JEQ Grow Your Career With ASA
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online 8 August 2008
Published in J Environ Qual 37:1719-1732 (2008)
DOI: 10.2134/jeq2006.0230
© 2008 American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Juhler, R. K.
Right arrow Articles by Rosenberg, P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Juhler, R. K.
Right arrow Articles by Rosenberg, P.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Juhler, R. K.
Right arrow Articles by Rosenberg, P.
Related Collections
Right arrow Agricultural Pesticides
Right arrow Pesticides
Right arrow Soil Models

TECHNICAL REPORTS

Ground Water Quality

Impact of Basic Soil Parameters on Pesticide Disappearance Investigated by Multivariate Partial Least Square Regression and Statistics

R. K. Juhlera,*, T. H. Henriksena,b, V. Ernstsena, F. P. Vintherc and P. Rosenberga

a Geological Survey of Denmark and Greenland, GEUS, Øster Voldgade 10, DK-1350 Copenhagen K, Denmark
b present address, University Hospital of Copenhagen, Dep. of Clinical Pharmacology, Blegdamsvej 9, 2100 Copenhagen, Denmark
c Aarhus Univ., Faculty of Agricultural Sciences, PO Box 50, DK-8830 Tjele, Denmark

* Corresponding author (rkj{at}geus.dk).

Received for publication June 16, 2006.

   ABSTRACT
TOP
 NOTES
 ABSTRACT
INTRODUCTION
 Materials and Methods
 Results
 Discussion
 Summary
 REFERENCES
 
Dissipation time is a key parameter when studying and modeling the environmental fate of pesticides. This study was conducted to characterize the variability of pesticide disappearance in soil and to identify possible controlling parameters related to intrinsic soil properties and microbiology. Multivariate data analysis was used to study spatial variability in three horizons from 24 sandy soil profiles. The time for 50% disappearance (DT50) was characterized for two herbicides, metribuzin (MBZ) and MCPA, and methyltriazine amine (MTA; transformation product of metsulfuron-methyl, tribenuron-methyl, thifensulfuron-methyl, and chlorsulfuron). Normal and log-normal distributions were compared for DT50 and soil properties and descriptive statistics were calculated. Conformity with log-transformed distributions was observed and assuming normality of the DT50 data would cause 5 to 35% overestimation. Mean DT50 were: MCPA 9.5, MBZ 168, and MTA 127. Significant effect of soil depth on DT50 was shown for MCPA and MBZ, with low values in deeper horizons. Simple linear correlation for combinations of MCPA, MTA, pH, and total organic carbon (TOC) was observed. Using partial least squares regression (PLS) 71 to 85% of the total DT50 variance was explained. A specific predictor variable could not be identified as the controlling components differed within horizons and compounds. For MCPA the overall important predictor variables were microbiology and TOC, whereas for MTA and MBZ it was inorganic variables (Al, Fe, cation exchange capacity, base saturation percent, and pH) and microbiology. The study indicates that PLS generated input data can improve pesticide fate modeling and reduce the uncertainty in dissipation estimation.

Abbreviations: CV, relative standard deviation • DT50, time for a 50% decline of the initial pesticide concentration • FDA, hydrolysis of fluorescein diacetate • k, rate constant for disappearance, first-order kinetics • MBZ, metribuzin • MCPA, methyl-4-chlorophenoxyacetic acid • MTA, methyltriazine amine • PLS, partial least squares regression • SIR, substrate-induced respiration • TP, transformation product


   INTRODUCTION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
Materials and Methods
 Results
 Discussion
 Summary
 REFERENCES
 
GROUND WATER protection is a fundamental issue in the EU Water Framework Directive (Anonymous, 2000) and the Ground Water Directive (Anonymous, 2006). A major concern is the possible contamination of the resource with xenobiotics such as pesticides. Much effort is put into characterizing potential contamination caused by leaching of active pesticide compounds and transformation products (TP). Methyl-4-chlorophenoxyacetic acid (MCPA) has been observed in 1.2% of 1121 samples of monitoring screens in the National Danish Ground Water Monitoring Program (Brüsch and Juhler, 2003), and leaching of metribuzin metabolites has been observed at a sandy test site in Denmark (Henriksen et al., 2004; Kjaer et al., 2005). The need for assessment of such pesticide contamination risk is evident and modeling of fate and leaching through the unsaturated zone is a fundamental tool.

Characterization of the persistence of pesticides and transformation products is a key issue in the risk assessment, and the time for a 50% decline of the initial pesticide concentration (DT50) is a central parameter used in numerous models describing soil hydrology and leaching such as MACRO, PELMO, SIMULAT, LEACHP, WAVE, and GLEAMS (Beulke et al., 2000; Gottesburen et al., 2000; Craven, 2000; Dubus et al., 2002). Causes of DT50 variability are intrinsic as well as extrinsic environmental factors, and quantifying the impact of variability as well as uncertainties in the primary data is a key issue in pesticide fate modeling (Dubus et al., 2003; Holden and Fierer, 2005; van der Keur and Iversen, 2006). The variability of pesticide dissipation may be caused by a complex set of interacting parameters (e.g., bioavailability of the compound, microbial diversity and adoption, degree of specificity of the degraders, and the basic soil properties). These parameters display horizontal as well as vertical variation. It is a matter of dispute how knowledge of probability distribution can be used for setting up modeling scenarios. The lack of information on probability distribution, variability, and uncertainty in input parameters such as DT50 is a critical issue in hydrological modeling and prediction of pesticide fate, in particular at a larger scale such as a river basin (Beven, 1993; Haan et al., 1998).

Ideally, site-specific data should be used for pesticide leaching risk assessments (Coquet et al., 2005). However, DT50 data is costly to derive and as an alternative realistic input data can be generated based on knowledge of ranges and controlling parameters. However, knowledge of variability and coupled data sets are hardly ever available. Typically, published DT50 values are limited to only one compound or a specific soil type. Likewise, there are only few studies presenting data that covers soils as well as subsurface samples (Charnay et al., 2005). As a consequence, there are limited possibilities to set realistic variation ranges in the conceptual model scenarios. Furthermore, in most DT50 studies the focus has been on the active herbicide compound and few data on DT50 for herbicide transformation products (TPs) are available. As the relevance of TPs in relation to environmental and ground water contamination has been recognized (Kolpin et al., 1996; Sinclair and Boxall, 2003; Anonymous, 2003; Boxall et al., 2004; Quevauviller 2005; Anonymous, 2006), the need for modeling fate of pesticides as well as TPs has become evident.

In the present study, two pesticide compounds, 2-methyl-4-chlorophenoxyacetic acid (MCPA) and metribuzin (MBZ), and a transformation product methyltriazine amine (MTA, a transformation product of metsulfuron-methyl, tribenuron-methyl, thifensulfuron-methyl, and chlorsulfuron), are studied. The compounds were selected to represent different sorption and degradation characteristics. Pesticides may be degraded by purely abiotic processes as exemplified by the hydrolysis of aldicarb, but a review of several pesticides has shown that in subsoils the degradation is primarily microbial (Fomsgaard, 1995). Regarding metribuzin the microbial degradation and the mobility in natural aquifers and soil columns has been characterized (Locke and Harper, 1991; Agertved et al., 1992; Locke et al., 1994; Buss et al., 2006). Likewise, the microbial degradation of MCPA has been described (Loos et al., 1979; Sorensen et al., 2006). The formation of MTA as a transformation product has been observed in several pesticide degradation studies (Li et al., 1999; Zanardini et al., 2002).

Classical univariate statistics have been used for generating estimates of variations useful for modeling and Monte Carlo simulation. However, there is an increasing awareness that multivariate methods can provide a deeper insight into processes and trends. Thus, multivariate procedures have been used for exploring and modeling water chemistry data (Buccianti and Pawlowsky-Glahn, 2005) and limestone geochemistry (Thomas and Aitchison, 2005) and the principles for compositional data analysis have been reviewed recently (Aitchison and Egozcue, 2005). Widely used pedotransfer functions are statistical regression equations expressing the relationship between soil properties (Bouma, 1989; Hamblin, 1991) and the use of such functions as an alternative to direct measurements is well known in soil hydrology studies (Wosten et al., 2001; van der Keur and Iversen, 2006). In the present study, a similar approach was used for studying the relationship between rate of disappearance (k) and basic soil properties and microbial characteristics.

Multivariate techniques such as partial least squares regression (PLS) can provide insight into the properties causing the variation of disappearance. A PLS analysis combines aspects from both the principal component analysis and multiple regression and does not have the constraints from the multiple regression regarding normal distribution and independence of the variables (Wold et al., 1987).

The objective of the present study was to use multivariate data analysis to characterize variability of pesticide disappearance in soil and to identify possible controlling parameters related to intrinsic soil properties. The PLS analysis of disappearance, basic soil parameters, and microbiology is improved by generating all data from the very same set of soil samples (i.e., the underlying datagrid consists of comparable data). Such consistent data sets are limited in the literature. An approach for addressing range as well as variability of compound disappearance in soil is described where pedotransfer functions are generated based on correlation between disappearance and predictors such as soil parameters. By combining these functions with knowledge on DT50 variability it is possible to generate model input data with a realistic range and variability for transport and fate modeling.


   Materials and Methods
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
Results
 Discussion
 Summary
 REFERENCES
 
Disappearance of metribuzin and MCPA and the transformation product MTA was studied. Soil samples were collected from the vadose zone at 24 field sites in Jutland, Denmark. Sites were located in four different morphological landscape units: late glacial marine foreland, outwash plain, and morainic landscape from the Saalian and the Weichselian glaciations. The experimental data provided a basis for the evaluation of variability in the DT50 parameter. Measurements of disappearance were extended by data describing fundamental soil characteristics of the A, B, and C horizons. Individual parameters were characterized using univariate statistics and possible correlations between these parameters and disappearance were investigated using multivariate data analysis.

Soil Sampling
Samples were taken from the A (5 to 45 cm, mean depth 15 cm), B (30 to 120 cm, mean depth 45 cm), and C horizons (40 to 200 cm, mean depth 105 cm), and the mean sample depth of all samples was 65 cm. Soil samples were analyzed by physical, chemical, and microbial methods. Several characteristics of the soils were measured including particle size distribution (texture), bulk density, water content, porosity, water retention, cation exchange capacity (pHCaCl2), TOC, iron extracted by acid ammonium oxalate (Feox), and aluminum extracted by acid ammonium oxalate (Alox). The microbial methods including microbial biomass and activity (substrate-induced respiration [SIR]) and hydrolysis of fluorescein diacetate (FDA) were as described by Vinther et al. (2008). In short, CEC was used to indicate the total sum of exchangeable cations that a soil can adsorb whereas base saturation (%) is the extent to which the adsorption complex of a soil is saturated with exchangeable cations other than hydrogen and aluminum. For more details on analytical procedures see Barlebo et al. (2002). The supporting parameters were arbitrarily grouped into sorption-related variables (i.e., pH, base saturation, CEC, Al, Fe, and TOC), microbiology (i.e., FDA, SIR), hydrology (i.e., porosity, bulk density, and soil water potential pF1, pF2, Pf4.2) and texture-related variables (i.e., clay and silt, and sand fractions).

Sample Spiking and Incubation
After field sampling the soil samples were homogenized, sieved (mesh size 2 mm) and stored at –20°C until the start of the experiment. Soil samples were thawed at room temperature and acclimated at 10°C for 1 wk before incubation. The DT50 experiments were made using single compound incubations with certified standards. Reference standards of 2-Methyl-4-chlorophenoxyacetic acid (MCPA, CAS RN. 94-74-6) and metribuzin (MBZ, CAS RN 21087-64-9) were purchased from Dr. Ehrenstorfer (Augsburg, Germany). Methyltriazine amine (MTA, 2-methoxy-4-methyl-6-(methylamino)-1,3,5-triazine, CAS RN 5248-39-5) was purchased from Sigma-Aldrich (St. Louis, MO). Water used for spiking was HPLC grade (Rathburn, Walkerburn, Scotland).

For the experiments a sample of thawed soil corresponding to 20 g dry weight were placed in 250 mL blue cap bottles (Schott Duran, Stafford, UK) and spiked with 1.0 mL of a 20 mg L–1 MCPA, MTA, or MBZ in an aqueous solution. Stock solutions were made at 5000 mg L–1 in acetonitrile. The working solutions used for spiking were prepared by pipetting the acetonitrile standard to a glass flask and leaving it uncapped overnight in the fume hood to evaporate the solvent. Afterward, spiked solutions were prepared by adding water to the flask. Concentrations in the spike solutions were confirmed by LC-MS analysis. Using this aqueous solution, a final pesticide concentration of 1.00 mg kg–1 was made for the spiked pesticide (soil dry weight) and soils were incubated at 10°C in darkness to exclude any possible photodegradation. For each experiment three to five samplings were collected over time. The maximum incubation time for MCPA and MTA was 90 d, and for MBZ it was 154 d. For each sampling time three true replicate experiments were made.

Extraction and Pesticide Analysis
After incubation the residues were extracted using pressurized liquid extraction (Dionex ASE 200, Sunnyvale, CA) and analyzed using LC-MS/MS according to Henriksen et al. (2002) with minor modifications. Residues were extracted with a mixture of methanol/water (9:1 for MCPA, 3:1 for MBZ, 1:1 for MTA) at elevated temperature and 1500 psi (MCPA at 120°C for horizon A soil, 100°C for horizon B and C, MBZ at 60°C for all horizons; for MTA it was 60°C for horizon A soil, 90°C for horizon B and C). Extracts were analyzed using ESI LC-MS/MS (Waters Alliance LC/Micromass Triple Quad Ultima system, Manchester, UK). The chromatographic analysis was made using an XTerra 3.5 µm RP18 100 mm x 2.1 mm id column from Waters (Milford, MA). The mobile phase was composed of methanol/water (80:20) for MCPA. For MTA and MBZ the ratio of organic to aqueous modifier was 25:75 and 50:50, respectively, and the aqueous phase was 0.2 vol% acetic acid. Detection limits were determined as the lowest level measured of the respective compounds in fortified soil samples from an A and a C horizon. The limit of detection (LOD) was in the range 2.5 to 25 µg kg–1 for the three pesticides, highest for MCPA in the A horizon. These levels correspond to 0.25 to 2.5% of the initial concentration used in the disappearance studies. All solvents used were HPLC grade from Romil (Cambridge, UK); water used for LC was HPLC grade (Rathburn, Walkerburn, Scotland).

Characterization of Disappearance
A single-compartment first-order kinetic model was used for the calculation of DT50 values in the soils investigated (Bowman, 1991; Locke et al., 1994; Soulas and Lagacherie, 2001), and linear regression on log-transformed concentration data was used for estimation of the rate constant k. In a first-order kinetic process, the concentration of the herbicide at a given time can be calculated using Eq. [1] (t = time [days], C0 the herbicide concentration at t = 0 [mg kg–1], C(t) the herbicide concentration at time t [mg/kg], k = rate constant [days–1]):

Formula 1[1]
Measuring the residual herbicide concentration at three to five sampling times the DT50 could be estimated using Eq. [2]:

Formula 2[2]

Data Analysis
Statistical analysis was made using SAS (SAS Institute Inc., 2006). The procedure univariate was used for calculating kurtosis, skewness, and for testing normality of untransformed and ln transformed data, whereas the Shapiro-Wilk W test (p = 0.05) was used for evaluation. Data with k = 0 were eliminated from this test due to the ln transform, and for statistics of DT50 an arbitrary upper limit was set to 2 yr (i.e., values larger than this limit were categorized as ‘no dissipation’). Analysis of variance (ANOVA, Kruskal–Wallis test using procedure npar1way) was used to evaluate the significance of difference in dissipation between soil horizons. Correlation analysis was made using data for pH (CaCl2), total organic matter, and DT50 for MBZ, MTA, and MCPA. For each pesticide compound results from A, B, and C horizons were analyzed individually as well as combined using simple linear regression, and F statistics were used for analysis of significance.

Partial least squares regression (PLS regression using Nipals algorithms) was used to identify predictors for disappearance of the compounds. The PLS analysis of the rate constant k (Eq. [1]) was made using MatLab (Mathworks Inc., 2005) with the PLS-toolbox software (Eigenvector Research Inc., 2005) on mean-centered and normalized data using the leave-one-out cross-validation procedure (Pantsar-Kallio et al., 2001; Blanck et al., 2003; Schmidtlein and Sassin, 2004). Initially, all variables were included in the PLS and regarded as active. In an iterative process each apparently less important variable was detected and made passive in the PLS (Martens and Martens, 2000). Outliers were identified as data objects having high influence on the model (leverage) and at the same time having high residual. These were removed from the PLS model.


   Results
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results
Discussion
 Summary
 REFERENCES
 
In the present study, 18 parameters were measured in the same soil samples used for measurement of disappearance of three pesticide related compounds. The study was made on a total of 62 soil samples (i.e., 23, 24, and 15 samples from A, B, and C horizons, respectively). In some instances, an analytical procedure failed causing missing values in the overall data set; for example, measurement of base saturation failed in two samples from the A horizons (Table 1 ), but overall a comparable dataset suitable for univariate and multivariate studies was generated.


View this table:
[in this window]
[in a new window]

  		Table 1. Univariate statistic analysis of parameters in three horizons from sandy Danish soils. Number of samples (N), missing values (N miss), mean, percent relative standard deviation (CV, calculated as 100%*standard deviation* mean–1), maximum, and minimum value are shown. Abbreviations used are hydrolysis of fluorescein diacetate (FDA), substrate induced respiration (SIR), soil water potential (pF1, pF2, Pf4.2) and cation exchange capacity (CEC).

 
Statistical Analysis of Basic Soil Parameters
The descriptive statistics analysis of soil parameters is shown in Table 1. In recent reviews, general variability of soil properties has been summarized (Mulla and McBratney, 2000; van der Keur and Iversen, 2006). The present study is based on soils sampled from four different morphological landscape units, and high variation in several supporting parameters could be expected. However, for several properties measured, the variability observed is comparable to ranges compiled from several scales (Table 2 ), e.g., the overall CV of pH for all horizons. For other properties such as texture (i.e., sand parameters), a higher CV for all three horizons was observed when comparing to general levels in the literature (van der Keur and Iversen, 2006).


View this table:
[in this window]
[in a new window]

  		Table 2. Variability in soil parameters from sandy soil deposits in Jutland, Denmark. The 24 locations analyzed represent four different morphological landscape units, and variability are compared to a general variability reported in the literature. The level of %CV in the present study is indicated (‘ > ’ larger than reference range, [ ] within reference range for river basin scale). Abbreviations used are soil water potential (pF1, Pf4.2) and percentage coefficient of variation (CV).

 
A more detailed analysis indicated that the CV for pH in the C horizons was outside the general range. Likewise, high CV values were observed for contents of organic matter (e.g., the CV = 141%) in the C horizon. This reflects the tendency toward a low content of organic matter in the B and C horizons. A tendency toward lower TOC for samples from deeper soils was also observed in other studies (Severson et al., 1991; Gaultier et al., 2006). Additionally, the present results indicate that the TOC range increases when the scale is changed to geological landscapes.

The microbial biomass (SIR) and activity (FDA) changed considerably from the A horizon to the B horizon, and an increase in %CV was observed going from topsoil to subsoil (Table 1). In the C horizons the levels of the microbial parameters were low and in most cases below the detection limit (SIR < 20 µg biomass-C g soil–1 d.w.; FDA < 5 µg nitrophenol g d.w. h–1). Consequently, statistics on SIR (B and C horizon) and FDA (C horizon) are not included in Table 1.

A prerequisite for most parametric univariate and multivariate statistical analyses is that the data are normally distributed and have equal variances. It has been suggested that geochemical data follows a log-normal distribution (Ahrens, 1954). Studies of soil data have demonstrated that a log-transform could reduce the variability of the data up to threefold (Brejda et al., 2000). In contrast, a study including several large datasets of regional geochemical data has demonstrated that many variables show neither normal nor lognormal data distribution (Reimann and Filzmoser, 2000). The distribution of the present dataset was tested using untransformed as well as log-transformed data (Table 3 ).


View this table:
[in this window]
[in a new window]

  		Table 3. Data distribution, test for normal, and log-normal probability distribution. Number of data analyzed, indication of rejection of the Shapiro-Wilk W test for normality (*, p = 0.05), p value for the test, skewness, and kurtosis are shown. Abbreviations used are hydrolysis of fluorescein diacetate (FDA), substrate induced respiration (SIR), soil water potential (pF1, pF2, Pf4.2), and total carbon (C total).

 
In the present dataset there were parameters for which the Shapiro-Wilk W test for normality was not rejected for either untransformed or log-transformed data, e.g., Alox in the A horizon (Table 3). The sample size required to distinguish between log-normal (geometric-normal) and arithmetic-normal distributions has been estimated using Monte Carlo simulations (Gingerich, 1995; Koziol, 1996). These simulations indicate that with the number of data and the variability observed for some parameters in the present study it is not possible to clearly distinguish between geometric-normal and arithmetic-normal distributions. Consequently, the comparison of the data compliance with normal vs. lognormal distribution is limited to a comparison of the number of parameters that rejects the Shapiro-Wilkov test using untransformed and log-transformed data.

A summary of the distribution analysis is shown in Table 4 . Analyzing all three horizons as a whole, the Shapiro-Wilk W test was rejected for most parameters for untransformed as well as log-transformed data (Table 3). Only six untransformed and three log-transformed tests were not rejected. In the analysis, normality of untransformed soil pH was observed, and this is in agreement with previous studies (Brejda et al., 2000). Fewer tests were rejected when grouping the data according to horizons (untransformed 5, 9, 8 and log-transformed 1, 5, 5 rejected for A, B, and C horizons, respectively). Examining the normality of a data set, the degree of symmetry (skewness) and thinness of the tail (kurtosis) of the probability distribution is also of interest. Thus, if the skewness deviates from 0 the distribution is skewed, and a kurtosis different from 3 indicates deviation from the peak form of the normal distribution. Overall, more skewed and deviating peak shapes were observed for untransformed data (Table 4).


View this table:
[in this window]
[in a new window]

  		Table 4. Summary of the test for probability distributions, arithmetic and geometric (log-normal). For all 18 supporting parameters the actual measured data and the log-transformed data were tested using the Shapiro-Wilk W test (p = 0.5). Number of parameters tested, number of tests rejected, and comparison of skewness and deviation from kurtosis = 3 normal vs. lognormal distributions are shown for each horizon.

 

View this table:
[in this window]
[in a new window]

  		Table 5. Descriptive statistics of DT50. Results of statistical analysis of log-transformed and untransformed data are shown. Mean DT50 value of the untransformed (Mno) and log-transformed (Mlog) data are compared and the relative difference between the estimated values is shown as % of the untransformed mean. For test of normality, the Shapiro-Wilk W test was used (Pr < 0.05). Abbreviations used are Methyl-4-chlorophenoxyacetic acid (MCPA), methyltriazine amine (MTA), and metribuzin (MBZ). Q1 is 25th percentile, Q3 is 75th percentile.

 
Statistical Analysis of Variability of Disappearance (DT50)
It has been stated that to improve the accuracy of hydrologic models, good estimates of the means and variances of the input parameters are of greater importance than the actual form of the distribution (Haan et al., 1998). Using hydrological models for studies of fate and transport of pesticides in soils, it is advantageous to have some insight into the distribution of the parameters describing disappearance. The alternative (i.e., generating larger datasets of pesticide disappearance at each study site) would be expensive. To address these issues the distribution of the dissipation rate and halftime was tested for normality.

For most untransformed data sets the test was rejected (i.e., data were not normally distributed). For the DT50 values, normality was rejected for all datasets except MBZ in the A and C horizon. Some examples of the distribution of untransformed and transformed DT50 data are shown in Fig. 1. In Fig. 1a and b the MCPA data from all horizons is shown. A clearly skewed distribution is observed for the untransformed data (Fig. 1a), whereas the log-transform improves the shape of the distribution even if the test for normality is still rejected. Analyzing the data in detail shows that there is a large difference in disappearance within the horizons, and this is a probable cause of the lack of normality. A similar result is observed for MTA when analyzing all horizons as a whole (Fig. 1c and d). In the log-transformed data (Fig. 1d) a possible bimodal distribution can be observed, caused by a group of slow dissipation rate observations in samples of the C horizon (Table 5). The DT50 data for MBZ exemplifies a successful log-transform. The test for normality is rejected for the DT50 data (part e) whereas the log-transformed data is not rejected (Fig. 1f).


Figure 1
View larger version (24K):
[in this window]
[in a new window]

  		Fig. 1. Distribution of dissipation time (DT50, days–1) determined in the study (details on number of data points etc. in Table 6). Data from all horizons are analyzed for each pesticide compound and the frequency percentages are shown for untransformed and log-normal transformed data, and a normal distribution is shown in each plot. The compunds are methyl-4-chlorophenoxyacetic acid (MCPA), methyltriazine amine (MTA) and metribuzin (MBZ).

 

View this table:
[in this window]
[in a new window]

  		Table 6. Test of normality of dissipation data. The halftime (DT50) and dissipation rates (k) were tested using the Shapiro-Wilk W test (p = 0.5). Similar tests were made using log-transformed data. Data were tested using the Shapiro-Wilk W test (p = 0.5), and ‘-’ indicates that the test for normality was rejected, ‘X’ that the test for normality could not be rejected. Abbreviations used are Methyl-4-chlorophenoxyacetic acid (MCPA), methyltriazine amine (MTA), and metribuzin (MBZ).

 
The many rejections of the normality test for untransformed data is noteworthy, since many data in literature are based on descriptive statistics assuming normally distributed data. The present study shows that use of log-transformed data is preferred for calculation of means and variances.

To evaluate the effect on descriptive statistics all dissipation data were analyzed using untransformed as well as log-transformed data (Table 5). From the initial analysis of distribution (Table 6), it was clear that in general the untransformed data were not normally distributed, and no further details on skewness etc. are included in Table 5. In contrast, only a few tests of normality were rejected for the log-transformed data and details on the probability distribution is included in Table 5. Using the log-transformed data, the overall mean DT50 for MCPA was 9.5 d, MBZ was 168 d, and MTA was 183 d, with standard deviations as indicated in Table 5. In general, the distributions of the DT50 values showed positive skewness less than 1, with the exception of MBZ having a negative skewness in the A and C horizons. Also, there was a general tendency toward thicker tails and lower peaks in the log-transformed distributions as indicated by the kurtosis values below 3. In general it was found that an overestimation in the range of 5 to 35% resulted from the use of untransformed data for the mean DT50 estimation.

In an analysis of literature values the DT50 was 16 d (range 1–134 d) and 42 d (range 3–285 d) for MCPA and MBZ, respectively (Linde et al., 2007). In the present study where log-transformed data was used, the DT50 was 9.5 d for MCPA and 168 d for MBZ for the upper soil layers and thus within the ranges reported elsewhere. Several other studies have been restricted to top soils. Consequently, the overall mean DT50 values obtained in the present study are rather high, mainly caused by slow disappearance in lower horizons.

Comparing the degradation in the three horizons, the fastest dissipation was observed in the A horizons for all compounds. The effect of soil depth on the mean log-transformed DT50 values was examined by analysis of variance. A significant effect of soil depth was observed for two compounds: MCPA (chi-square with two degrees of freedom = 8.18, p = 0.017) and MBZ (chi-square with two degrees of freedom = 15.9, p = 0.0004). In contrast, no significant difference was observed for MTA (chi-square with two degrees of freedom = 0.09, p = 0.995). Comparing the mean DT50 of MTA (127 d) to the actual active compounds related to MTA formation, the MTA disappearance is slower (literature data DT50 median, range): metsulfuron-methyl (20 d, 3–135 d), tribenuron-methyl (13 d, 8–21 d), and chlorsulfuron (28 d, 6–147 d).

Simple correlation analysis has been used previously to identify relationships between pesticide degradation and basic soil properties. In the present study, data for total organic matter and pH were selected for further analysis, as correlation between soil pH and DT50 of isoproturon was observed in previous studies (Walker et al., 2002; Sebai et al., 2007), and the effect of organic carbon on transport and degradation of pesticides has been shown in several studies (Ahmad et al., 2001; Haberhauer, 2002; Beulke and Brown, 2006; Ahmad et al., 2006). Also, changes in organic C with soil depth have been proposed for estimation of changes in degradation rate as a function of soil depth (Veeh et al., 1996). Analyzing the horizons individually, no significant correlation was observed, whereas significant correlations were observed for the DT50 of MCPA and MBZ when analyzing all horizons as a whole (Table 7 , only significant correlations are shown). In contrast, no significant correlations were found for MTA. Previous studies on soil organic matter composition have demonstrated large variations in the molecular nature of these compounds, and detailed sorption studies are clearly needed to extend the knowledge on pesticide/organic carbon interactions (Ahmad et al., 2006). In this context, the correlations identified in the present study should be considered indicative as also indicated by the R2 values and the root mean square error (RMSE) values. Nevertheless, for MCPA and MBZ the correlation analysis supports the previous observations of pH and soil organic carbon effect on soil dissipation of pesticides.


View this table:
[in this window]
[in a new window]

  		Table 7. Simple linear correlation analysis of MBZ and MCPA, time for a 50% decline of the initial pesticide concentration (DT50, days), pH, and total organic matter. The table shows number of samples (N), square of the multiple correlation coefficient (R2), square root of the error mean square (RMSE), the correlation coefficient, and the p value for the F statistics. Only combinations with significant p values are shown (level p = 0.1). Dissipation values DT50 < 2 yr were used for censoring the data used for correlation analysis. Abbreviations used are Methyl-4-chlorophenoxyacetic acid (MCPA), methyltriazine amine (MTA), metribuzin (MBZ), and total organic matter (TOC).

 
Multivariate PLS Analysis of Dissipation
The PLS required a full dataset for all parameters in the data grid. Some datasets had to be omitted from the analysis due to missing analytical results in one or more parameters and the analysis was limited to samples where dissipation was observed (k < 0). For this reason, differences in number of datasets occurs within the PLS models and also in relation to the univariate descriptive statistics described in previous sections. The number of data sets used in each PLS is indicated in Table 8 . Also, for each model the part of the percent k variance captured by each component and by the regression model in total is listed.


View this table:
[in this window]
[in a new window]

  		Table 8. Overview of PLS models for dissipation rates (k < 0). For each compound and horizon, the number of outlier and used data sets, and Root Mean Square Error of Cross Validation (RMSECV, leave one out) are shown. For each model the part of the percent k variance captured by each component and by the regression model in total (%) is listed. Abbreviations used are Methyl-4-chlorophenoxyacetic acid (MCPA), methyltriazine amine (MTA), and metribuzin (MBZ).

 
All models were made from five or less components, and the general variance captured was 70 to 90%. The regression vectors calculated by the PLS (Table 9 ) is a numerical representation of the link between the captured variation of the predictor variables and variation in the response. For all PLS models, the loadings of each variable are shown in Table 9. As k is describing the rate of disappearance, the values are negative. Consequently, a positive correlated variable indicates a k value approaching zero when increasing the variable, i.e., larger DT50 values (Eq. [2]) and vice versa for a negatively correlated variable. For example, for MCPA in the A horizon, TOC has a negative loading indicating that an increase in content of carbon correlates with more negative (i.e., numerically larger), k values and thereby lower DT50 values. The ability of MCPA to strongly bind to carbon compounds in the soil is a possible interpretation of the result from this PCA analysis. The existence of strong carbon-MCPA bonding has been demonstrated, and effects of sorption strength on mineralization have been documented (Jensen et al., 2004), and the authors conclude that mineralization only proceeded from the water-extractable pool of MCPA.


View this table:
[in this window]
[in a new window]

  		Table 9. The regression vectors calculated by partial least squares regression. The loading of variables on the dissipation rate (k) in horizon A, B, and C are shown for each compound and horizon. Abbreviations used are Methyl-4-chlorophenoxyacetic acid (MCPA), methyltriazine amine (MTA), metribuzin (MBZ), and total organic matter (TOC), hydrolysis of fluorescein diacetate (FDA), substrate induced respiration (SIR), soil water potential (pF1, pF2, Pf4.2), and total organic carbon (Org Ctot).

 
For the present study, compounds were selected to represent different properties based on literature values, i.e., a combination of low sorption and high degradation potential (MCPA), high sorption and low degradation (MTA), and metribuzin representing low sorption and low degradation. For this reason, it is not surprising that differences in the important variable pattern are observed in the PLS analysis of k variation, considering individual pesticides differences related to horizons are also evident. For example, for MCPA the important variables in the uppermost soil (A horizon) are made up by a different set of variables than PLS models in the deeper soils (B and C horizons). In the A horizon the important variables are related to organic carbon, microbial activity as described by FDA and SIR, and the hydrology-related variables bulk density, porosity, pF1, pF2, and pF4.2. The impact of hydrologic variables on studies made in batch experiments may seem unanticipated. Furthermore, it may be expected that these parameters could be replaced by texture variables (clay, silt, and sand). However, it was not possible to substitute the hydrologic variables with texture variables. This indicates that even if the experiments were made in batch, the hydrologic variables contained information on soil properties. It may be linked to microbiology properties of the soil but no simple correlation could be identified from the present data material.

The degradation of MCPA at different depths has been studied in agricultural soil in Spain (Crespin et al., 2001). The MCPA degradation was strongly influenced by soil moisture in the upper 40 cm of the soil. The first-order rate constant observed was 0.135 d–1 corresponding to a DT50 of 5.1 d. Microbial degradation and photodecomposition was causing the degradation. The present study also identified important microbial predictors for all three compounds investigated in the A horizon (Table 9). A number of publications have shown the importance of carbon fractions in relation to sorption and modeling of MCPA leaching (Thorstensen et al., 2001; Haberhauer et al., 2001; Haberhauer, 2002; Lindahl et al., 2005; Fredslund et al., 2008; Vinther et al., 2008), and sorption to soil matrix components is a known mechanism for disappearance of pesticides and TPs. In other regression studies on pesticide sorption, both pH and organic carbon content were correlated with pesticide sorption. In a Norwegian study, the highest Freundlich adsorption coefficient values were observed in soil with high organic content and low pH (Thorstensen et al., 2001), and the presence of humic acids has been shown to increase and accelerate the movement of MCPA in a sandy soil whereas fulvic acids caused the opposite effect (Haberhauer, 2002). Also, it has been demonstrated that the mobility of MCPA increases with the application of fertilizers, in particular on soil with low clay and low organic matter content (Horvat, 2003). In the present study, organic carbon was identified as an important sorption-related parameter in the upper layers for MCPA, and the mineralogical components Fe and Al were identified as important in several PLS results from B and C horizons. Compared to the A horizons, the content of organic C is low in the B and C horizons. This indicates that for the three compounds investigated the sorption to mineralogical components such as Fe and Al becomes a more important mechanism in relation to disappearance with increasing soil depth.

For MBZ the disappearance was relatively slow (Table 5). With mean DT50 values of 116, 203, and 411 d for the A, B, and C horizon, respectively, the estimates are beyond the duration of the experiment. Thus, these results should be considered indicative rather than exact. However, even if the disappearance rate is low, the variability of the k measurements is informative when analyzed by PLS. Microbiology is of importance (FDA, SIR), and compared to MCPA more sorption-related predictors are identified when analyzing the MBZ data (pH, Al, CEC, and base saturation).

A similar result is found for MTA, i.e., microbiology is of importance in the A horizon, and the inorganic sorption predictors are of importance rather than organic C. Similar effects of inorganic soil components on subsurface degradation of isoproturon have been documented (Alletto et al., 2006) and the authors reported that clay had a major controlling effect on IPU degradation in subsoils.


   Discussion
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results
 Discussion
Summary
 REFERENCES
 
In the literature several studies on pesticide disappearance have been published (Fomsgaard, 1995; Beulke et al., 2000). However, to improve the DT50 data for use in modeling there is a need for identifying parameters related to variation in dissipation, making the results comparable between studies, and improving reliability of the model output. Thus, in the present study, the objective was to address the underlying cause of variability rather than establishing specific DT50 values.

Overall, the need for data transformation before descriptive statistical analysis was found to be of essence, and the use of the log-transform was suitable for the present dataset. As the untransformed data gives up to 35% higher DT50 estimates (Table 5) it can be argued that it is still acceptable to use untransformed data for worst case scenarios. However, aiming for realistic values and standard deviations, descriptive statistics should be based on log-transformed calculations. Also, non-symmetric variance intervals are generated as result of the back transformation of the log-transformed DT50 data. This observation is of particular relevance when setting up ranges for disappearance in Monte Carlo scenarios and similar studies.

A major concern in relation to modeling and multivariate analysis is the degree of uncertainty and representativeness of the data generated in dissipation studies (Beulke et al., 2000; Dubus et al., 2002; Dubus et al., 2003; Massey and Lenoir, 2003; van der Keur and Iversen, 2006). Thus, sampling from deeper soil horizons and incubating in batch may cause a bias in the results, and the representativeness of the results generated by techniques such as batch experiments, column experiments, and in-field experiment is an important issue for debate. Some comparison studies have been made (Di et al., 1998; Beulke et al., 2000; Mills et al., 2001; Albrechtsen et al., 2001; Gaus and Casteele, 2004; McDonald et al., 2006) and it is clear that further studies are required to identify and characterize the sources of uncertainty in the output generated by the models (Soulas and Lagacherie, 2001; Dubus et al., 2003; Massey and Lenoir, 2003; Bending et al., 2006). However, a review of the present state of the methodology has identified laboratory dissipation studies as the preferred approach (Soulas and Lagacherie, 2001).

The use of multivariate analysis seems to be the appropriate approach for studying interactions and the effect on variability of dissipation. In the present study, a significant effect of depth was observed on the DT50 value for MCPA and MBZ. The multivariate PLS of MCPA data identified microbial variables as important for the dissipation variance in the A horizon. Hydrolysis of flourescein diacetate was also measured in the B horizon, but it was not identified as important in the PLS of MCPA in the B horizon. A similar result was obtained for MTA even if the dissipation of this compound was slow. For MTA the microbial variables FDA and SIR was only identified as important in the A horizon. Further, sorption-related variables (CEC, Fe, and Al) and pH were important variables in relation to MTA. This is in agreement with strong sorption characteristics of this compound.

Overall, the results from the PLS of the C horizons are less conclusive. It may be related to the slower dissipation at these depths or to a generally higher variability of many of the parameters at this depth (Table 1). Summarizing the PLA analysis of MCPA dissipation it is clear that sorption and microbial activity are important in upper soil layers whereas texture becomes more important in deeper layers. The important texture parameters identified in the C horizon may be related to sorption effects on dissipation or direct effects on the microbial soil community such as availability of nutrients or geometrical constraints in the deep soils affecting the soil-bacteria mechanical interactions (Rebata-Landa and Santamarina, 2006).

The source of variability in pesticide dissipation may be linked to the quantitative as well as diversity aspects of the microbial community in the soil. Several factors have been identified as affecting the abundance and diversity of microbes in the unsaturated zone (Blume et al., 2002; Holden and Fierer, 2005), but in general, knowledge about spatial variability of microbial diversity and activity in subsoil is limited. Within recent years, studies on small scale variation in soil microbiology have been published (Zhang et al., 1997; Pallud et al., 2004), and geostatistic analysis of the variation in 2,4-D mineralization ranging from micro scale to field scale has demonstrated spatial heterogeneity at all scales investigated (Gonod et al., 2006).The authors observed an increase in variability when going from field to microhabitat scale. Considering the effect of scale on variability of FDA and SIR measurements used in the present PLS analysis, a detailed study on microbial parameters demonstrated that in-field variation was of the same order of magnitude as the variation between fields with CVs ranging from 23 to 42% in the A horizons and from 98 to 114% in the B horizons (Vinther et al., 2008).

Depending on the specificity of the transformation process, the variability may be related to a general decrease in microbial number when going from upper toward lower soils, a trend observed in the present study as well as elsewhere (Fierer et al., 2003; Holden and Fierer, 2005; Rebata-Landa and Santamarina, 2006). In contrast, if the transformation is dependent on the presence of specific microbes or microbial consortia, the variability of transformation is evidently related to presence and activity of specific microbes rather than to overall microbial parameters such as FDA and SIR. For example, a study of isoproturon (IPU) degradation has demonstrated the correlation between specific IPU degrading organism and the pH of the soil samples (Walker et al., 2002). Using molecular techniques it has been possible to trace this variability all the way to the genetics of the degrading bacteria (de Lipthay et al., 2003; Shi and Bending, 2007).

The presence of pesticide compounds may also induce variability in the microbial community. Thus, promotion of bacteria capable of MCPA degradation has been demonstrated at field scale in a shallow aquifer. In a degradation study of phenoxy herbicides it has been demonstrated that the presence of phenoxy herbicides could increase the abundance of specific phenoxy acid degrading Pseudomonas bacteria, and thereby promote variability at field scale (Tuxen et al., 2002; de Lipthay et al., 2003).

Overall, the different PLS patterns observed in the A and B horizons suggests that the processes in the A and B soils may differ radically, even if it may be the very same microorganisms that are involved in degrading the pesticide compound. The degradation of both MCPA and MBZ is mediated by the microbes in the soil, and previous studies have demonstrated that the microbial biomass and pesticide biodegradation declines strongly with depth (Ekelund et al., 2001; Fierer et al., 2003; Rodriguez-Cruz et al., 2006; Rebata-Landa and Santamarina, 2006; Allison et al., 2007). For example, variability of herbicide dissipation for the pesticide isoproturon has been related to sorption to clay (Alletto et al., 2006) and extractability (Charnay et al., 2005), and several studies have demonstrated interactive effects between organic matter content, microbial degradation, and sorption processes (Bollag et al., 1992; Di et al., 1998; Madsen et al., 2000). Also, several publications provide a descriptive approach to the effect on pesticide degradation, i.e., bioavailability (Shelton and Doherty, 1997; Reid et al., 2000; Guo et al., 2000; Delle Site, 2001; Saffih-Hdadi et al., 2006). In contrast, it has recently been suggested that the decrease in pesticide biodegradation with soil depth may be caused by an increase in the lag phase of the microbial degradation process rather than a decrease in bioavailability (Bending and Rodriguez-Cruz, 2007). Irrespective of the mechanism, the present study confirms the significant effect of depth on pesticide degradation for two of the three compounds investigated (MCPA and MTA), and the PLS demonstrates that these differences are taking effect within the upper layers of the soil, the A and B horizon.

A simple correlation analysis was made for dissipation (DT50), pH, and TOC. Significant correlations were identified between the two soil parameters and dissipation of MCPA and MBZ (Table 7). From this analysis, it could be anticipated that the mechanisms causing the variability of MCPA and MBZ dissipation were identical or at least similar. In contrast, the multivariate data analysis indicates that the underlying mechanisms are different. From the PLS it was shown that variation in MCPA dissipation rate is related to TOC and microbiology, whereas for MTA and MBZ inorganic sorption components and microbiology were identified as important controlling parameters. This exemplifies the advantage of using multivariate data analysis for studies of pesticide dissipation in soil. There is still a need to address variability in further detail, but the present study demonstrates that dissipation rates can be modeled using basic soil properties and general microbial measurements. Tools for prediction of vulnerable soils may be developed by combining PLS, Monte Carlo based modeling, and measurement of simple, fundamental parameters describing texture, hydrology, exchange capacity, sorption, and microbial activity. Furthermore, there is a potential for implementing more specific measurements, such as sorption measurements based on molecular structure of the soil organic carbon fractions and characterization of microbial diversity and specificity based on methods from molecular biology. By combining the multivariate data analysis with such specific methods, the influence of technical uncertainty can be further reduced, and the modeling results improved.


   Summary
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results
 Discussion
 Summary
REFERENCES
 
The mechanisms related to spatial variability of pesticide dissipation were investigated using descriptive statistics and PLS analysis. Overall, this study demonstrates that when a moderate rate of disappearance can be observed, it is possible to identify important predictor variables for the disappearance of pesticides and a transformation product. For MCPA and MBZ a significant decrease in dissipation (DT50) was observed going from the top soil toward deeper B and C horizons, whereas a similar trend could not be observed for MTA. The mean DT50 value for MTA was 127 d and to our knowledge, no previous DT50 values for the transformation product MTA have been published.

In the upper soils, dissipation rate was related to microbial parameters in general. Simple correlation was observed between dissipation of MCPA and MTA and the soil parameters pH and total organic carbon. Based on the multivariate PLS analysis it is suggested that mechanisms causing the variability of dissipation differ among the compound/horizon combinations investigated. Variability in MCPA dissipation rate was related to TOC and microbiology, whereas for MTA and MBZ inorganic sorption components and microbiology were identified as important controlling parameters. Improved descriptive statistics were generated when using log-transformed data for basic soil properties as well as for calculation of DT50. Using untransformed data for DT50 estimation caused an overestimation in the range of 5 to 35%. Although the mechanisms causing these differences have not been identified definitively, the PLS analysis of the data clearly suggest that there are fundamental differences in the processes causing dissipation of MCPA and MTA in soil from the A versus the B horizon.

Ideally, site-specific data should be used for pesticide leaching risk assessments and the results of the present study demonstrate that basic soil parameters and simple microbiology measurements can be used for estimating such data. Compared to the widely used approach where literature DT50 values are used as input data for pesticide fate models the availability of site-specific, estimated DT50 mean values and ranges are likely to improve the precision of the modeling processes. Furthermore, the prospect of substituting expensive dissipations with simple, less costly variables brings site-specific modeling within reach of a general research budget.


   ACKNOWLEDGMENTS
 
We wish to acknowledge the skilled technical assistance of M. Andersen, P.B. Jacobsen, S. Kopalski, C.R. Lynge, and P. Stockmarr, as well the work performed by several people involved in sampling and processing the soils. This work was supported by the Danish Ministry of the Environment and the Danish Ministry of Food, Agriculture, and Fisheries.


   NOTES
TOP
 NOTES
ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results
 Discussion
 Summary
 REFERENCES
 
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher.


   REFERENCES
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 Materials and Methods
 Results
 Discussion
 Summary
 REFERENCES
 





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Juhler, R. K.
Right arrow Articles by Rosenberg, P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Juhler, R. K.
Right arrow Articles by Rosenberg, P.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Juhler, R. K.
Right arrow Articles by Rosenberg, P.
Related Collections
Right arrow Agricultural Pesticides
Right arrow Pesticides
Right arrow Soil Models

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Journal of Natural Resources
and Life Sciences Education		 Vadose Zone Journal
Soil Science Society of America Journal Journal of Plant Registrations The Plant Genome