SEARCH

SEARCH BY CITATION

Keywords:

  • Herbicide;
  • Long-term trend;
  • Mann–Kendall test;
  • River load;
  • Trend significance

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Many policies have been established and actions have been taken to reduce pesticide pollution of surface waters. However, the effectiveness of these initiatives was rarely tested on an empirical basis. This study suggests that evidence on the policy effects can be evaluated by analyzing the temporal changes in the loads rather than the concentrations of active ingredients in rivers. It is shown that the long-term change of pesticide emissions into surface waters can be tested statistically by the number of upward versus downward trends in river load. To evaluate the situation in Germany, 57 concentration time series of 14 herbicide substances at 7 river monitoring stations with a minimum of 24 analyses per year were assembled and annual substance river loads were calculated. The longest time course was 17 years (1990–2006). The significance of trends of data rows was analyzed by an univariate Mann–Kendall test that evaluates 27 (47.4%) of the 57 time series as statistically significant downward trended. It took a period of 10 years and longer before the high annual atrazine and simazine loads measured in the years 1990–1991 had been diminished to a drastically lowered level after ban of the herbicides. Data are available on the yearly consumption of 8 substances used in German agriculture. A total of 36 time series for this subset were tested with a partial Mann–Kendall test with the consumption as covariance factor, which reduces the number of significant trends in river load noticeably. Based on this test, only 7 (19.4% of 36) declining time series remain. As a result, the intended effect of measures to reduce surface water contamination by the use of pesticides seems to be only partially successful, however, the database to justify this statement is small. For the water monitoring strategies in Germany, it is recommend to enhance the sampling frequency at river stations. A minimum of a semimonthly sampling interval would facilitate the calculation of valuable annual river loads and would therefore allow a pronounced validation of the long-term change of pesticide emission into rivers. Integr Environ Assess Manag 2012; 8: 543–552. © 2012 SETAC


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Ever since the 1980s, awareness of the ecological risks of pesticide use has risen substantially, and in parallel, improved analytical methods increased the number of substance findings in surface waters considerably. As a result, numerous programs were initiated in the aftermath, and various measures were taken to minimize the potential hazards to aquatic biocoenosis caused by the application of pesticides. In Germany, the very persistent and mobile substances atrazine and simazine have been banned since 1990. For other substances, application restrictions were issued or have been strengthened. For field sprayers, an obligatory biannual inspection was implemented in 1993 (Osteroth 2004). Furthermore advanced sprayer technology has been launched, e.g., drift-reducing nozzles and equipment for on-field sprayer cleaning, helping farmers reduce pesticide release coupled with sprayer handling.

In the past decade, activities have been focused more on point sources as a relevant origin of pesticide entry into surface waters. Information campaigns were accordingly addressed to farmers to enhance their environmental awareness and to provide best management practices to reduce the contamination of waters by plant protection products. The H2OK campaign (no drop into the gully; IVA 2010a) and the PAMIRA campaign (refuse packaging collection at point of sale; IVA 2010b) are 2 prominent examples in Germany. The Train the Operators to Prevent Pollution from Point Sources (TOPPS) project (Roettele 2008) forms a transnational platform dealing with this problem. The European Commission (2009) recently established a framework for community action to achieve a sustainable use of pesticides.

Considering the long-lasting list of programs and activities, the question has to be asked whether their intended effects have been reached or not. The German National Action Plan (BMELV 2008) uses the risk indicator SYNOPS (Gutsche and Strassmeyer 2007) to evaluate the effects of national policies toward the sustainable use of pesticides. Risk indicators are helpful tools to characterize the probability of pesticide losses into waters and to evaluate the relevance of substances parameters, soil properties, climate, application pattern, etc., on this process. However, a risk indicator quantifies only the probability of water contamination but does not reflect the factual substance emission into the river network. Various pesticide risk indicators are established (Reus et al. 2002), but the correlation between a respective risk indicator and the “true values,” i.e., the river contamination in reality has been not ascertained up to now.

To improve water quality policies, empirical evidence is needed, and measures have to be evaluated based on hard facts regarding data on water contamination via pesticides. However, concentration data deliver information on water quality and the exposure of the aquatic system to pesticides, yet do not indicate directly the quantity of pesticide entries into surface waters. In terms of emission intensity that is addressed by mitigation measures and schemes, only the mass fluxes of a substance transported in a river give valuable information.

Long-term observations of pesticide loads in rivers and/or trend analyses of pesticide occurrences in surface waters are sparsely reported in literature. The occurrence of 38 pesticides in a small agricultural catchment in southern Sweden was correlated to the amounts used in the catchment (Kreuger 1998). The total pesticide load in water decreased markedly from 1990 to 1996 in accordance with decreasing amounts applied during spring and early summer. For selected pesticides and streams in the United States, Vecchia et al. (2008) developed a regression model to assess the variability and long-term trends in concentration. In 8 of 10 modeled pesticide site combinations, the trend over two 6-year periods, starting in 1995 and 2001, respectively, was significantly downward. In northeastern Greece, 8 sites along 3 rivers were sampled from 1999 to 2007, but no trend was reported for the 28 compounds detected frequently (Vryzas et al. 2009). The concentrations and loading rates of pesticides used in paddy fields were investigated over a period of 5 years in the Seta River (Japan) without detecting any interannual tendency (Sudo et al. 2002). In the Venetian Lagoon (Italy) during 3 sampling campaigns in 1987, 1993, and 1998, the concentrations of organochlorine pesticides were analyzed in the uppermost sediment layer collected at 25 sites (Secco et al. 2005). The results statistically indicate that pesticide concentrations in the sediments tended to decrease during each period of time. For the annual load of 3 particle-bound organic pollutants, Pohlert et al. (2011) detected decreasing trends during the period 1995–2008 at 2 of 8 monitoring stations along the Rhine using the Mann–Kendall (MK) test.

The study presented here gathers all long-lasting data series on herbicide river loads available for Germany and tests the significance of their trends. The number of decreasing river load trends forms the basis for evaluating long-term changes in herbicide release and/or leakage into the respective river systems and for judging the effectiveness of efforts to reduce pesticide pollution in surface waters.

METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Data criteria and inquiry

Water quality monitoring in Germany is the responsibility of the water authorities of the federal states (Bundesland). Furthermore, water companies using bank filtration for drinking water supply analyze surface waters on pesticides frequently. These institutions transferred pesticide concentration data for 26 river monitoring sites in Germany. To estimate annual pesticide loads in rivers, the following selection criteria for concentration data were specified: 1) a minimum of 24 analyses per year (2 samples per month or biweekly) in the case of grab samples, or 2) composite sampling over the entire year or at least during the period of main pesticide entries into the river system. Furthermore stations and/or substances with a monitoring duration shorter than 7 years or with large data gaps were excluded from further interpretation. The final database consists of 57 time series of annual river pesticide loads covering a total of 14 herbicide substances (no data on fungicides and insecticides were available) at 7 stations (Figure 1), the longest time span is 17 years (1990–2006). Daily discharge data for the sampling stations was part of the concentration recordings or was taken from the German Federal Institute of Hydrology (Bundesanstalt für Gewässerkunde, Koblenz).

thumbnail image

Figure 1. River basins and river quality monitoring stations in Germany with pesticide load time series used for statistical analysis (upper case characters correspond to the description in Table 1).

Download figure to PowerPoint

Table 1 presents some characteristics of the monitoring sites and river basins, whereas Table 2 lists the substances together with the number of sites at which each substance was measured. The time series of the metabolite desethyl-atrazine are not considered in the trend analyses, because the river load of a metabolite is not an independent variable in terms of statistical analysis, and the significance of its trend cannot be counted for the evaluation of trend frequencies.

Table 1. Characteristics of the water quality stations in Germany with pesticide concentration data sets used for trend analysisa
Stationb (km)River basinInstitutionBasin area (km2)Longest time seriesNr substances analyzedcNr analyses per yeard
  • a

    See Methods for selection criteria.

  • b

    Upper case letter corresponds to site in Figure 1.

  • c

    Number of substances used for statistical evaluation, for some substances the time series is shorter than the length indicated in column 6.

  • d

    Minor figure: minimal number of analyses for a substance in 1 of the sampling years; major figure: regular number of analyses per year at this station; (g), grab sampling; (c), continuous sampling.

A: Köln (685.8)RhineRheinEnergie144 2321990–20061395–104 (g)
B: Bischofsheim (0.8)MainHLUG27 1401990–20061229–35 (c)
C: Frankfurt-Nied. (37.2)MainMainova AG24 7641991–2006924–26 (g)
D: Schweinfurt (330.8)MainLfU Bayern12 7151991–2003321–24 (g)
E: Frankfurt-Praunh. (6.5)NiddaMainova AG19201996–20061025–26 (g)
F: Hengsen (108.4)RuhrWW Westfalen19491990–2006446–52 (g)
G: Westhofen (95.8)RuhrWW Westfalen20981990–2006642–52 (g)
Table 2. Evaluated substances and number of stations with time series
SubstanceNr stations
  • a

    Time series of the metabolite desethyl-atrazine are not to be considered in the trend analyses but for reasons of information data are given in Tables S1 to S7.

2,4-D2
2,4-DP4
Atrazine6
desethyl-Atrazine(5)a
Bentazone4
Chlortoluron4
Diuron5
Isoproturon6
MCPA5
Mecoprop5
Metabenzthiazuron1
Metazachlor3
Metolachlor1
Simazine4
Terbuthylazine7

In addition to the stations presented in Table 1, 3 more stations in the Weser river basin (Verden, Porta Westfalica, and Wahnhausen) together with 34 time series conducted from 1996 to 2000 were evaluated. However, from these time courses the trend was statistically significant in only 1 case, which is mainly an effect of the limitations of the MK test for short time series. Thus these 34 time series are not considered in the overall statistical trend evaluation.

Table 3. Yearly consumption of plant protection products applied to agricultural land in Germany, market panel results of Kleffmann GmbH (Lüdinghausen, Germany)a
SubstanceYear
19961997199819992000200120022003200420052006
  • a

    Index values relative to the average consumption in 1999–2001 (baseline = 100).

  • b

    Chlortoluron had no registration from 2004 to 2006.

  • c

    Sum of mecoprop and mecoprop-P consumption.

2,4-D18414112211987943355675461
Bentazone7590114120958585112926761
Chlorotoluronb1681601809712381336110
Isoproturon105122127103107908163646666
MCPA981091079110510411811312811695
Mecopropc11613112711299898589604742
Metazachlor88819110295103115126127134140
Terbuthylazine9911411210193106104107128134130

Annual river load calculation

Estimating the annual flux of solutes in rivers is an intensively discussed issue in hydrological sciences, and numerous approaches have been introduced over the past decades (Gilroy et al. 1990, Hooper et al. 2001, Phillips et al. 1999). In the study presented here, the total annual load Lyear of a pesticide at a station was calculated as

  • equation image(1)

where K is a dimension conversion factor, Qyear is the total annual discharge, Ci and Qi are the instantaneous values of substance concentration and discharge, respectively, at the time of sampling, and ns is the number of samples. Analytical results below the limit of quantification (LOQ) are set to zero, thus the calculation of the annual load represents a minimum estimation. This approach equates directly to the basic definition of flux and corresponds to Method No. 18 described by Phillips et al. (1999). A more refined flux estimation approach as presented for the pesticide measurements of the US Geological Survey's National Stream Quality Accounting Network (Hooper et al. 2001) is not possible due to the lack of high-discharge sampling periods in the German data set.

Uncertainties in pesticide river load calculation

Uncertainties in the estimation of annual solute loads in rivers at the point of sampling arise during all steps of the procedure: discharge measurement, sampling frequency, sample storage, laboratory analysis, and the annual load calculation method (Harmel et al. 2006; Johnes 2007). Topping (1972) proposes a root mean square error propagation method to estimate the cumulative uncertainty for river load mass calculation

  • equation image(2)

where Ep is the probable total range in error (± %), n is the number of sources of potential error, and Ei are the ranges in error of the individual sources of error. For pesticide load data used for the study presented here in the following the range of the error sources river discharge, sampling frequency and calculation method, and quantification uncertainty is estimated separately.

The German water authority responsible for river gauging evaluates the uncertainty of its river discharge data by ± 5% (95% confidence interval; LAWA 1997). The combined effects of sampling frequency and the annual load interpolation method on uncertainty are explained for the example of P river loads in 17 catchments in the United Kingdom by Johnes (2007). Assuming that the temporal dynamic and interbasin variation of herbicide occurrence in rivers is similar to total P then the probable range in error for sampling frequency and the annual load interpolation method is assessed roughly as −25%/ + 50% for biweekly sampling based on the results by Johnes (2007) who calculated the effect of sampling frequency for various intervals.

Pesticide concentration analysis for water quality control by water authorities and water companies in Germany generally has to meet the requirements of good laboratory practice. The analytical methods for pesticide substance detection are specified in the German standard methods for the examination of water, wastewater and sludge (GDCh 1981). All institutions whose analysis results are respected for the trend tests presented here regularly have to pass an interlaboratory quality control. The guideline for interlaboratory comparison in Germany demands that 80% of the substance concentration results of an individual laboratory have to fall within the tolerance range −12.5%/ + 25% (LANUV 2011) of the true value (i.e., the corrected mean from the results of all labs) if the laboratory is certified for state measurements. The values of −12.5%/ + 25% are used as probable ranges in error for all uncertainties resulting from laboratory analysis.

The pesticides of concern in this study exist predominantly in the dissolved phase, so an underestimation due to sampling and analysis of filtered water only is insignificant. Possible effects on pesticide concentration due to improper preservation and storage of water samples are not quantified in literature, thus no figure can be derived for this error component.

Introducing the individual ranges in error as specified, the total range in error Ep is finally calculated as −17%/ + 33% according to Equation 1. In addition to these load calculation uncertainties due to sampling, pesticides are subject to in-stream losses by physical, chemical, and biological transformation processes. Based on the characteristics of “standard” stream combined with the physico-chemical pesticide properties Capel et al. (2001) presents a methodology to assess the potential in-stream losses and to recalculate the probable overall mass of a substance that entered the surface water system upstream the point of measurement. In the study presented here, we abstain from this approach because for trend analysis by the MK test, but the ratios between the annual loads of a time series (not the absolute differences) are relevant. The correction of the measured loads at a given monitoring station by a (constant) factor for in-stream losses do not change the load ratios between the years.

MK test of time series trend

Each time series of substance annual loads was tested to see if a statistically significant increasing or decreasing trend exists. The evaluation was carried out using the univariate MK test for the detection of trend in time series. For time courses with available data on yearly substance consumption (see below), a partial MK test was additionally carried out to correct the test result by the influence of the covariate consumption.

The MK test is used widely in environmental science because it is a simple, robust, and nonparametric test that can cope with missing values and values below a detection limit. Since its initial description by Mann (1945), the test has been extended to include seasonality (Hirsch et al. 1982), multiple monitoring sites (Lettenmaier 1988), and covariates representing natural fluctuations (Libiseller and Grimvall 2002). The univariate MK test sums up the sign of data differences rather than evaluating the data values themselves (Gilbert 1987). The MK score S for a time series {xk, k = 1,2,…, n} of data is defined as

  • equation image(3)

where

  • equation image(4)

for all xj, xk with j < k. If no ties between the observations exist and no trend is present in the time series, the test statistic is asymptotically normally distributed with

  • equation image(5)

where E(S) is the expectancy value of S and n is the number of data points in the time series. The initial value of the MK test statistic, the score S, is assumed to be 0 (i.e., no trend). The normalized test statistic Z is computed as follows

  • equation image(6)

The trend is said to be downward (upward) if Z is negative (positive), and the computed probability is greater than the level of significance. If the p value is less than the level of significance, there is no trend. The test decision gives only information about the existence of a significant trend but no figure to describe its strength or slope.

Although the univariate MK test proves only the responding variable itself, the partial MK test detects a trend with correction for an influencing variable. A derivation of the partial MK test is presented in detail by Libiseller and Grimvall (2002), who remark that the MK test adjusted by a covariate can result in a substantial gain of power. In a partial MK test the annual substance consumption in Germany (see below) and the annual river discharge at the monitoring stations were introduced as covariates. For an analysis of pesticide river loads, univariate and partial (bivariate) MK test statistics are generated with the program MULTMK/PARTMK (Libiseller 2002).

Data on pesticide consumption

An official cadaster on pesticide use in agriculture and other sectors does not exist for Germany. The market research company Kleffmann GmbH (Lüdinghausen, Germany) regularly collects a panel survey that provides representative data on the application of plant protection products to agricultural land in Germany on a yearly basis (data not published). For the study presented here, Kleffmann GmbH forwarded panel results on the consumption of 8 substances during the time period 1996–2006 totaled for Germany (Table 3). To protect its commercial interests, Kleffmann GmbH transferred the consumption data not as absolute quantities but as index values, relative to the average 1999–2001 that was set as an index baseline = 100. For the time period 1996–2001 the 4 substances 2,4-D, chlortoluron, isoproturon, and mecoprop showed a significant downtrend in consumption according to a univariate MK test; the changes in bentazone, MCPA and metazachlor are not significant, and the terbuthylazine usage increased significantly over the 11-year period.

The Kleffmann market panel does not cover the entire list of substances in Table 2. For active ingredients banned in Germany such as atrazine and simazine, no data were requested from the farmers; diuron is mainly applied on urban and private land and consequently is not subject of an agricultural market panel, and some additional substances are excluded due to limited survey capacity.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

The annual loads for individual substances, stations, and years calculated according to Eq. 1 are provided as Supplemental Data (Tables S1–S7). Figure 2 presents graphs of the river load time series for 11 substances (using data from at least 3 stations) together with the yearly consumption in Germany (only from 1996 to 2006). To allow comparability among substances and river stations, all load and consumption data have been standardized and expressed as index values relative to the mean period of 1999–2001, set to 100, for each substance at its respective station. Obviously the time series for the majority of herbicides at most stations are characterized by high fluctuations. A qualitative interpretation of the graphs suggests that the fluctuations of substances such as 2,4-DP, diuron, isoproturon, and mecoprop were stronger in the 1990s and have leveled out during recent years. Other herbicides (e.g., bentazone, MCPA (2-methyl-4-chlorophenoxyacetic acid), metazachlor, terbuthylazine) show no distinctive shift in their interannual variations.

thumbnail image

Figure 2. Time series for annual river loads for 11 herbicides at sampling stations in Germany (cf. Table 1) together with yearly consumption of the respective substances. All figures as relative values (baseline = 100, which is the average load and consumption from 1999 to 2001).

Download figure to PowerPoint

Results of the univariate and partial MK test statistics for all 57 time series are given in Table 4. Based on the univariate MK test, the statistical analyses show a significant downward trend for 27 of the 57 time series (47%); in 13 of these cases (23%) the decrease is highly significant (p < 0.001). For the remaining 30 cases (53%), the trend was insignificant according to the applied test.

Table 4. Results of the univariate and partial MK test of time series for pesticide annual river loads at 7 stations in German river basins
St.aSubstanceTime seriesNrbUnivariate MK testPartial MK test
Kendall's score SSDTest statistic ZProbability (p value)SignificancecProbability (p value)Significancec
  • MK = Mann-Kendall; ND = no data on substance consumption available (multivariate MK not applicable); NS = not significant (p ≤ 0.05); SD = standard deviation; St. = Station.

  • a

    Character corresponds to Table 1 and site in Figure 1.

  • b

    Number of years with load data.

  • c

    Level of significance: *p < 0.05; **p < 0.01; ***p < 0.001.

A2,4-D1990–2006171822.300.8070.4206NS0.1357NS
 2,4-DP1996–200611−3512.50−2.7990.0051**ND 
 Atrazine1990–200617−11024.28−4.531<0.0001***ND 
 Bentazone1990–200617−8024.28−3.2950.0010***0.0017**
 Chlortoluron1991–200616−2521.56−1.1590.2463NS0.8437NS
 Diuron1991–200616−6122.19−2.7490.0060**ND 
 Isoproturon1991–200616−3422.21−1.5310.1258NS0.5352NS
 MCPA1990–200617222.890.0870.9304NS0.6235NS
 Mecoprop1990–200617−9024.10−3.7350.0002***0.0281*
 Metazachlor1990–200617−7222.89−3.1450.0017**0.3814NS
 Metolachlor1990–200617−4623.93−1.9220.0546NSND 
 Simazine1990–200617−10523.69−4.433<0.0001***ND 
 Terbuthylazine1990–200617−7523.35−3.2130.0013**0.0036**
B2,4-D1996–200611−1612.82−1.2490.2115NS0.8086NS
 2,4-DP1996–200611−4312.85−3.3480.0008***ND 
 Atrazine1990–200615−8720.21−4.305<0.0001***ND 
 Bentazone1994–2006132616.391.5860.1127NS0.0755NS
 Chlortoluron1990–200610−511.18−0.4470.6547NS0.6186NS
 Diuron1996–200611−3912.85−3.0360.0024**ND 
 Isoproturon1990–200616−5222.21−2.3410.0192*0.5078NS
 MCPA1996–200611−512.85−0.3890.6971NS0.8870NS
 Mecoprop1996–200610−3511.18−3.1300.0018**0.7246NS
 Metazachlor1996–200711−1912.85−1.4790.1391NS0.6448NS
 Simazine1990–200711−2812.70−2.2040.0275*ND 
 Terbuthylazine1990–200616922.130.4070.6842NS0.1249NS
C2,4-DP1993–200512−2014.58−1.3710.1702NSND 
 Atrazine1991–200615−7620.12−3.7780.0002***ND 
 Bentazone1993–200613−1216.36−0.7330.4633NS0.5754NS
 Chlortoluron1996–200611−2312.18−1.8880.0590NS0.9751NS
 Diuron1996–200611−3912.85−3.0360.0024**ND 
 Isoproturon1991–200616−2622.21−1.1710.2418NS0.4150NS
 MCPA1996–200610−710.41−0.6730.5012NS0.6342NS
 Mecoprop1991–200615−3320.21−1.6330.1025NS0.9630NS
 Terbuthylazine1991–200615−2520.18−1.2390.2155NS0.3522NS
DAtrazine1991–200312−5014.58−3.4290.0006***ND 
 Simazine1991–20037−176.66−2.5530.0107*ND 
 Terbuthylazine1991–2003635.320.5640.5730NS0.7663NS
E2,4-DP1996–20069−189.59−1.8770.0606NSND 
 Atrazine1996–200610−2011.02−1.8160.0694NSND 
 Bentazone1996–2006101311.181.1630.2449NS0.3732NS
 Chlortoluron1996–200611−1912.50−1.5200.1286NS0.5917NS
 Diuron1996–200611−4312.85−3.3480.0008***ND 
 Isoproturon1996–200611−2312.85−1.7910.0734NS0.5216NS
 MCPA1996–200610−1511.18−1.3420.1797NS0.0599NS
 Mecoprop1996–200610−2111.18−1.8780.0603NS0.0718NS
 Metazachlor1996–200610411.020.3630.7165NS0.2746NS
 Terbuthylazine1996–200610−79.83−0.7120.4765NS0.6280NS
FIsoproturon1991–200616−8322.19−3.7410.0002***0.0029**
 MCPA1991–200616−5422.21−2.4310.0151*0.0134*
 Methabenz-thiazuron1992–200613−1815.50−1.1610.2456NSND 
 Terbuthylazine1990–200615−8019.49−4.104<0.0001***0.0009***
GAtrazine1990–200615−8719.79−4.396<0.0001***ND 
 Diuron1990–200616−6322.13−2.8470.0044**ND 
 Isoproturon1990–200616−7022.21−3.1520.0016**0.0386*
 Mecoprop1992–20039−89.59−0.8340.4045NS0.5426NS
 Simazine1990–200615−8219.49−4.207<0.0001***ND 
 Terbuthylazine1990–200615−4217.79−2.3610.0182*0.2164NS

For the subset of 36 time series with data on yearly substance usage in German agriculture, a partial MK test with consumption as the covariance factor was carried out. The results (p value and significance) are presented in Table 4. Table 5 summarizes the frequencies of trend significance according to both MK tests. As can be seen, the number of significant downward trends declines considerably when analyzed by the partial MK test; only 7 of 36 time series (19%) remain significant. The consequences of including herbicide consumption as a covariate in the test can be explored when the results of the univariate and partial MK tests are compared for the same data series. For nearly all time series, the partial MK test leads to a loss of significance (an increase in p value) for the river load trend in contrast to the univariate MK test. Although the univariate MK test gives 11 declining time series, in 4 cases the trend is no longer significant when the covariance of the trend in herbicide consumption is extracted from the target variable river load. No modification in trend significance results when including the annual river discharge at the monitoring stations as further covariate in the partial MK test.

Table 5. Frequency of significant downward trends in herbicide river load time series in Germany, based on univariate versus partial MK test (n = 57 time series, longest period 1990–2006)
Significance level (p value)All substancesOnly substances with data on national consumption
Univariate MK testUnivariate MK testPartial MK testa
n(%)n(%)n(%)
  • MK = Mann-Kendall.

  • a

    Covariate: yearly substance consumption in German agriculture.

Significant27(47.4)11(30.6)7(19.4)
 p < 0.00113 4 1 
 p < 0.019 4 3 
 p < 0.055 3 3 
Not significant30(52.6)25(69.4)29(80.6)
Total57(100)36(100)36(100)

To evaluate a possible effect the river load calculation method may have on test results, the time-proportional annual loads were calculated according to Method No. 15 of Phillips et al. (1999) in addition to the flow-proportional load. A univariate MK test using the time-proportional load data does not change the test decision in any case (results not presented). Generally the deviation between the substance loads calculated according to both methods is very small; the mean proportion of flow-proportional to time-proportional load is 0.979 on averaging the 638 individual loads (>LOQ) with a standard deviation of 0.187. The very high consistency between flow-weighted and time-weighted annual loads indicates that the substance concentration in rivers is not correlated with the river discharge at the dates of sampling.

Considering individual substances, the river loads of only 2 active ingredients, simazine and diuron, significantly declined at all monitored stations, whereas atrazine showed a significant downward trend for 4 of the 5 time series. For isoproturon, the substance with the highest market volume in Germany of all analyzed herbicides as well as for 2,4-DP, mecoprop, and terbuthylazine, the results are heterogeneous. Stations with a significant trend and with no significant changes in annual load series are represented in nearly identical proportions. The trend of MCPA is significant at only 1 of 5 stations. For chlortoluron no significant trend was detected at any of the 4 stations, and time series for 2,4-D, metolachlor, and metabenzthiazuron were also insignificant. This information, however, is based on a smaller number of stations.

With respect to the monitoring sites, the results vary considerably as well. At Station A (Cologne, the Rhine) the annual loads decrease significantly for 8 of 13 substances whereas at Station E (Frankfurt-Praunheim, the Nidda), a significant reduction was observed for only 1 of 8 time series via the univariate MK test.

DISCUSSION AND CONCLUSIONS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

A comprehensive survey of the results may lead one to the conclusion that an essential proportion of herbicide river loads have significantly declined over a period of 10–17 years in Germany, according to the univariate MK test, so that the intended outcomes of the pesticide reduction measures seem to be achieved. However, the database used to evaluate this statement is small, and some restrictions have to be mentioned for a more detailed interpretation.

All monitoring sites of the study are located in the river Rhine basin. Other main German catchments, namely the Weser and Elbe river basins, are not captured by evaluable measurements. Therefore, the situation of herbicide emission from agricultural land is only covered by river load data for less than half of all German territory. Furthermore, each pair of monitoring stations (Stations B and C as well as F and G, as discussed below) are located in close proximity and measure nearly congruent river basin areas. If 2 load time series of a substance are observed at adjacent sites, then they are not independent from each other in a statistical sense but represent a doubling of 1 single data row. Thus the resulting overall number of significant load trends is in fact based on a smaller amount of independent samples than the number of analyzed time series suggests. Evaluation is further constrained by the monitoring at Station A (Cologne), which does not unambiguously represent herbicide river input from areas under German jurisdiction but does also reflect emissions in bordering countries (France, Luxembourg, Switzerland).

With atrazine and simazine, the database comprises 2 herbicides that have been banned in Germany since 1990. Thus any other tendency than a decline in river loads of these substances would be astonishing. Nevertheless it is remarkable that it took a period of 10 years or more at nearly all stations before the high annual atrazine and simazine loads measured in 1990 and 1991 were diminished to a level slightly above or below the limit of quantification. With 53% of total annual loads (based on the average of 1990–2006) for atrazine and 51% for simazine, respectively, the findings show a distinctive peak during the months of May–July, which corresponds to the application period of these herbicides onto corn in Germany. Thus the input of these banned substances into surface waters has to be attributed mainly to recent applications during the respective years.

If the statistical evaluation is confined to the 8 substances with consumption data (2,4-D, bentazone, chlortoluron, isoproturon, MCPA, mecoprop, metazachlor, and terbuthylazine), then 36 time series remain, which form a relatively small set of empirical data for testing the overall long-term change in herbicide river loads in Germany. Of these 36, the downward trend is significant according to the univariate MK test in only 11 cases (31%). If the consumption trend is included as covariate in a partial MK test, then the significance level of river load trends is reduced in the nearly all cases, leaving only 7 statistically significant time series. This finding emphasizes that the amount of substance use is a dominant factor determining the quantity of herbicide river load. However, the trend in consumption and in river load of an herbicide differs in numerous cases. Four substances (2,4-D, chlortoluron, isoproturon, and mecoprop) with 17 time series significantly decrease in consumption, but the results of the trend tests for consumption and for river load coincide only for 5 time series. The time course of chlortoluron particularly illustrates this problem: the application has been restricted in Germany since 2002, and in following years the river loads rapidly decreased below the detection limit at all 4 stations (Figure 2). However, in contrast to the visual interpretation of the load graphs, the univariate MK test judges the trends of the 4 river load time series to be insignificant. Obviously the statistical power of the MK test for trend detection is limited and fails for data rows with certain characteristics.

Regarding the 7 load series downwardly trended according to the partial MK test, 1 (or more) additional factor besides the decline in consumption is obviously effective enough to cause the gradual river load reduction. The annual river discharge was measured by the partial MK test as a second covariate, but it turned out to be insignificant. Therefore, several factors such as enhancing the environmental awareness of farmers, technological improvements (e.g., drift-reducing nozzles, equipment for on-field sprayer cleaning), strengthened application restrictions, or minimized pesticide input via point sources could help improve water contamination. Nevertheless the extent to which one of these “soft” factors might contribute to the river load decrease of an individual substance cannot be concluded based on the analyzed monitoring data.

The annual herbicide river load is estimated via 2 steps. First the daily loads are calculated as the product of measured substance concentration and river discharge. This is followed by discharge-weighted extrapolation to a period of 1 year. Both input quantities as well as the extrapolation step imply uncertainties. For the uncertainties associated with the procedural steps discharge measurement, sample collection, sample storage, and laboratory analysis, an error propagation method calculates −17%/ + 33% as the probable range in total error. However, the authors have not found any reference to what extent the result of the MK test depends on the inaccuracy of the underlying load values. An indirect validation of the reliability of the MK test result applied to the pesticide load series might be deduced from paired sampling sites. Two pairs of monitoring stations are located close to each other and cover nearly the same catchment area; Stations B (Bischofsheim) and C (Frankfurt-Niederrad) at the Main River, as well as Stations F (Hengsen) and G (Westhofen) at the Ruhr. Nine herbicides are measured at both Stations B and C; here the trends of 7 substances are consistently judged by the univariate MK test. The discrepancy in test results for isoproturon and mecoprop between Stations B and C can be explained by different sampling strategies (grab sampling at Station B versus continuous sampling at Station C). The analyzed substance spectrum at Stations F and G on the Ruhr only match for 2 substances, isoproturon and terbuthylazine; the univariate MK test shows identical trend significance for both.

In summary, there is some evidence that a downward trend in substance emission into surface waters for several herbicides and river basins in Germany can be stated. However, the change has to be attributed mainly to the decline in substance consumption. The calculation of a quantity such as the “annual load as percent of use” (LAPU) estimated for 39 pesticides in United States streams by Capel et al. (2001) eliminates the effects of annual changes in pesticide application amounts on the river load trend, because the yearly load is related to the substance consumption of the respective year. However, LAPU quantities cannot be calculated for German rivers, because no spatially distributed data on pesticide use are available. For this purpose, a correction of the river loads at the point of measurement by in-stream losses would be needed.

The starting question whether the various programs and activities to reduce pesticide contamination of rivers are effective cannot be conclusively judged on the basis of the data set presented here. The findings are contradictory, and the spectrum of analyzed herbicides covers only a small sector of the large variety of active ingredients registered in Germany for agricultural use. Emerging herbicide substances with growing consumption such as sulfonyl-ureas especially cannot be evaluated due to a lack in river monitoring data.

Referring to the water monitoring strategies in Germany, the authors recommend improving river sampling schemes in Germany with respect to trend analyses of pesticide river load. A minimum of semimonthly sampling, as proposed by the Hilden (2003) and judged to be an appropriate strategy for long-term studies by Robertson and Roerish (1999), facilitates the calculation of valuable substance river loads, and would allow for a pronounced evaluation of the temporal shift in intensity of pesticide emissions into surface waters in the future. Furthermore, the evaluation can be improved by introducing an spatially index on substance consumption in Germany such as the LAPU (Capel et al. 2001) into the environment monitoring system.

SUPPLEMENTAL DATA

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Supporting Table S1.

Supporting Table S2.

Supporting Table S3.

Supporting Table S4.

Supporting Table S5.

Supporting Table S6.

Supporting Table S7.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

The authors are grateful to the following individuals and institutions for data delivery and helpful assistance: Fr. Jöris-Vieten, RheinEnergie AG, Köln; Dr. Post, Fr. Hiller, Mainova AG, Frankfurt a.M.; Dr. Schöttler, Fr. Zullei-Seibert, Wasserwerke Westfalen; Dr. Seel, Fr. Gabriel, Hessisches Landesamt für Umwelt und Geologie; and Fr. Dr. Wolf, Dr. Wagner, Bayer. Landesamt für Wasserwirtschaft. Many thanks to Kleffmann & Partner GmbH, Lüdinghausen, who kindly contributed the market panel results. The study was financially supported by the Association of the Crop Protection and Fertilizer Industry in Germany (Industrieverband Agrar). The authors are also indebted to the reviewers for their careful examination of the manuscript and very helpful comments.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information
  • [BMELV] Bundesministerium Ernährung, Landwirtschaft und Verbraucherschutz. 2008. Nationaler Aktionsplan zur nachhaltigen Anwendung von Pflanzenschutzmitteln. Bundesministerium für Ernährung, Landwirtschaft und Verbraucherschutz (BMELV), Bonn. 32 p.
  • Capel PD, Larson SJ, Winterstein TA. 2001. The behavior of 39 pesticides in surface waters as a function of scale. Hydrol Proc 15: 12511269.
  • European Commission. 2009. Directive 2009/128/EC of the European Parliament and of the Council of 21 October 2009 establishing a framework for Community action to achieve the sustainable use of pesticides. Official Journal of the European Union L 309/71, 16 p.
  • [GDCh] Gesellschaft Deutscher Chemiker. 1981. Deutsche Einheitsverfahren zur Wasser-, Abwasser- und Schlamm-Untersuchung, 1st ed. Weinheim, Germany: Wiley-VCH. Looseleaf in Binder, 10 Volumes, 6728 p.
  • Gilbert RO. 1987. Statistical methods for environmental pollution monitoring. New York (NY): Van Nostrand Reinhold.
  • Gilroy EJ, Hirsch RM, Cohn TA. 1990. Mean square error of regression based constituent transport estimates. Water Resour Res 26: 20692077.
  • Gutsche V, Strassmeyer J. 2007. SYNOPS: A model to assess the environmental risk potential of pesticides. Nachrichtenbl Deutsch Pflanzenschutzdienstes 59: 197201.
  • Harmel D, Cooper RJ, Slade RM, Haney RL, Arnold JG., 2006. Cumulative uncertainty in measured streamflow and water quality data for small watersheds. Trans ASABE 49: 689701.
  • Hilden M. 2003. Ermittlung von Stoff-Frachten in Fließgewässern. Berlin, Germany: LAWA-Bericht, Kulturbuch-Verlag. 62 p.
  • Hirsch RM, Slack JR, Smith RA. 1982. Techniques of trend analysis for monthly water quality data. Water Resour Res 18: 107121.
  • Hooper RP, Aulenbach BT, Kelly VT. 2001. The National Stream Quality Accounting Network: A flux-based approach to monitoring the water quality of large rivers. Hydrol Proc 15: 10891106.
  • [IVA] Industrieverband Agrar. 2010a. H2OK—Spritzmittelreste gehören auf das Feld. [cited 2010 September 22]. Available from: http://www.h2ok.org
  • [IVA] Industrieverband Agrar. 2010b. PAMIRA—Return of Agricultural Packaging. [cited 2010 September 22]. Available from: http://www.pamira.de
  • Johnes PJ. 2007. Uncertainties in annual riverine phosphorus load estimation: Impact of load estimation methodology, sampling frequency, baseflow index and catchment population density. J Hydrol 332: 241258.
  • Kreuger J. 1998. Pesticides in stream water within an agricultural catchment in southern Sweden, 1990–1996. Sci Total Environ 216: 227251.
  • [LANUV] Landesamt Natur Umwelt Verbraucherschutz Nordrhein-Westfalen. 2011. Länderübergreifender Ringversuch—PSM in Grundwasser. Düsseldorf, Germany, 163 p.
  • [LANUV] Landesamt Natur Umwelt Verbraucherschutz Nordrhein-Westfalen. 1997. Pegelvorschrift Stammtext; 4th ed. Berlin, Germany: Kulturbuchverlag.
  • Lettenmaier DP. 1988. Multivariate nonparametric tests for trend in water quality. Water Resour Bull 24: 505512.
  • Libiseller C. 2002. MULTMK/PARTMK—A program for the computation of multivariate and partial Mann-Kendall test. [cited 2010 February 15]. Available from: www.mai.liu.se/∼cllib/welcome/PMKtest.html.
  • Libiseller C, Grimvall A. 2002. Performance of partial Mann-Kendall test for trend detection in the presence of covariates. Environmetrics 13: 7184.
  • Mann H. 1945. Nonparametric tests against trend. Econometrica 13: 245259.
  • Osteroth HJ. 2004. In: Proceedings First European Workshop on Standardized Procedure for the Inspection of Sprayers in Europe; 2004 27–29 April; Braunschweig, Germany. Braunschweig, Germany: Mitteilungen Biologische Bundesanstalt Land- Forstwirtschaft. p 6873.
  • Phillips JM, Webb BW, Walling DE, Leeks GJL. 1999. Estimating the suspended sediment loads of rivers in the LOIS study area using infrequent samples. Hydrol Proc 13: 10351050.
  • Pohlert T, Hillebrand G, Breitung V. 2011. Trends of persistent organic pollutants in the suspended matter of the River Rhine. Hydrol Proc 25: 38033817.
  • Reus J, Leendertse P, Bockstaller C, Fomsgaard I, Gutsche V, Lewis K, Nilsson C, Pussemier L, Trevisan M, van der Werf H, et al. 2002. Comparison and evaluation of eight pesticide environmental risk indicators developed in Europe and recommendations for future use. Agric Ecosyst Environm 90: 177187.
  • Robertson DM, Roerish ED. 1999. Influence of various water quality sampling strategies on load estimates for small streams. Water Resour Res 35: 37473759.
  • Roettele M. 2008. Strategies to reduce point source losses of ppp to water focus on “behaviour, technique and infrastructure”: Results and lessons learned from the TOPPS project. Aspects Appl Biol 84: 357368.
  • Secco T, Pellizzato F, Sfriso A, Pavoni B. 2005. The changing state of contamination in the Lagoon of Venice. Part 1: Organic pollutants. Chemosphere 58: 279290.
  • Sudo M, Kunimatsu T, Okubo T. 2002. Concentration and loading of pesticide residues in Lake Biwa basin (Japan). Water Res 36: 315329.
  • Topping J. 1972. Errors of observation and their treatment, 4th ed. London (UK): Chapman and Hall. 119 p.
  • Vecchia AV, Martin JD, Gilliom RJ. 2008. Modeling variability and trends in pesticide concentration in streams. J Am Water Resour Assoc 44: 13081324.
  • Vryzas Z, Vassiliou G, Alexoudis C, Papadopoulou-Mourkidou E. 2009. Spatial and temporal distribution of pesticide residues in surface waters in northeastern Greece. Water Res 43: 110.

Supporting Information

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHODS
  5. RESULTS
  6. DISCUSSION AND CONCLUSIONS
  7. SUPPLEMENTAL DATA
  8. Acknowledgements
  9. REFERENCES
  10. Supporting Information

Additional Supporting Information may be found in the online version of this article.

FilenameFormatSizeDescription
ieam_1276_sm_SupplInfo.doc240KSupplementary Information

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.