parameters are added continuously to measurement programs, based on new insights into possible pollutants. Parameters are less often discarded. Figure 4 shows   Table S1, Supplementary Information). Notwithstanding that, Figures 3 and 4 still raise the question of whether more micropollutants are found to be exceeding because in time more micropollutants are measured in these monitoring programs, or if these newly exceeding micropollutants were added because they were suspected to be new or emerging. We think the latter is more likely. Nevertheless, we will be careful in interpreting a deterioration, and will be more confident in an improvement. This issue is discussed in more detail in the discussion.

Values and trends in the WQI RR
We calculated the WQI RR values for source waters for drinking water production at the five locations along the river Rhine in the period 2000 to 2018 and determined if these show a (significantly) improving trend, according to a linear regression model. In Table 2 the significance value (p-value) of these calculated trends in WQI RR is shown.
The WQI RR does not seem to improve in any of the locations.
In Table 2 it can be seen that historically, Haringvliet has the best water quality on average. Andijk has the lowest WQI RR in 2018, and therefore has the best recent water quality with regard to the index for removal requirement.

Contribution of substance categories to the WQI RR
To get an idea of the kind of substances that contribute to the trend in WQI RR in Table 2, we looked at the excee-  Table S1, Supplementary Information).
more than one label, some micropollutants are counted in more than one parameter category. year to year at all the locations.
Lobith is of strategic importance because this is the where the Rhine enters the Netherlands. As an example, for this location, we plot the individual removal requirement for micropollutants per parameter category (Table 3)

The purification treatment effort index WQI PTE
The effort of meeting the removal requirement, as explained in the Methodology section, can differ depending on the difficulty of removing micropollutants. We address this aspect by calculating WQI PTE (see Equation (4)) for all locations.
A high value indicates that the substances in the WQI RR are hard to remove in drinking water treatment.  A 'þ' indicates the WQI RR is increasing, which implies there is no improvement. No 'þ' or 'À' is given for trends with significance level p > 0.5.   The required purification treatment level index WQI

RR_PTE
The two indices WQI RR and WQI PTE aggregated into the WQI RR_PTE (see Equations (5) and (6)) can evaluate the required purification treatment level for water. Figure 6 shows the development of the WQI RR_PTE for the five  Table 5), which implies the water quality in terms of required purification treatment level is not improving.
The WQI RR_PTE is more sensitive to the removal requirement than to the purification treatment effort, and this inequality increases with more exceeding micropollutants. This is because the first is a summation, which increases with every new micropollutant with a removal requirement, and the latter is an average that will not change dramatically with an extra value.
In 2018, the order of locations based on their WQI RR_PTE from low (better), to high (worse) was: Andijk (  value and/or difficult removal. As an example how the WQI RR_PTE results from individual micropollutants, we show the micropollutants that contribute to the calculated WQI RR_PTE for Lobith in Figure 7. On the left bar are the intrinsic PTEs modeled for the micropollutants from low (yellow/light, easy removal) to high (red/dark, difficult removal). On the right is their contribution to the WQI RR_PTE (Equation (5)), the darker, the higher.
Logically, substances with high intrinsic PTE (red/dark, in the left bar) have a tendency to require more extensive purification treatment. Most of the substances in Figure 7 have a high intrinsic PTE. In contrast, most substances with a low intrinsic PTE (easy removal) do not contribute to the WQI RR_PTE (not shown). These are substances such as bis(2-ethylhexyl)phthalate (DEHP) and benzo(a) pyrene. Some substances with medium PTE also do not contribute to the WQI RR_PTE (not shown). This is because these micropollutants exceeded their DWB standard in this location with a low percentage (RR) and the PTE was enough to cover this. These are substances such as bentazon and monolinuron. Iron and aluminum also have a low PTE (left bar in Figure 7, yellow/light). However, these micropollutants exceeded the DWB standard to such a high extent in this location (represented in their RR) that their PTE was insufficient for complete removal. This is why they still A 'þ' indicates the WQI PTE is increasing, which implies there is no improvement. No 'þ' or 'À' is given for trends with significance level p > 0.5.
contribute to the WQI RR_PTE, shown in Figure 7. Substances that have improved over the years (these have a dark color mostly in earlier years) are glyphosate, NO 2 , atrazine, ammonium (see Figure 7). Substances that contribute  the comparison of the WQI RR between locations will be hampered by gaps in the reported micropollutants. The framework described here can still be used, but in those cases it is recommended to keep to a fixed set of micropollutants in calculating the indices.
One uncertainty introduced by also including 'all' newly reported substances is that we cannot exclude the possibility that some of the micropollutants have already been exceeding their standard before they were or could be measured. In these cases, the indices are underestimated in the early years and will seem to have a tendency to increase, but the appropriateness of the measurement program affects how large this effect is. An estimation of the likelihood of emission routes existing for these micropollutants could indicate the potential of underestimation, but this is beyond the scope of this paper. In the future, the likelihood that newly measured parameters are added because there was some indication for their potential novel threat to water quality will increase with the introduction of risk-based monitoring.
For now, it enough to point out that the indexes were at least not improving for the locations under investigation and that there were differences in water quality between the locations.
In addition to a complete measurement program, the measurement frequency in the monitoring program is important. The fewer times a micropollutant is measured within a period, the higher the chance that some peak concentrations will be missed, therefore underestimating the value of the indices. The current indices are not suitable for indicating adverse effects on the environment or people, as the exposure period to the pollution is not incorporated, which is important for risk assessment. In addition, we did not use water quality standards directly linked to human health. The index presented may rather be seen as an index for the relative complexity of treatment processes required to produce safe, good quality drinking water according to the Dutch DWB, to a point at which water utilities do not have to interrupt water intake because micropollutants exceed their quality standard.

CONCLUSION
Despite all assumptions and potential drawbacks of our indices and calculations, applying such a calculation framework reveals differences between locations and in time in an objective and reproducible manner, which allows us to study trends in very complex data. We conclude that it is possible and useful to base the indices on data from evolving monitoring programs, to also capture new and upcoming threats. The approach and indices applied to drinking water can also be applied to the influent and effluent of WWTPs, and to (industrial) reuse of water that has to meet quality standards.
Based on the calculations of these three indices, we conclude that the objectives of the WFD have not led to improvement of the water quality in the river Rhine.
Clearly, extra effort is needed in the field of emission reduction, with a focus on new and emerging substances and their removal, in order to reduce the purification treatment level required for the preparation of drinking water.
Measurement programs must remain aligned to this, for instance with the help of risk-based monitoring for early signals of new and possibly problematic substances. In addition, measures should be taken to limit emissions of these substances before they become a real problem.