How a Complete Pesticide Screening Changes the Assessment of Surface Water Quality

A comprehensive assessment of pesticides in surface waters is challenging due to the large number of potential contaminants. Most scientific studies and routine monitoring programs include only 15-40 pesticides, which leads to error-prone interpretations. In the present study, an extensive analytical screening was carried out using liquid chromatography-high-resolution mass spectrometry, covering 86% of all polar organic pesticides sold in Switzerland and applied to agricultural or urban land (in total 249 compounds), plus 134 transformation products; each of which could be quantified in the low ng/L range. Five medium-sized rivers, containing large areas of diverse crops and urban settlements within the respective catchments, were sampled between March and July 2012. More than 100 parent compounds and 40 transformation products were detected in total, between 30 and compounds in each two-week composite sample in concentrations up to 1500 ng/L. The sum of pesticide concentrations was above 1000 ng/L in 78% of samples. The chronic environmental quality standard was exceeded for 19 single substances; using a mixture toxicity approach, exceedances occurred over the whole measurement period in all rivers. With scenario calculations including only 30-40 frequently measured pesticides, the number of detected substances and the mixture toxicity would be underestimated on average by a factor of 2. Thus, selecting a subset of substances to assess the surface water quality may be sufficient, but a comprehensive screening yields substantially more confidence.

Regional and local scale pesticide transport in the alluvial plain of the Swiss Rhône River Valley

Kirsten Meiwirth

The herbicide transport and the groundwater vulnerability to pesticide contamination were studied in the alluvial aquifer of the Rhône River Valley near Martigny (Southwest Switzerland). The low clay and organic matter content of the soil, the intensive agricultural use, and the low depth of the groundwater table, lead to an alleged high vulnerability of the groundwater to contamination. Groundwater is used for irrigation, drinking water supply, and industrial purposes; it is of great economic and ecological importance which makes it particularly interesting to study. The work consists of local scale transport experiments and a regional monitoring of the groundwater over a two year period. At the local scale, two herbicides (atrazine and isoproturon) and a tracer were applied to instrumented field plots (4 m2) in two consecutive years and the transport through the vadose zone was studied in detail. Water flow and solute transport are closely linked to climatic factors. After the application, the pesticides remain at the soil surface as long as no precipitation occurs. Following to the first heavy rainfall, the chemicals are quickly transported through the vadose zone and part of them reaches the groundwater in a short time. During dry periods, the concentrations decrease steadily in the soil profile and the groundwater. After further rainfall, additional concentration peaks are observed in the groundwater, while only small peaks appear near the soil surface. Approximately 2.5 months after the application, the chemical concentration in the soil and in the groundwater has decreased considerably. The experimental results have been used to evaluate a mechanistic deterministic root zone model (HYDRUS-1D). The aim of the simulations was to define the processes involved in pesticide transport and, if possible, to predict the fate of chemicals applied at the soil surface. The MIM (mobile-immobile water) concept was used in order to account for the rapidity of the observed transport. In the study area, the shallow groundwater influences considerably the water conditions in the unsaturated zone; apparently, in such cases the use of a one-dimensional model to simulate the water flow and the chemical transport in the vadose zone is hindered due to difficulties in defining the lower boundary condition. Groundwater flow is typically three-dimensional and therefore, a global (saturated - unsaturated) 3-D model or the coupling of an unsaturated 1-D model to a 3-D saturated model would be more appropriate. Nevertheless, HYDRUS-1D allowed to describe qualitatively the observed results and to confirm the assumption that accelerated flow occurs on the experimental plots. At the regional scale, 13 piezometers were installed in an observation area (400 ha), and the groundwater was regularly analysed for the concentration of 6 herbicides (atrazine, terbutylazine, simazine, deethylatrazine, isoproturon, and diuron). Herbicides were detected in 12 piezometers, sometimes at high concentrations, but during a short period of time (mid May to mid July). Throughout the rest of the year, the concentrations remained below the drinking water limit. The observations are in good agreement with the conclusions of the local experiments (rapid transport, important influence of the climatic conditions, quick decrease of concentration peaks). Furthermore, the groundwater characteristics play an important role for the observation of herbicides. Near the Rhône river, high hydraulic gradients quickly dilute chemicals leaching from the fields above and herbicides coming from more distant fields may also be observed. Near the slope, an inflow of contaminated water from the hill side (vineyards) contributes to the groundwater contamination. Within the plain, high concentrations of herbicides applied to an above located field can temporarily be observed. Contaminations are confined in time and space and different substances are detected from one piezometer to another. This lack of continuity in pesticide concentrations will complicate the prediction of the solutes' fate in the groundwater, and a very precise knowledge of the pesticides use, both in space and time, is required.

EPFL2003

Assessing toxic impacts on aquatic ecosystems in life cycle assessment (LCA)

Jérôme Payet

In the last decade, several Life Cycle Assessment (LCA) methods for assessing impact of products on living resources have been developed. Beyond the quantified assessments of impacts on living systems, it also checked the feasibility of the impact assessment on human health and ecosystems quality and helps to identify the limits of such methods. Among the different impact categories, that of toxic substances on ecosystems occupies an important place. The extent of these impacts has been stressed on many occasions and the necessity of preserving ecological areas and biodiversity has become a major issue on an international level. By focusing on aquatic ecosystems, this thesis aims at identifying constraints connected with assessment of the impact of chemical substances on ecosystems in LCA and setting up a method for assessing impacts of toxic substances on aquatic ecosystems which meets the requirements of a comparative approach like Life Cycle Assessment. The overall purpose of the thesis is to propose a comparative method for the Life Cycle Impact Assessment of toxics on aquatic ecosystems. With that aim, the dissertation is going throughout 6 major issues: The feasibility of the comparative impact assessment on ecosystems and the identification of associated constraints. The development of a statistical method for comparing impact on ecosystems; The review of the data availability for calculation of Effect Factors. The choice of the most relevant ecotoxicity measure (ECxs1, NOECs2 and LOECs3) for a comparative purpose. The development of best-estimate extrapolation factors for assessing chronic effects based on acute data. The analysis of the ecological realism of the comparative assessment method. These points are analysed throughout the 7 chapters of the thesis. Chapter 1 aims at introducing the thesis. A general presentation of Life Cycle Assessment is proposed, following by a detailed description of the Life Cycle Impact Assessment on ecosystems. This description covers the state of the art of researches and identifies the development needed. Therefore the scope of the thesis and the main points that must be addressed by this research are presented. Chapter 2 starts with a review of existing methods for Life Cycle Impact Assessment on ecosystems (LCIA), the chapter presents the parametric version of the AMI method (Assessment of the Mean Impact), which has been developed during the PhD for the assessment of impact on aquatic ecosystems. For this purpose, a framework and the main requirements for the development of this method are presented. For a comparative assessment, the Hazardous Concentration of a toxic affecting 50% of the species over their chronic EC50 (Effect Concentration affecting 50% of tested individuals), also called HC50EC50, is selected for the calculation of Effect Factors to be implemented in current LCIA methods. The Confidence Interval on the HC50EC50 is provided, enabling comparison between the impact values obtained as results of a Life Cycle Assessment study. The choice of EC50s is based on review of the main ecotoxicological databases, and analysis of the availability and reliability of test results. Moreover, bearing in mind that mostly acute data are available, while LCA deals mainly with chronic exposure, best-estimate extrapolation factors for the HC50EC50 and the associated uncertainty are provided for inorganics, nonpesticide organics, and pesticide organics. Concerning the method itself, in order to find the best methods for calculation of a toxicity indicator, several statistical estimators, parametric and non-parametric approaches are compared, identifying their properties and respective strengths for a comparative method. The analysis relates to both the reliability of the estimator and its Confidence Interval, especially in terms of statistical robustness and Effect Factor stability. Based on these findings, the AMI method is described in detail, and an example of application comparing two wheat crop scenarios differing by the pesticides used is presented. Chapter 3 presents in detail the four methods currently used for the development of Effect Factors for the Life Cycle Impact Assessment (LCIA) on Ecosystems: the parametric version of AMI (Assessment of the Mean Impact) based on HC50EC50s; the Eco-Indicator based on HC50NOECs; USES-LCA based on both HC5NOECs and the Most sensitive species, and the PNEC (Predicted No- Effect Concentration) based on the Most sensitive species. After presentation of the LCIA framework and its main divergences from Environmental Risk Assessment for chemical regulation, the four methods are detailed and applied for the calculation of Effect Factors for 83 substances, covering inorganics, non-pesticide organics, and pesticide organics. Each method is therefore analysed concerning three key points: applicability in the LCA framework, environmental relevance, and statistical reliability. Particular attention is paid to possible bias and the uncertainty, highlighting the following findings: (1) HC5NOECs are on average 50 times higher than the most sensitive species, and this difference in conservatism introduces a bias in the analyses for the method mixing HC5NOECs and most sensitive species. (2) Effect Factors based on the most sensitive species increase the relative weight of the most toxic chemicals by two orders of magnitude, depending on whether the study is based on US or European ecotoxicity databases. (3) the methods based on HC50EC50s and HC5NOECs are the only ones able to provide a Confidence Interval on the Effect Factor, but the Confidence Interval on the HC5NOECs can be more than 10 orders of magnitude greater than that of the HC50EC50s. (4) compared with the Confidence Interval on the HC50EC50s, the most sensitive species cannot be distinguished from HC50EC50s for chemicals characterised by fewer than 5 species, and the HC5NOECs cannot be distinguished from the HC50EC50s for chemicals characterised by fewer than 8 species. Chapter 4 compares two statistical estimators, aiming at calculate the average toxicity of substances on biological species. The two methods provide an estimation of the HC50EC50 and the associated Confidence Interval. On the one hand, parametric method using the geometric mean and a calculation of the confidence interval with Student is considered. On the other hand, a distribution-free method calculates the HC50EC50 based on the median response of species and the confidence interval based on bootstrap. In order to facilitate the use of the non-parametric method, a table linking the number of species tested and the size of the confidence interval is provided for samples from 5 to 500 species. The comparison is based on actual data concerning 191 substances covering inorganics, non-Pesticide organics, and Pesticide organics. The mean and width of the chronic EC50s samples for all the substances are presented. The Shapiro-Wilk test is performed for the 191 EC50s samples and the assumption of log-normality of the distribution failed in more than 20% of the cases. Two causes of this non Log-normality are identified; (1) the skewness, which is shown to be an important issue for the assessment of the average toxicity of chemicals while (2) the multi-modal distributions, which are not likely to influence considerably the final result. A detailed application of the two methods is done with the comparison of two herbicides, the Sulfosulfuron and the Prosulfuron, where the distribution-free method appears to be more powerful than the parametric for a substance-to-substance comparison. Nevertheless, the distribution-free method requires a minimum of 5 chronic EC50s, that cannot be satisfied in most cases. Chapter 5 aims at illustrating the previous chapter in using the nonparametric version of the AMI method for the comparative assessment of the impacts of metals on aquatic ecosystems. This chapter briefly describes the method, then it focuses on the comparative analysis of 9 metals sometimes tested with different salts and speciations. Two interesting results can be highlighted: (1) the toxicity of metals covers the whole range of toxicity of chemicals; (2) the confidence interval of the HC50EC50 for metals is on average twice as great for metals compared with other chemicals. This increase in the variability of ecotoxicological responses from species is likely to be due to the change in bioavailability of metals associated with a change of test conditions (e.g. pH, or Organic Matter). Chapter 6 reviews and analyses the reliability of existing aquatic toxicity databases which can be used for the calculation of Effect Factors for Life Cycle Impact Assessment (LCIA). For that purpose, the main LCIA methods are presented focusing on their data requirement. It concerns: EDIP4 (based on the PNEC); AMI5 (based on parametric HC50EC50); Eco-Indicator (based on the HC50NOEC); USES-LCA (based on the HC5NOEC). Moreover 6 ecotoxicity databases available in an electronic format are analysed: Aquire; Pesticide Ecotoxicity Database (PED); IUCLID; Acute Toxicity Database (ATD); Fathead Minnow database (FMD); and ECETOC Aquatic Toxicity Database (EAT). The analysis especially focuses on the identification of the substances and organisms, the definition of the tests conditions, and the control procedure of the database. A selection of tests is done, retaining a dataset of 128,864 tests results, acute, sub-chronic and chronic. A description of the data availability on the basis of the selected test is performed, considering the available EC50s (Effect Concentration affecting 50% of the individuals tested), LOECs and NOECs (Lowest or No Observed Effect Concentration). The number of covered substances is also analysed regarding the number of species or phyla considered. On that basis, an estimation of the maximum number of possible Effect Factors is performed. The results highlight the discrepancy between the large number of test results available (128,864), and the relatively restricted number of Effect Factors (betweeen 34 and 4959 depending on the method)) that can be calculated for a comparative purpose like LCIA. Chapter 7 provides a conclusion to the work, answering the points underlined in introduction. Then, the fey features of the AMI method are restated and the perspectives and output of the work are presented. In the last part of the thesis, the AMI HC50EC50 database is presented. It provides acute and chronic HC50EC50 data calculated with the parametric version of AMI (geometric mean of the EC50s and confidence interval based on Student) for 522 susbtances. ------------------------------ 1 ECx: concentration of susbtance that affects 50% of the individuals tested for a given effect. 2 NOEC: No Observable Effect Concentration 3 LOEC: Lowest Observable Effect Concentration 4 EDIP: Environmental Design for Industrial Product 5 AMI : Assessment of the Mean Impact

EPFL2004

Regional and local scale pesticide transport in the alluvial plain of the Swiss Rhône River Valley

Kirsten Meiwirth

EPFL2003

Assessing toxic impacts on aquatic ecosystems in life cycle assessment (LCA)

Jérôme Payet

EPFL2004