- Email: [email protected]
Re-evaluation of groundwater monitoring data for glyphosate and bentazone by taking detection limits into account
Science of the Total Environment 536 (2015) 68–71
Contents lists available at ScienceDirect
Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv
Short Communication
Re-evaluation of groundwater monitoring data for glyphosate and bentazone by taking detection limits into account Claus Toni Hansen a, Christian Ritz b, Daniel Gerhard c, Jens Erik Jensen d, Jens Carl Streibig e,⁎ a
Ålegårdsvej 29, 2740 Skovlunde, Denmark Nutrition, Exercise and Sports, University of Copenhagen, Rolighedsvej 26, 1958 Frederiksberg C, Denmark Mathematics and Statistics, University of Canterbury, Private Bag 4800, Christchurch 8140, New Zealand d SEGES P/S, Crops & Environment, Agro Food Park 15, 8200 Aarhus N, Denmark e Plant and Environmental Sciences, University of Copenhagen, Hoejbakkegaard Alle 13, 2630 Taastrup, Denmark b c
H I G H L I G H T S
G R A P H I C A L
• It is imperative to include all samples – including those falling below detection levels. • Samples with pesticide concentrations below detection limits result in leftcensored observations. • Groundwater pesticide medians are 104–105 lower when including than when excluding “non-detect” • Excluding “non-detect” samples significantly overestimates pesticide load in groundwater.
The size of glyphosate contamination in Danish groundwater without including samples with glyphosate below detection level is 10,000 to 100,000 higher than doing a parametric event-time model, where non-detect samples are left censored. The trend lines are rather similar. The broken lines are the EU drinking water directive maximum value 0.1 μg l− 1 and the dotted line the detection limit of 0.01 μg L− 1. Numbers are number of samples above detection line and total number of samples.
a r t i c l e
a b s t r a c t
i n f o
Article history: Received 28 May 2015 Received in revised form 8 July 2015 Accepted 8 July 2015 Available online xxxx Editor: D. Barcelo Keywords: Groundwater Herbicides Drinking-water supply Contaminants
⁎ Corresponding author. E-mail address: [email protected] (J.C. Streibig).
http://dx.doi.org/10.1016/j.scitotenv.2015.07.047 0048-9697/© 2015 Elsevier B.V. All rights reserved.
A B S T R A C T
Current regulatory assessment of pesticide contamination of Danish groundwater is exclusively based on samples with pesticide concentrations above detection limit. Here we demonstrate that a realistic quantification of pesticide contamination requires the inclusion of “non-detect” samples i.e. samples with concentrations below the detection limit, as left-censored observations. The median calculated pesticide concentrations are shown to be reduced 104 to 105 fold for two representative herbicides (glyphosate and bentazone) relative to the median concentrations based upon observations above detection limits alone. © 2015 Elsevier B.V. All rights reserved.
C.T. Hansen et al. / Science of the Total Environment 536 (2015) 68–71
1. Introduction Since 1989, Danish groundwater has annually been analysed for pesticide residues in order to safeguard that the EU drinking water directive limit of 0.1 μg pesticide L−1 is met (Thorling et al., 2013). By contrast to other countries in the EU, the Danish drinking water is minimally treated groundwater. Therefore, the result of this long time monitoring of pesticide residues in groundwater is a crucial factor in the various Pesticide Action Plans agreed upon in the Danish Parliament in order to protect the groundwater (Kudsk and Jensen, 2014). However, in contrast to normally accepted practice the samples below the detection limits (non-detect samples) were excluded from the statistical analysis (Thorling et al., 2013). As examples of normal practice using non-detect samples, Köck-Schulmeyer et al. (2014) in a four-year monitoring of polar pesticides in groundwater included samples below the detection limit as a surrogate value of half the detection limits for the pesticides, although they did not explicitly describe the statistical method used. Fram and Belitz (2011) explicitly described how the data, including a surrogate concentration for samples below the detection limit, were analysed by the method outlined by Helsel and Hirsch (2002). The herbicides bentazone and glyphosate are important in a number of crops and they can be found in the upper groundwater. Thus, they are appropriate representative pesticides for the evaluations carried out in this study. The weakly acidic herbicide, bentazone, is mobile and considered moderately persistent in soil. The Freundlich adsorption isotherm does not vary much among soils with clay contents, whether it is used under conventional or reduced tillage cultivation. The first order degradation constant in the soils does not significantly change in response to clay soils. However, tillage system has been demonstrated to affect macro-pore connectivity in some soils (Larsbo et al., 2009). Glyphosate is the most used herbicide in agriculture. It is a zwitterion with three pKa values. In contrast to bentazone it is rarely freely dissolved in the soil water. In the pH range 4–8 the mono- and divalent anion is adsorbed to aluminium and ferric oxides (Borggaard and Gimsing, 2008). The leached glyphosate is likely to be colloidally adsorbed in wet soils with preferential flow (Vereecken, 2005). Degradation of glyphosate appears to be dependent on microbial activity (Borggaard and Gimsing, 2008). The objective of this communication is twofold: First, we re-evaluate monitoring data for the two representative herbicides bentazone and glyphosate by estimating median concentrations in groundwater through the utilization of all available monitoring data (including non-detect samples) (Helsel, 2006, 2012). We demonstrate that a reevaluation of monitoring data on this basis will provide a more balanced indication of both actual median herbicide concentration levels in groundwater as well as changes in concentration levels over time. Secondly, we discuss implications beyond Danish environmental policies.
2. Materials and methods We re-evaluated data from the monitoring report (Thorling et al., 2013) for the two representative herbicides, bentazone and glyphosate, by including all samples. Specifically, we included data covering the period 1995 to 2012 for bentazone and 2000 to 2012 for glyphosate. Since 2003, sampling was predominantly from monitoring points where the groundwater was formed after 1950. From 2007, sampling focus was on wells where previous samples had shown pesticide concentrations above detection limits. For the latter, monitoring points with no pesticides above detection limit were only sampled every third year (2007–2010), and only twice for 2011–2015. Thus the monitoring plan changed to an increased focus on “groundwater at risk” (Thorling et al., 2013). Consequently, the distribution of wells was not representative of the whole country, but targeted regions with a track record of high herbicide concentrations.
69
Pesticide concentrations were assumed to follow a log-normal distribution, i.e., concentrations were assumed to be normally distributed when transformed to the logarithmic scale (e.g. Gilbert, 1987; Helsel, 1990, 2012). Specifically, for each herbicide we fitted a parametric event-time model assuming a linear trend in time and log-normally distributed concentrations. Concentrations below the detection limit of 0.01 μg L−1 were treated as left-censored observations, i.e. the concentrations were not measured precisely and it was only known that their value is smaller than the detection limit (Helsel, 2006, 2012). Thus, for each herbicide a joint model that included monitoring data from all years was fitted; thus data from all years were used to estimate the standard deviation of a common log-normal distribution. In fact, the model may be viewed as a simple linear regression model of logarithmtransformed concentrations and with year as a quantitative explanatory variable except for the large proportion of left-censored observations. By means of back-transformation using the exponential function, estimated medians and percentile concentrations were obtained on the original scales. Approximate z-tests were used to evaluate the linear trends. Sensitivity analyses were carried out assuming log-logistic and Weibull distributions, which are both suitable for right-skewed data, but in practice less used than log-normal distribution (Cox and Oakes, 1984). Additionally, logistic regression models for the reduced binary endpoint detected/non-detected were also fitted; these analyses did not assume any distribution for the concentrations. A significance level of 5% was used. Statistical analysis was carried using R (R Core Team, 2014). 3. Results and discussion It is illustrative to compare estimated herbicide concentrations in the groundwater to the actual use of the herbicides. Sales of bentazone and glyphosate in Denmark are illustrated in Fig. 1. Field rates of pure bentazone products (Fig. 1A) were reduced by 33% of the original recommended rate in 1995 in order to protect the groundwater against potential contamination; and from 1995 bentazone use has been declining and is focused on crops such as maize, field peas, grass and clover for seed production as well as spring cereals with various under-sown crops. Due to increasing problems with Geranium sp. weeds in maize, use of bentazone is increasing in maize where it is the only active ingredient with high efficacy. A large proportion of glyphosate (Fig. 1B) is used pre-harvest in cereals, cruciferous crops and peas and preemergence in maize, potatoes and other slow germinating crops in order to effectively manage the first flush of weeds. It is also used extensively in reduced tillage cropping systems (conservation tillage). During the whole period the median trend for bentazone concentrations estimated on the basis of all samples increased 5% per year (p = 0.005; 95% CI: 2–9%). The estimated median trends are shown in Fig. 2A. By contrast results based on samples with concentrations above the detection limit showed a negative trend (p = 0.025) in medians (Thorling et al., 2013). Moreover, estimated medians from analyses with and without non-detect samples differed 104–105 fold. For glyphosate the discrepancy between estimated medians from the analyses with and without non-detect samples was also about 104–105 fold. The estimated trend line for glyphosate based on all samples (Fig. 2B) showed an increase by 11% (95% CI: 4–20%) per year (p = 0.004). This positive trend followed the total sold amount of glyphosate (Fig. 1B). However, if the year 2009 is omitted due extremely high concentrations among the actual observed concentration, the positive trend was weakened substantially (p = 0.16). In comparison the analysis without non-detects showed a positive trend in the medians (p = 0.017) (Thorling et al., 2013). Using alternative distributions of the concentrations (log-logistic and Weibull distributions) resulted in estimated medians in the range from 10− 5 to 10−8 for glyphosate compared to approximated 10− 6 for the assumed log-normal distribution (Fig. 2B) and from 10−5 to 10−7 for bentazone compared to approx. 10− 6 for the assumed
70
C.T. Hansen et al. / Science of the Total Environment 536 (2015) 68–71
Sold Bentazone
A
Bentazone
100
1e+01
7 12 18 23 10 12 16 20 14 12 23 29 25 27 25 25 22 24
A
75
Concentration ( µg l−1)
Ton
1e−02
50
Percentile 0.75 0.50 1e−05 0.25
25 1e−08
103 517 829 797 787 827 799 641 639 301 824 853 796 645 860 709 509 691
0 1990
1995
2000
2005
2010
1995
2000
2005
Year
2010
Year
Sold Glyphosate
B
Glyphosate
2000 1e+01
B
8
5
6
9
3
13
9
14
10
27
8
5
6
1500
Concentration ( µg l−1)
Ton
1e−02
1000
Percentile 0.75 0.50 1e−05 0.25
500 1e−08
0
837 782 788 769 630 813 847 800 703 639 509 638 691 1990
1995
2000
2005
2010
Year Fig. 1. Total sold amount of bentazone and glyphosate in Denmark. The black bars show the period analysed in this study. The restriction on bentazone (A) took effect in 1995 because of groundwater protection, which is reflected in the declining amount purchased. Glyphosate (B) has increased due to its use in pre-harvest cereals, conservation tillage and pre-emergence spraying in maize. The non-agricultural use of the two herbicides is negligible (https://www.middeldatabasen.dk/Middelvalg.asp).
log-normal distribution. Likewise, logistic regression analyses confirmed the significant trends over time with p = 0.002 and p = 0.01 for bentazone and glyphosate when including non-detect samples, respectively. From 2007 monitoring data were not based on randomly sampling, but focused on so-called “high risk groundwater”. Danish environmental policies have over the last decades been driven largely by concerns regarding groundwater contamination with pesticides. Since 2011, groundwater status presented by official monitoring reports (e.g. Thorling et al., 2013) have included graphs of median and mean concentrations without considering the sample below detection limits over time, corresponding to the upper regression lines in Fig. 2.
2000
2004
2008
2012
Year
Fig. 2. Bentazone concentrations in Danish groundwater (A) between 1995 and 2012 (Figur 39 page 107; Thorling et al., 2013). Glyphosate concentrations in Danish groundwater (B) between 2000 and 2012 (Figur 40 page 108, Thorling et al., 2013). The dots indicate median concentrations for samples above the detection limit. The fitted regression trends of medians are indicated by the grey solid lines. The upper horizontal broken line indicates the limit according to the EU Drinking Water Directive (Council Directive 98/83/EC, URL: http://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:31998L0083) of 0.1 μg L−1. The punctured horizontal line at 0.01 μg L−1 indicates the current detection limit. The grey solid line below the detection limit is the regression line for all samples. The redcoloured gradient indicates the range encompassing the middle 98% of the distribution of concentrations (range between 1% and 99% percentiles in the distribution of concentrations) estimated from the entire dataset. The numbers at top of the graphs are number of concentration above the detection limit; the numbers below are total numbers of measurements per year.
4. Conclusions Omitting non-detect samples resulted in heavily upwards biased estimated median pesticide concentrations, distorting the picture of
C.T. Hansen et al. / Science of the Total Environment 536 (2015) 68–71
the level of contamination of Danish groundwater. Furthermore, we have demonstrated that medians estimated on samples above the detection limit are not appropriate as indicators for temporal development in the groundwater contamination in general. Omitting non-detect samples significantly distorts public and political perception of the groundwater status. There has been little appreciation of the implications of exclusively basing calculations on samples with concentrations above the detection limit. In the EU context, it is imperative to report groundwater and drinking water contamination on an appropriate statistical basis, taking all samples into consideration. The methods used in this study yield robust results, which are insensitive to few outliers and the statistical analysis accommodated the realistic, right-skewed distribution of the concentrations. The current detection limit for both herbicides in soil water is 0.01 μg L− 1. The trend lines in Fig. 2 do not necessarily show a realistic trend because they are affected by the changed sampling strategy from 2007. After 2007 focus was directed to sample sites that previously had shown concentrations of the herbicide above the detection limit. This means that data were not randomly sampled and any trend in the development of herbicide load in the groundwater is biased. Our empirical results show that for bentazone the trend line based on samples excluding non-detects contradicts the trend line derived from all samples. In the case of glyphosate trend lines are corroborative. However, not too much reliance should be put on the trend line based on detects only as the sample size is very small, and if the single year with very high median glyphosate concentration is discarded no significant trend exists. In this study, we have limited our calculations to bentazone and glyphosate, but the same statistical methods should also be applied to other pesticides, xenobiotics and perhaps plant nutrient as well as toxic element concentrations (e.g., nitrate or arsenic compounds) in the soil water. It is a limitation that the distributional assumptions may not be strongly supported by the data. However, support for the chosen assumptions was found in the literature. In addition, we carried out several sensitivity analyses to ensure that our findings were not merely a consequence of the imposed model assumptions, which were in turn not sustained by the data. We considered analyses with other distributional assumptions as well as robust distribution-free analyses using logistic regression.
71
Conflicts of interest We confirm that there are no known conflicts of interest associated with this publication and that there has been no financial supports for this work that could have influenced its outcome. Acknowledgement We thank the Geological Survey of Denmark and Greenland for providing the data for this analysis. References Borggaard, O.K., Gimsing, A.L., 2008. Fate of glyphosate in soil and the possibility of leaching to ground and surface waters: a review. Pest Manag. Sci. 64, 441–456. Cox, D.R., Oakes, D., 1984. Analysis of Survival Data. CRC Press. Fram, M.S., Belitz, K., 2011. Occurrence and concentrations of pharmaceutical compounds in groundwater used for public drinking-water supply in California. Sci. Total Environ. 409, 3409–3417. Gilbert, R.O., 1987. Statistical Methods for Environmental Pollution Monitoring. Van Nostrand Reihold. Helsel, D.R., 1990. Less than obvious — statistical treatment of data below the detection limit. Environ. Sci. Technol. 24, 1766–1774. Helsel, D.R., 2006. Fabricating data: how substituting values for nondetects can ruin results and what can be done about it. Chemosphere 65, 2434–2439. Helsel, D.R., 2012. Statistics for Censored Environmental Data Using Minitab and R. 2nd ed. Wiley. Helsel, D.R., Hirsch, R.M., 2002. Statistical Methods in Water Resources. U.S. Geological Survey Techniques of Water-Resources Investigations, bk.4:chap.A3 (510 pp. Available at: URL: http://water.usgs.gov/pubs/twri/twri4a3/). Köck-Schulmeyer, M., Ginebreda, A., Postigo, C., Garrido, T., Fraile, J., López de Alda, M., Barceló, D., 2014. Four-year advanced monitoring program of polar pesticides in groundwater of Catalonia (NE-Spain). Sci. Total Environ. 470–471, 1087–1098. Kudsk, P., Jensen, J.E., 2014. Experiences with Implementation and Adoption of Integrated Pest Management in Denmark. In: Pimentel, David, Peshin, Rajinder (Eds.), Integrated Pest Management vol. 4, pp. 467–485. Larsbo, M., Stenström, J., Etana, A., Börjesson, E., Jarvis, N.J., 2009. Herbicide sorption, degradation, and leaching in three Swedish soils under long-term conventional and reduced tillage. Soil Tillage Res. 105, 200–208. R Core Team, 2014. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria (URL http://www.R-project.org/). Thorling, L., Brüsch, W., Hansen, B., Larsen, C.L., Mielby, S., Troldborg, L., Sørensen, B.L., 2013. Grundvand. Status og udvikling 1989–2012. Teknisk rapport, GEUS 2013, English Summary (URL: http://www.geus.dk/publications/grundvandsovervaagning/ index.htm). Vereecken, H., 2005. Mobility and leaching of glyphosate: a review. Pest Manag. Sci. 61, 1139–1151.
Contents lists available at ScienceDirect
Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv
Short Communication
Re-evaluation of groundwater monitoring data for glyphosate and bentazone by taking detection limits into account Claus Toni Hansen a, Christian Ritz b, Daniel Gerhard c, Jens Erik Jensen d, Jens Carl Streibig e,⁎ a
Ålegårdsvej 29, 2740 Skovlunde, Denmark Nutrition, Exercise and Sports, University of Copenhagen, Rolighedsvej 26, 1958 Frederiksberg C, Denmark Mathematics and Statistics, University of Canterbury, Private Bag 4800, Christchurch 8140, New Zealand d SEGES P/S, Crops & Environment, Agro Food Park 15, 8200 Aarhus N, Denmark e Plant and Environmental Sciences, University of Copenhagen, Hoejbakkegaard Alle 13, 2630 Taastrup, Denmark b c
H I G H L I G H T S
G R A P H I C A L
• It is imperative to include all samples – including those falling below detection levels. • Samples with pesticide concentrations below detection limits result in leftcensored observations. • Groundwater pesticide medians are 104–105 lower when including than when excluding “non-detect” • Excluding “non-detect” samples significantly overestimates pesticide load in groundwater.
The size of glyphosate contamination in Danish groundwater without including samples with glyphosate below detection level is 10,000 to 100,000 higher than doing a parametric event-time model, where non-detect samples are left censored. The trend lines are rather similar. The broken lines are the EU drinking water directive maximum value 0.1 μg l− 1 and the dotted line the detection limit of 0.01 μg L− 1. Numbers are number of samples above detection line and total number of samples.
a r t i c l e
a b s t r a c t
i n f o
Article history: Received 28 May 2015 Received in revised form 8 July 2015 Accepted 8 July 2015 Available online xxxx Editor: D. Barcelo Keywords: Groundwater Herbicides Drinking-water supply Contaminants
⁎ Corresponding author. E-mail address: [email protected] (J.C. Streibig).
http://dx.doi.org/10.1016/j.scitotenv.2015.07.047 0048-9697/© 2015 Elsevier B.V. All rights reserved.
A B S T R A C T
Current regulatory assessment of pesticide contamination of Danish groundwater is exclusively based on samples with pesticide concentrations above detection limit. Here we demonstrate that a realistic quantification of pesticide contamination requires the inclusion of “non-detect” samples i.e. samples with concentrations below the detection limit, as left-censored observations. The median calculated pesticide concentrations are shown to be reduced 104 to 105 fold for two representative herbicides (glyphosate and bentazone) relative to the median concentrations based upon observations above detection limits alone. © 2015 Elsevier B.V. All rights reserved.
C.T. Hansen et al. / Science of the Total Environment 536 (2015) 68–71
1. Introduction Since 1989, Danish groundwater has annually been analysed for pesticide residues in order to safeguard that the EU drinking water directive limit of 0.1 μg pesticide L−1 is met (Thorling et al., 2013). By contrast to other countries in the EU, the Danish drinking water is minimally treated groundwater. Therefore, the result of this long time monitoring of pesticide residues in groundwater is a crucial factor in the various Pesticide Action Plans agreed upon in the Danish Parliament in order to protect the groundwater (Kudsk and Jensen, 2014). However, in contrast to normally accepted practice the samples below the detection limits (non-detect samples) were excluded from the statistical analysis (Thorling et al., 2013). As examples of normal practice using non-detect samples, Köck-Schulmeyer et al. (2014) in a four-year monitoring of polar pesticides in groundwater included samples below the detection limit as a surrogate value of half the detection limits for the pesticides, although they did not explicitly describe the statistical method used. Fram and Belitz (2011) explicitly described how the data, including a surrogate concentration for samples below the detection limit, were analysed by the method outlined by Helsel and Hirsch (2002). The herbicides bentazone and glyphosate are important in a number of crops and they can be found in the upper groundwater. Thus, they are appropriate representative pesticides for the evaluations carried out in this study. The weakly acidic herbicide, bentazone, is mobile and considered moderately persistent in soil. The Freundlich adsorption isotherm does not vary much among soils with clay contents, whether it is used under conventional or reduced tillage cultivation. The first order degradation constant in the soils does not significantly change in response to clay soils. However, tillage system has been demonstrated to affect macro-pore connectivity in some soils (Larsbo et al., 2009). Glyphosate is the most used herbicide in agriculture. It is a zwitterion with three pKa values. In contrast to bentazone it is rarely freely dissolved in the soil water. In the pH range 4–8 the mono- and divalent anion is adsorbed to aluminium and ferric oxides (Borggaard and Gimsing, 2008). The leached glyphosate is likely to be colloidally adsorbed in wet soils with preferential flow (Vereecken, 2005). Degradation of glyphosate appears to be dependent on microbial activity (Borggaard and Gimsing, 2008). The objective of this communication is twofold: First, we re-evaluate monitoring data for the two representative herbicides bentazone and glyphosate by estimating median concentrations in groundwater through the utilization of all available monitoring data (including non-detect samples) (Helsel, 2006, 2012). We demonstrate that a reevaluation of monitoring data on this basis will provide a more balanced indication of both actual median herbicide concentration levels in groundwater as well as changes in concentration levels over time. Secondly, we discuss implications beyond Danish environmental policies.
2. Materials and methods We re-evaluated data from the monitoring report (Thorling et al., 2013) for the two representative herbicides, bentazone and glyphosate, by including all samples. Specifically, we included data covering the period 1995 to 2012 for bentazone and 2000 to 2012 for glyphosate. Since 2003, sampling was predominantly from monitoring points where the groundwater was formed after 1950. From 2007, sampling focus was on wells where previous samples had shown pesticide concentrations above detection limits. For the latter, monitoring points with no pesticides above detection limit were only sampled every third year (2007–2010), and only twice for 2011–2015. Thus the monitoring plan changed to an increased focus on “groundwater at risk” (Thorling et al., 2013). Consequently, the distribution of wells was not representative of the whole country, but targeted regions with a track record of high herbicide concentrations.
69
Pesticide concentrations were assumed to follow a log-normal distribution, i.e., concentrations were assumed to be normally distributed when transformed to the logarithmic scale (e.g. Gilbert, 1987; Helsel, 1990, 2012). Specifically, for each herbicide we fitted a parametric event-time model assuming a linear trend in time and log-normally distributed concentrations. Concentrations below the detection limit of 0.01 μg L−1 were treated as left-censored observations, i.e. the concentrations were not measured precisely and it was only known that their value is smaller than the detection limit (Helsel, 2006, 2012). Thus, for each herbicide a joint model that included monitoring data from all years was fitted; thus data from all years were used to estimate the standard deviation of a common log-normal distribution. In fact, the model may be viewed as a simple linear regression model of logarithmtransformed concentrations and with year as a quantitative explanatory variable except for the large proportion of left-censored observations. By means of back-transformation using the exponential function, estimated medians and percentile concentrations were obtained on the original scales. Approximate z-tests were used to evaluate the linear trends. Sensitivity analyses were carried out assuming log-logistic and Weibull distributions, which are both suitable for right-skewed data, but in practice less used than log-normal distribution (Cox and Oakes, 1984). Additionally, logistic regression models for the reduced binary endpoint detected/non-detected were also fitted; these analyses did not assume any distribution for the concentrations. A significance level of 5% was used. Statistical analysis was carried using R (R Core Team, 2014). 3. Results and discussion It is illustrative to compare estimated herbicide concentrations in the groundwater to the actual use of the herbicides. Sales of bentazone and glyphosate in Denmark are illustrated in Fig. 1. Field rates of pure bentazone products (Fig. 1A) were reduced by 33% of the original recommended rate in 1995 in order to protect the groundwater against potential contamination; and from 1995 bentazone use has been declining and is focused on crops such as maize, field peas, grass and clover for seed production as well as spring cereals with various under-sown crops. Due to increasing problems with Geranium sp. weeds in maize, use of bentazone is increasing in maize where it is the only active ingredient with high efficacy. A large proportion of glyphosate (Fig. 1B) is used pre-harvest in cereals, cruciferous crops and peas and preemergence in maize, potatoes and other slow germinating crops in order to effectively manage the first flush of weeds. It is also used extensively in reduced tillage cropping systems (conservation tillage). During the whole period the median trend for bentazone concentrations estimated on the basis of all samples increased 5% per year (p = 0.005; 95% CI: 2–9%). The estimated median trends are shown in Fig. 2A. By contrast results based on samples with concentrations above the detection limit showed a negative trend (p = 0.025) in medians (Thorling et al., 2013). Moreover, estimated medians from analyses with and without non-detect samples differed 104–105 fold. For glyphosate the discrepancy between estimated medians from the analyses with and without non-detect samples was also about 104–105 fold. The estimated trend line for glyphosate based on all samples (Fig. 2B) showed an increase by 11% (95% CI: 4–20%) per year (p = 0.004). This positive trend followed the total sold amount of glyphosate (Fig. 1B). However, if the year 2009 is omitted due extremely high concentrations among the actual observed concentration, the positive trend was weakened substantially (p = 0.16). In comparison the analysis without non-detects showed a positive trend in the medians (p = 0.017) (Thorling et al., 2013). Using alternative distributions of the concentrations (log-logistic and Weibull distributions) resulted in estimated medians in the range from 10− 5 to 10−8 for glyphosate compared to approximated 10− 6 for the assumed log-normal distribution (Fig. 2B) and from 10−5 to 10−7 for bentazone compared to approx. 10− 6 for the assumed
70
C.T. Hansen et al. / Science of the Total Environment 536 (2015) 68–71
Sold Bentazone
A
Bentazone
100
1e+01
7 12 18 23 10 12 16 20 14 12 23 29 25 27 25 25 22 24
A
75
Concentration ( µg l−1)
Ton
1e−02
50
Percentile 0.75 0.50 1e−05 0.25
25 1e−08
103 517 829 797 787 827 799 641 639 301 824 853 796 645 860 709 509 691
0 1990
1995
2000
2005
2010
1995
2000
2005
Year
2010
Year
Sold Glyphosate
B
Glyphosate
2000 1e+01
B
8
5
6
9
3
13
9
14
10
27
8
5
6
1500
Concentration ( µg l−1)
Ton
1e−02
1000
Percentile 0.75 0.50 1e−05 0.25
500 1e−08
0
837 782 788 769 630 813 847 800 703 639 509 638 691 1990
1995
2000
2005
2010
Year Fig. 1. Total sold amount of bentazone and glyphosate in Denmark. The black bars show the period analysed in this study. The restriction on bentazone (A) took effect in 1995 because of groundwater protection, which is reflected in the declining amount purchased. Glyphosate (B) has increased due to its use in pre-harvest cereals, conservation tillage and pre-emergence spraying in maize. The non-agricultural use of the two herbicides is negligible (https://www.middeldatabasen.dk/Middelvalg.asp).
log-normal distribution. Likewise, logistic regression analyses confirmed the significant trends over time with p = 0.002 and p = 0.01 for bentazone and glyphosate when including non-detect samples, respectively. From 2007 monitoring data were not based on randomly sampling, but focused on so-called “high risk groundwater”. Danish environmental policies have over the last decades been driven largely by concerns regarding groundwater contamination with pesticides. Since 2011, groundwater status presented by official monitoring reports (e.g. Thorling et al., 2013) have included graphs of median and mean concentrations without considering the sample below detection limits over time, corresponding to the upper regression lines in Fig. 2.
2000
2004
2008
2012
Year
Fig. 2. Bentazone concentrations in Danish groundwater (A) between 1995 and 2012 (Figur 39 page 107; Thorling et al., 2013). Glyphosate concentrations in Danish groundwater (B) between 2000 and 2012 (Figur 40 page 108, Thorling et al., 2013). The dots indicate median concentrations for samples above the detection limit. The fitted regression trends of medians are indicated by the grey solid lines. The upper horizontal broken line indicates the limit according to the EU Drinking Water Directive (Council Directive 98/83/EC, URL: http://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:31998L0083) of 0.1 μg L−1. The punctured horizontal line at 0.01 μg L−1 indicates the current detection limit. The grey solid line below the detection limit is the regression line for all samples. The redcoloured gradient indicates the range encompassing the middle 98% of the distribution of concentrations (range between 1% and 99% percentiles in the distribution of concentrations) estimated from the entire dataset. The numbers at top of the graphs are number of concentration above the detection limit; the numbers below are total numbers of measurements per year.
4. Conclusions Omitting non-detect samples resulted in heavily upwards biased estimated median pesticide concentrations, distorting the picture of
C.T. Hansen et al. / Science of the Total Environment 536 (2015) 68–71
the level of contamination of Danish groundwater. Furthermore, we have demonstrated that medians estimated on samples above the detection limit are not appropriate as indicators for temporal development in the groundwater contamination in general. Omitting non-detect samples significantly distorts public and political perception of the groundwater status. There has been little appreciation of the implications of exclusively basing calculations on samples with concentrations above the detection limit. In the EU context, it is imperative to report groundwater and drinking water contamination on an appropriate statistical basis, taking all samples into consideration. The methods used in this study yield robust results, which are insensitive to few outliers and the statistical analysis accommodated the realistic, right-skewed distribution of the concentrations. The current detection limit for both herbicides in soil water is 0.01 μg L− 1. The trend lines in Fig. 2 do not necessarily show a realistic trend because they are affected by the changed sampling strategy from 2007. After 2007 focus was directed to sample sites that previously had shown concentrations of the herbicide above the detection limit. This means that data were not randomly sampled and any trend in the development of herbicide load in the groundwater is biased. Our empirical results show that for bentazone the trend line based on samples excluding non-detects contradicts the trend line derived from all samples. In the case of glyphosate trend lines are corroborative. However, not too much reliance should be put on the trend line based on detects only as the sample size is very small, and if the single year with very high median glyphosate concentration is discarded no significant trend exists. In this study, we have limited our calculations to bentazone and glyphosate, but the same statistical methods should also be applied to other pesticides, xenobiotics and perhaps plant nutrient as well as toxic element concentrations (e.g., nitrate or arsenic compounds) in the soil water. It is a limitation that the distributional assumptions may not be strongly supported by the data. However, support for the chosen assumptions was found in the literature. In addition, we carried out several sensitivity analyses to ensure that our findings were not merely a consequence of the imposed model assumptions, which were in turn not sustained by the data. We considered analyses with other distributional assumptions as well as robust distribution-free analyses using logistic regression.
71
Conflicts of interest We confirm that there are no known conflicts of interest associated with this publication and that there has been no financial supports for this work that could have influenced its outcome. Acknowledgement We thank the Geological Survey of Denmark and Greenland for providing the data for this analysis. References Borggaard, O.K., Gimsing, A.L., 2008. Fate of glyphosate in soil and the possibility of leaching to ground and surface waters: a review. Pest Manag. Sci. 64, 441–456. Cox, D.R., Oakes, D., 1984. Analysis of Survival Data. CRC Press. Fram, M.S., Belitz, K., 2011. Occurrence and concentrations of pharmaceutical compounds in groundwater used for public drinking-water supply in California. Sci. Total Environ. 409, 3409–3417. Gilbert, R.O., 1987. Statistical Methods for Environmental Pollution Monitoring. Van Nostrand Reihold. Helsel, D.R., 1990. Less than obvious — statistical treatment of data below the detection limit. Environ. Sci. Technol. 24, 1766–1774. Helsel, D.R., 2006. Fabricating data: how substituting values for nondetects can ruin results and what can be done about it. Chemosphere 65, 2434–2439. Helsel, D.R., 2012. Statistics for Censored Environmental Data Using Minitab and R. 2nd ed. Wiley. Helsel, D.R., Hirsch, R.M., 2002. Statistical Methods in Water Resources. U.S. Geological Survey Techniques of Water-Resources Investigations, bk.4:chap.A3 (510 pp. Available at: URL: http://water.usgs.gov/pubs/twri/twri4a3/). Köck-Schulmeyer, M., Ginebreda, A., Postigo, C., Garrido, T., Fraile, J., López de Alda, M., Barceló, D., 2014. Four-year advanced monitoring program of polar pesticides in groundwater of Catalonia (NE-Spain). Sci. Total Environ. 470–471, 1087–1098. Kudsk, P., Jensen, J.E., 2014. Experiences with Implementation and Adoption of Integrated Pest Management in Denmark. In: Pimentel, David, Peshin, Rajinder (Eds.), Integrated Pest Management vol. 4, pp. 467–485. Larsbo, M., Stenström, J., Etana, A., Börjesson, E., Jarvis, N.J., 2009. Herbicide sorption, degradation, and leaching in three Swedish soils under long-term conventional and reduced tillage. Soil Tillage Res. 105, 200–208. R Core Team, 2014. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria (URL http://www.R-project.org/). Thorling, L., Brüsch, W., Hansen, B., Larsen, C.L., Mielby, S., Troldborg, L., Sørensen, B.L., 2013. Grundvand. Status og udvikling 1989–2012. Teknisk rapport, GEUS 2013, English Summary (URL: http://www.geus.dk/publications/grundvandsovervaagning/ index.htm). Vereecken, H., 2005. Mobility and leaching of glyphosate: a review. Pest Manag. Sci. 61, 1139–1151.