Multi-Level Approach for the Integrated Assessment of Polar Organic

For the German and Austrian part of the catchment, this value was estimated by the catchment-area extrapolation method HYDMOD.(28) Thirteen out of 125...
0 downloads 10 Views 4MB Size
Download PDF
Article pubs.acs.org/est

Multi-Level Approach for the Integrated Assessment of Polar Organic Micropollutants in an International Lake Catchment: The Example of Lake Constance Christoph Moschet,†,‡ Christian Götz,† Philipp Longrée,† Juliane Hollender,†,‡ and Heinz Singer†,* †

Eawag, Swiss Federal Institute of Aquatic Science and Technology, 8600 Dübendorf, Switzerland Institute of Biogeochemistry and Pollutant Dynamics, ETH Zürich, 8092 Zürich, Switzerland

‡

S Supporting Information *

ABSTRACT: Polar organic micropollutants (MPs) can have ecotoxicological effects on aquatic ecosystems and their occurrence in drinking water is a threat to public health. An extensive exposure assessment of MPs in large river and lake catchments is a necessary but challenging proposition for researchers and regulators. To get a complete picture of MP exposure in a large catchment, we employed a novel integrated strategy including MP measurement in the international catchment of Lake Constance and mass-flux modeling. A comprehensive screening of 252 MPs in the lake water by high-resolution mass spectrometry was used to identify the most commonly present MPs for the study site. It was found that the wastewater borne MPs diclofenac, carbamazepine, sulfamethoxazole, acesulfame, sucralose, benzotriazole, and methylbenzotriazole accounted for the most frequent and prominent findings. The concentration pattern of these compounds in the catchment was calculated based on regionalized inputs from wastewater treatment plants (WWTPs) and substance specific elimination rates. In 52, 8, and 3 of the 112 investigated river locations the concentration exceeded the predicted no-effect levels for diclofenac, sulfamethoxazole and carbamazepine, respectively. By coupling the catchment and lake model the effect of future trends in usage as well as possible mitigation options were evaluated for the tributaries and the lake. The upgrade of the major WWTPs in the catchment with a postozonation step would lead to a load reduction between 32% and 52% for all substances except for sucralose (10%).

■

INTRODUCTION Polar organic micropollutants that are typically present at low levels (μg L−1 to ng L−1) raise ecotoxicological concerns even at these low concentrations.1 Due to increasing research and monitoring activities in recent years, an increasing number of MPs have been detected in surface waters worldwide, for example, refs 2−7. Household and down-the-drain chemicals used in large quantities are continuously released via urban sewer systems to wastewater treatment plants (WWTPs) and ultimately to receiving surface waters. The potential ecotoxicological and public health issues associated with the occurrence of MPs in water resources point to a critical need for strategies that enable efficient and comprehensive evaluation of surface water quality. For sustainable management of water quality, exposure assessments have to be performed within a river basin considering all sources of emission. Analyzing single sampling sites within a catchment only delivers nonrepresentative point information. The catchment based approach also represents the key principle of the European Water Framework Directive (WFD). However, in large watersheds of big rivers and lakes, measurements for all substances that can potentially enter the surface water are neither practical nor feasible. Additionally, the © 2013 American Chemical Society

spatial heterogeneity of water quality can hardly be covered by a reasonable number of measurements in large catchments. Therefore, assessment of chemical water quality and the evaluation of installed mitigation measures are often based on a limited and inflexible data set that inadequately represents the actual situation. Recent advances in liquid chromatography coupled to highresolution mass spectrometry (LC-HRMS) dramatically opened the observation window for a vast array of organic micropollutants.8,9 These techniques enable the identification of a large number of targeted analytes and to identify suspected and unknown compounds at low concentrations. With this, exposure assessment is no longer confined to the very restricted and fixed number of priority substances recommended by risk assessment calculations, by regulations and laws (i.e., priority substances of the WFD), or by measurements of a few and Special Issue: Rene Schwarzenbach Tribute Received: Revised: Accepted: Published: 7028

November 2, 2012 January 23, 2013 February 5, 2013 February 26, 2013 dx.doi.org/10.1021/es304484w | Environ. Sci. Technol. 2013, 47, 7028−7036

Environmental Science & Technology

Article

The performance of the tiered approach is demonstrated with the example of the Lake Constance catchment (a subcatchment of the Rhine River). Lake Constance was selected as it offers multiple challenges caused by the division of the catchment among three different countries and the use of its deep water for drinking water production for approximately four million people.22

preselected compounds. As sufficient screening measurements in large rivers and lakes are economically almost impossible, a well-balanced combination of modeling and measurements as well as the careful selection of sampling sites is necessary for a representative water quality investigation.10 A lake as an integrator of its tributaries stores chemical water quality information of all emission sources in the catchment over a long time period in relation to its hydraulic retention time.11 Therefore a screening analysis of relatively few lake water samples can elucidate the catchment-relevant substances without extensive measurement efforts within individual tributaries. It has been shown that with simple two-box compartment models past trends of contaminants can be adequately calculated in a lake with a restricted set of input data.12 Additionally, to model future trends, scenarios with different assumptions can be tested. Although lakes are perfectly suited to monitor long-term activities in the watershed by lake-modeling and screening measurements, the spatial distribution of chemicals in the catchment and the identification of highly impacted river stretches can only be depicted by geography-referenced modeling. For this purpose in-streamwater quality models for point-source emitted chemicals can be used.10,13 Whereas some in-stream models require a rather diverse set of input data and therefore are more designed for a detailed analysis of regional and well-defined river basins (i.e., GREAT-ER,14 PhATE,15 Mike11,16 QUAL2E 17), other models are more dedicated for the evaluation of water quality in large catchments using only essential input data like the river network topology, consumption data, and WWTP elimination rates (i.e., models of refs 18−21). Although most of the large-scale river network models can be used to calculate different exposure scenarios, they have shortcomings when variations of consumption data over different countries, states, and regions in a large watershed have to be considered. As pointed out, several tools for the assessment of polar organic micropollutants in lakes and rivers are available and numerous analytical methods for selected priority substances are published. However, an integrated and effective approach is still needed which allows a fast evaluation of the contamination situation for the most relevant chemicals in a specific catchment and enables an evaluation of reduction strategies. Therefore, we investigated a three-step approach that merges the advantages of lake screening analysis as well as lake and river modeling. To fully profit from the presented approach, the investigated river catchments have to discharge into a lake, preferably one with water residence time of several years. We hypothesize that combining these strategies would lead to cost-effective monitoring, modeling of various scenarios, and assessment of reduction strategies in large lake catchments. The methods involved were arranged as follows: (i) comprehensive multicompound screening in selected lake samples using liquid chromatography coupled to high-resolution mass spectrometry in order to identify catchment-relevant MPs; (ii) mass-flow modeling of catchment-relevant MPs in the tributaries complemented by measurements to assess the ecotoxicological risk of the selected chemicals in the river network and to identify hotspots in the catchment; and (iii) lake modeling using the calculated input loads from the river network model to predict long-term trends of the lake concentrations and to evaluate the impact of different reduction strategies on the chemical status of the lake and the tributaries.

■

MATERIALS AND METHODS Site Description. Lake Constance is the second largest lake in Central Europe in terms of water volume and is located at the border between Switzerland, Germany and Austria (see Supporting Information (SI) Figure SI-1.1 for the geographical location). It is separated into two lake basins, the large Upper Lake Constance in the east (47°34′ N to 9°28′ E, surface area 472 km2, max. depth 253 m, mean residence time 4.3 years) and the smaller Lower Lake Constance in the west (47°45′ N to 9°01′ E, 62 km2, 40 m, 27 days). The Upper Lake Constance is a holomictic lake with a thermocline at a depth of around 9 m and with deep mixing between December and April.23 The large catchment area (11 500 km2) comprises parts of Germany (D), Switzerland (CH) and Austria (A). From the total population in the catchment (1.56 million), 46% lives in Germany, 31% in Switzerland, and 23% in Austria (information provided by the International Commission on Water Protection of Lake Constance, IGKB). A large part of the catchment in the south of the lake is located in alpine areas. The main tributary entering Lake Constance is the river Rhine with 62% of the total inflow.22 In addition, there are 11 other important tributaries (see SI Figure SI-1.2). Multicompound Screening. For the preliminary investigations and identification of relevant MPs, grab samples at four different spots (see SI Figure SI-1.2) were taken by the Institute for Lake Research at Langenargen (D) on August 24th, 2008. At each spot, one sample was taken from the epilimnion (1 m) and from the hypolimnion (20 to 230 m, depending on the spot). The 252 compounds were chosen due to their high use rate, stability and mobility in the environment and ecotoxicological concerns. The substances consisted of pesticides and their transformation products (TPs), pharmaceuticals/-TPs, biocides/-TPs, industrial chemicals, corrosion inhibitors, artificial sweeteners, personal care products, additives, and one tracer (see SI SI-3 for a list of all screened substances). The screening was performed with a high-resolution mass spectrometer (LTQ-Orbitrap, Thermo Fisher Scientific) coupled to a reversed-phase high-performance liquid chromatography (HPLC) system. Samples were enriched with a mixed-layered cartridge that contains four different solid phase extraction (SPE) materials: Oasis HLB, Isolute ENV+, StrataXCW, and StrataXAW. The mixed layered cartridge was used to capture a broad range of MPs, including anionic and cationic chemicals. The method is described in ref 24. Sampling. For the detailed investigation of the catchment, four grab samples from the twelve main tributaries were taken on April 28th, August 26th, September 15th, and September 24th, 2009 by the local authorities (see SI Figure SI-1.2). Samples were taken over the cross-section of the river close to the river mouth. Additional lake samples for the evaluation of the lake model were taken in the middle of the Upper Lake Constance on May 4th, June 8th, July 6th, August 3rd, and October 5th, 2009 at depths of 1 and 230 m. In order to estimate per capita consumption and WWTP elimination data 7029

dx.doi.org/10.1021/es304484w | Environ. Sci. Technol. 2013, 47, 7028−7036

Environmental Science & Technology

Article

Because we assessed the situation under worst-case conditions, we chose the base flow discharge Q95% as relevant discharge. This is the discharge that is exceeded at 347 days of the year. The Q95% were provided by the Swiss Federal Office for the Environment (FOEN) for the Swiss part of the catchment. For the German and Austrian part of the catchment, this value was estimated by the catchment-area extrapolation method HYDMOD.28 Thirteen out of 125 discharge values could not be determined due to the small catchment size of the particular river locations. Ji̅ was derived from national sales data (see SI Table SI-2) as well as from back calculations of measured WWTP inflow loads. For the back calculation, the measured input loads of the 19 investigated WWTPs were divided by the corresponding fex. The values of fex for pharmaceuticals and artificial sweeteners were collected from the literature (see SI Table SI-2). The national sales data and the back calculated consumption data were in good agreement (see SI Table SI-2). No sales data were available for benzotriazole, methylbenzotriazole, acesulfame, and sucralose. Ji̅ was taken from the back calculated consumption data for all compounds. The WWTP elimination (ei̅ ) was calculated from the relative difference between influent and effluent concentrations of the 19 WWTPs. In order to distinguish differences in consumption and WWTP elimination rates among the three countries, an analysis of variance (ANOVA) according to Chambers et al.29 (using the computer program R, www.r-project.org) was performed with the measurements of the 19 WWTPs (4 in Switzerland, 10 in Germany and 5 in Austria). For the consumption data, different values between the countries were implemented if the analysis of variance showed a signif icant value (Pr-value: 0.01−0.05). However, for the WWTP elimination, different values between the countries were only implemented if the analysis of variance showed a highly signif icant value (Pr < 0.001). This was done because differences in consumption data were expected, whereas in the WWTP elimination it was not. In this study, only parent compounds were analyzed and modeled because no water quality criteria have been established for transformation products yet. Only for sulfamethoxazole, its transformation product N4-acetyl-sulfamethoxazole was included in WWTP influent sample measurements, because this transformation product is completely cleaved back to sulfamethoxazole in the WWTP.30 The uncertainty for the substance load from the WWTP effluent into the river (±43%) was determined by error propagation of the standard deviations of consumption (±29%) and WWTP elimination (±32%) determined by the measurement of the 19 WWTPs. To assess the water quality of the tributaries, calculated river concentrations after the 112 WWTP effluents at base flow discharge (Q95%) were compared to predicted no effect concentrations (PNECs), if available. PNECs were adopted from the water quality criteria recommended by the Swiss Centre for Applied Ecotoxicology31 which were derived according to the EU’s Technical Guidance Document for Deriving Environmental Quality Standards,32 and were reviewed by international experts: PNECdiclofenac = 0.05 μg L−1, PNECsulfamethoxazole = 0.12 μg L−1, PNECcarbamazepine = 0.5 μg L−1, PNECbenzotriazole = 30 μg L−1, PNECmethylbenzotriazole = 50 μg L−1. For sucralose and acesulfame, no PNEC value was available in the literature at that time. Lake Model Setup. The lake model used, MASAS light,12 was only applied for the Upper Lake Constance because (i) the Upper Lake Constance is separated from the Lower Lake

for the selected MPs, 24 h time-proportional composite samples of the influent, as well as of the effluent, of 19 large WWTPs, distributed over the three countries, were taken on November 12th, 2009, during dry weather conditions (see SI Figure SI-1.2). All samples were stored unfiltered in glass bottles in the dark at −20 °C until analysis. Measurement of selected MPs. For the quantification of carbamazepine, diclofenac, sulfamethoxazole, its transformation product N4-acetylsulfamethoxazole, benzotriazole, and 4-/5methylbenzotriazole, a fully automated solid phase extraction coupled online to high performance liquid chromatographytandem mass spectrometry (online-SPE-HPLC-MSMS) was used. The hardware setup is described in ref 25; the method is described in ref 26 (second method). For the quantification of acesulfame and sucralose, an online-SPE-HPLC-Orbitrap method according to ref 27 was used. The artificial sweeteners in the river were only analyzed on August 28th and September 24th 2009; and in the lake on May 4th, August 3rd, and October 5th, 2009. For all analytes the isotope labeled analogues were used as internal standards to guarantee method precision of 5−15% (relative standard deviation) and an accuracy of 80−120% (recovery of spiked samples). River Network Model. The applied river network model is based on a model developed by Ort et al.,19 which has originally been developed for Switzerland. It was modified to account for country-specific consumption and WWTP elimination. It was expected that there are differences in consumption pattern between the three countries because of different product sales. However, no differences in WWTP elimination were expected because most considered WWTPs in the catchment are equipped with the same technology (88% with nitrification, 97% with phosphorus elimination, 64% with denitrification). Nevertheless, the hypothesis of a countryspecific WWTP elimination was tested. The model has been previously demonstrated to give a reasonable simulation for polar organic MPs that constantly enter the environment through WWTPs without undergoing significant physical or chemical reactions in the rivers.19 Loads and concentrations of the selected substances in each river location downstream from a WWTP effluent are calculated according to eq 1 (adapted from ref 19). All model input parameters are shown in SI Table SI-2. c(S)j =

L(S)j Q

=

1000 ·∑i Ji ̅ ·fex ·Pi ̅ ·(1 365

Q

− ei̅ ) ·Tij̅ (1)

c(S)j [μg L−1] is the predicted concentration/L(S)j [μg d−1] is the predicted load of substance S in the river location j after each considered WWTP i, Q [L d−1] is the discharge of the river at this river location, Ji̅ [mg CA−1 a−1] is the countryspecific per capita consumption in the WWTP i, P̅i [CA] is the number of people connected to each WWTP i, fex −] is the fraction of parent compound excreted by humans, ei̅ [−] is the country-specific elimination in the WWTP i and T̅ ij denotes the topology matrix. T̅ ij is determined by the WWTP network and is explained in more detail in Ort et al.19 From the physicochemical properties of the selected substances (SI Table SI-2), it can be concluded that neither significant degradation nor sorption processes take place in the river.19 Thus, the load of the selected MPs in the river sums up whenever a downstream located WWTP effluent enters the river. 7030

dx.doi.org/10.1021/es304484w | Environ. Sci. Technol. 2013, 47, 7028−7036

Environmental Science & Technology

Article

Figure 1. Concentration range of detected MPs in the epilimnion and hypolimnion of the Upper and Lower Lake Constance separated into three substance classes. PE: pesticides, PA: pharmaceuticals, V: Various (biocides, industrial chemicals, corrosion inhibitors, artificial sweetener), TP: transformation product, thin line: single values, thick line: median of all single values. The retrospectively processed concentrations of acesulfame are not shown (see text).

As shown in literature data the photolytic degradation assumed to be irrelevant for the other substances.34,36,37 Initial concentrations of the MPs in the epilimnion and hypolimnion were known from investigations in May 2009 (SI Table SI-2). The model was validated for the year 2009 by comparing the model output with measured lake concentrations throughout the year. To calculate the trend of lake concentrations over a 10 year time scale, a yearly stratification period of 200 days followed by 165 days of complete mixing was set in the model.22 All physical parameters of the river and lake model as well as the number of inhabitants and the consumption of the investigated compounds were kept constant within the 10 year forecast period. In the different WWTP upgrade scenarios, only the input loads of the MPs were varied depending on the reduction strategy.

Constance by a short stream section and (ii) the Upper Lake Constance has a 60 times larger water volume. The total load of the selected MPs entering the Upper Lake Constance was taken from the results of the river network model described above. The model includes the following processes: (i) flushing of the lake, (ii) water exchange between epilimnion and hypolimnion, and (iii) direct photolytic degradation. Other removal processes such as sorption, volatilization, abiotic, biotic, or indirect photolytic degradation were neglected because all selected substances show low Kow and KH values as well as high persistence in the environment (see SI Table SI2). Concentrations in the epilimnion and hypolimnion were calculated according to eqs 2 and 3.33 Jin,E ⎛Q Q Q dc E(t ) = + Ex ·cH(t) − ⎜ out + Ex dt VE VE VE ⎝ VE ⎞ + kdeg,photo⎟ ·c E(t ) ⎠

Jin,H Q Q dc H(t) = + Ex ·c E(t) − Ex ·cH(t ) dt VH VH VH

■

RESULTS AND DISCUSSION Comprehensive Multicompound Screening. In the comprehensive screening, 40 of 252 chemicals in the free water of Lake Constance were detected (pesticides: 10/76, pesticide-transformation products (TPs): 8/54; pharmaceuticals (PA): 10/65; pharmaceutical-TPs: 3/22; biocides: 4/13; biocide-TPs: 1/1; industrial chemicals: 1/11; corrosion inhibitors: 2/2; artificial sweeteners: 1/1). Eighty-four percent of all measurements of the detected chemicals were found in levels below 10 ng L−1 (see Figure 1, concentrations of all chemicals are in SI Table SI-3). Among the four sample locations, differences in MP concentrations were minor, indicating a fast horizontal mixing. Highest concentrations were found for benzotriazole (140 ng L−1), methylbenzotriazole (41 ng L−1), 4-acetamidoantipyrine (33 ng L−1), 2naphthalenesulfonic-acid (21 ng L−1), atenolol acid (19 ng L−1), sucralose (15 ng L−1), carbamazepine (14 ng L−1), and sulfamethoxazole (13 ng L−1). This means that all concentrations are considerably below the target values for drinking water which were proposed by the Danube, Meuse, and Rhine Memorandum of the International Association of Waterworks in the Rhine Catchment Area38 (0.1 μg L−1 for anthropogenic substances with known biological effects, 1.0 μg

(2)

(3)

cE and cH are the epilimnion and hypolimnion concentrations at the time t, Jin,E and Jin,H are the calculated load inputs into the epilimnion (through the lateral inflow of tributaries; 63% of the total load) and into the hypolimnion (through the inflow of directly discharging WWTPs and deep water inflow of the river Rhine; 37%), VE and VH are the volumes of the epilimnion (4.3 km3) and hypolimnion (43 km3),22 QEx is the water exchange between the epilimnion and hypolimnion (6.4 km3 y−1), Qout is the discharge of the lake (14 km3 y−1 in summer, 7.9 km3 y−1 in winter)23 and kdeg,photo is the elimination rate by direct photolysis (diclofenac: 0.046 d−1, sulfamethoxazole 0.0046 d−1, during the stratification period). The photolysis rate of diclofenac was adapted from data of Lake Greifensee.34 For sulfamethoxazole, an experimentally derived photolysis rate from Lake Josephine, Minnesota (45° latitude) 35 was extrapolated to the epilimnion volume of Upper Lake Constance taking the cloud attenuation (50%) into account. 7031

dx.doi.org/10.1021/es304484w | Environ. Sci. Technol. 2013, 47, 7028−7036

Environmental Science & Technology

Article

Figure 2. Top: Measured consumption data per capita and year (mg CA−1 a−1). D: n = 10, CH: n = 4, A: n = 5. Middle: WWTP elimination (%). D: n = 10, CH: n = 4, A: n = 5. The whiskers in the boxplots extend to the most extreme data point which is no more than 1.5 times the interquartile range. Data points outside the whiskers (outliers) are shown as black dots. The asterisks indicate statistical significance of the differences between the countries derived by the analysis of variance: * = significant (Pr-value: 0.01−0.05), ** = very significant (0.001−0.01), *** = highly significant (