Environ. Sci. Technol. 2007, 41, 2840-2846
Relating Atrazine Degradation Rate in Soil to Environmental Conditions: Implications for Global Fate Modeling K A T H R I N F E N N E R , * ,† VALENTIN A. LANZ,‡ MARTIN SCHERINGER,§ AND MARK E. BORSUK| Eawag, Swiss Federal Institute for Aquatic Science and Technology, CH-8600 Du ¨ bendorf, Switzerland, Institute of Biogeochemistry and Pollutant Dynamics, Swiss Federal Institute of Technology (ETH), CH-8092 Zu ¨ rich, Switzerland, Institute for Chemical and Bioengineering, Swiss Federal Institute of Technology (ETH), CH-8093 Zu ¨ rich, Switzerland, Empa, Swiss Federal Laboratories for Materials Testing and Research, CH-8600 Du ¨ bendorf, Switzerland, and Department of Biological Sciences, Dartmouth College, Hanover, New Hampshire
With the ready availability of georeferenced environmental data, regional and global chemical fate models have become increasingly spatially explicit. However, the description of how chemical fate properties such as degradation rate constants and partition coefficients depend on environmental conditions has not kept up with these developments. Consequently, model results are often subject to large uncertainty stemming from inherent variability in these properties. Here, we present an extensive meta-analysis of soil degradation of one exemplary compound, the herbicide atrazine. In the first part of the paper, we present the results of an in-depth statistical analysis of the dependence of atrazine degradation rate constants on various environmental factors. In the second part, the resulting estimation equation for atrazine degradation rate constants is implemented in CliMoChem, a model for the prediction of global chemical fate, which we supplemented with spatial information on various soil descriptors, such as temperature, sand and clay content, organic carbon content, and pH. Estimates of polar accumulation, an important indicator of global chemical fate, were then compared between this model setup and estimates obtained when the degradation rate constant is represented by a single value or as being dependent on temperature only. Results for the three rate estimation methods demonstrate that a spatially explicit description of the soil degradation process results in 4-fold higher estimates of polar accumulation, while reducing uncertainty in the prediction of this endpoint by more than 40%.
* Corresponding author phone: +41-44-823 50 85; fax: +41-44823 54 71; e-mail:
[email protected]. † Eawag and Institute of Biogeochemistry and Pollutant Dynamics, ETH. ‡ Empa. § Institute for Chemical and Bioengineering, ETH. | Dartmouth College. 2840
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 8, 2007
Introduction To assess environmental exposure to chemical contaminants, predictive fate models are routinely used in addition to monitoring data. Such models range from site-specific models of chemical fate for assessing local leaching and runoff potential (1, 2), to multimedia models for assessing potential global hazards such as overall persistence and long-range transport potential (LRTP) (3, 4). All of these models critically rely on estimates of the degradation rate constants of the chemical contaminants in the environmental compartments of concern. Such rate constants exhibit small- and largescale variability (e.g., (5)), which might cover several orders of magnitude (6). Different experimental approaches for determining rate constants and uncertainty in analytical techniques further add to the difficulty in selecting appropriate values for exposure modeling. In some modeling studies, single values from reference books are used to represent rate constants, while in others average values from a collection of data are employed. More sophisticated approaches use probabilistic techniques such as Monte Carlo analysis to propagate a distribution of values to the model endpoints of interest (7-9). While this might yield a more realistic estimate of uncertainty in the environmental behavior of a single compound, it often leads to predicted uncertainty ranges that are larger than the predicted differences in behavior between compounds (10). This hampers prioritization of chemicals according to their exposure potential and hence reduces the usefulness of exposure indicators such as persistence or LRTP for risk management purposes. Through the advent of geographical information systems and more advanced remote sensing techniques, the amount of information on spatially and temporally varying environmental conditions that can be incorporated into environmental fate models is expanding rapidly. As a consequence, spatially resolved models including georeferenced information on various environmental factors have been developed, claiming a more accurate representation of chemical fate (11, 12). However, as we demonstrated in an earlier study (10), the outcomes of such models will still be highly uncertain unless spatial variability in chemical fate properties is accounted for by explicitly representing their dependence on environmental factors. In the present study, we corroborate this point by demonstrating how knowledge on the dependence of degradation rate constants on environmental conditions can be used, e.g., in a global fate model, to reduce uncertainty in model results. This is exemplified for the case of soil degradation of the globally used herbicide atrazine, a compound for which the number of soil degradation studies is especially high. We first present a statistical approach for developing a multivariate model for rate constant prediction based on experimental data from the literature. Second, we apply the statistical model to estimate global patterns of atrazine soil degradation rate constants accounting for the spatial variability of soil properties. These patterns are then implemented in CliMoChem (13), a zonally averaged, global multimedia chemical fate model with spatial resolution of environmental conditions along the South-North axis. The purpose of this study is to show how the inclusion of spatially explicit rate constants both shifts the mean prediction and narrows the uncertainty interval for polar accumulation of atrazine, an exemplary endpoint of global chemical fate. 10.1021/es061923i CCC: $37.00
2007 American Chemical Society Published on Web 03/21/2007
Environmental Factors Influencing the Atrazine Soil Degradation Rate Constant Literature Review and Collection of Degradation Rate Constants. In an extensive literature search, 469 degradation rate constants for atrazine in soil were collected from 38 published and 17 unpublished data sources. For the statistical analysis of these data, we gathered comprehensive information on experimental conditions of environmental factors (e.g., soil pH, moisture content, sample depth, organic carbon (OC) content) and methodological details (e.g., study duration, sampling year), which amounted to 27 possible predictor variables for each data point (Table S1 in the Supporting Information, SI). In parallel, literature was reviewed to formulate a set of hypotheses regarding the expected relationships between the various environmental factors and atrazine degradation rates in soil. These hypotheses are used as a starting point for the multivariate model used in the statistical analysis. Collected rate constants refer to the total mass of atrazine in bulk soil. They include rate constants for mineralization and primary dissipation (i.e., the disappearance of atrazine, including degradation and transfer out of the studied system), for surface and subsurface soils, as well as those measured in both laboratory experiments and field studies. The distribution of all rate constants was approximately lognormal, with a median half-life of 39.3 d and a 95th-percentile range of 1.2 to 1300 d. Citra (7) analyzed 275 degradation rate constants of atrazine and calculated an average half-life of 68.2 d and a standard deviation of 77.3 d. Given the lognormal distribution of his dataset, this translates into a median half-life of 45.1 d, which is similar to our value. Our literature review indicated that, in soil, atrazine is degraded most efficiently by microbially mediated processes, whereas abiotic processes such as hydrolysis are less important (14, 15). The degradation rate is therefore mainly regulated by the bioavailability of the compound, the viability of the soil bacteria, and the presence of the specific enzymes involved. However, due to difficulties in quantitatively measuring the latter two attributes, most studies only report indirect factors influencing the activity of microbes, such as soil humidity, carbon and nitrogen content, and temperature. Based on our literature review, we hypothesize the following qualitative dependences of the degradation rate of atrazine in soil: (i) low soil nitrogen content can trigger microorganisms to use atrazine as a nitrogen source, thus enhancing atrazine mineralization (16); (ii) increasing clay and OC contents of soils can result in decreasing atrazine degradation rates, most likely due to a higher extent of sorption and hence lower bioavailability; (iii) at very low OC content, there can be a negative effect on the degradation rate due to nutritional limitations (17) and reduced microbial activity due to a lack of sites for attachment and growth (18); (iv) extreme pH values can inhibit microbial activity and thus reduce atrazine degradation (19); (v) there is probably an optimal region where moisture is sufficiently high so that atrazine is highly available and microbes are active (18), but below saturation, which could lead to anoxic conditions (20); (vi) under most natural conditions, temperature can be expected to have a strong positive influence, with reported activation energies between 30 and 70 kJ/mol (21, 22); (vii) increasing soil depth has been associated with decreasing degradation rates (20), most likely due to decreasing temperature, availability of energy sources and oxygen, and hence decreasing microbial activity with soil depth; (viii) the method, rate, and frequency of atrazine application are also suspected to influence degradation rates, but so far no clear relationships have been identified in the literature. Data Processing and Exploratory Data Analysis. Because not all 27 predictor variables were reported in all studies, our
data set had a total of 58% missing values for the predictor variables. Since such a data set is not well conditioned for statistical analysis, we employed established relationships from soil physics and chemistry to fill in missing values and consolidate the number of possible predictors (for details see SI-1). Exploiting these relationships reduced the proportion of missing values in our data set to 27% for 15 consolidated predictor variables (Table S1 in SI). It has been reported in several instances that dissipation rates observed in the field are higher than those measured in laboratory batch experiments. This has been attributed mainly to soil preparation and storage for laboratory experiments, which tend to reduce the number of viable microorganisms (23), as well as to the presence of additional loss processes such as leaching or volatilization in the field. It might also be expected that primary biodegradation, which for atrazine mostly involves dealkylation of the amine side chains, would proceed faster than mineralization, which involves breaking the triazine ring structure. We therefore carried out an analysis of variance (ANOVA) to test whether the two methodological variables study.type (field or laboratory) and endpoint (mineralization or primary biodegradation) split the data set into significantly different groups. We found that the resulting groups are not significantly different from each other (p ) 0.44, n ) 469). Based on these findings, degradation rates were not distinguished according to the methodological variables study.type and endpoint in the further analysis. Multivariate Regression. We used the set of hypotheses described above to develop a generic model for atrazine soil degradation rate constants bm k ) k0‚ f(T)‚g(pH)‚h(corg)‚sandbs‚nbtotn ‚waterbw‚dmin ‚ d dbdiff ‚atr.condba (1)
In this equation, k is the measured atrazine degradation rate constant (d-1); k0 is a constant; f(T), g(pH), and h(corg) are initially unspecified functions of temperature T (K), pH, and corg (mass %), respectively; sand is the gravimetric sand content (mass %); ntot is the total nitrogen content (mass %); water is the volumetric water content (vol %); dmin is the minimum depth of the sample (cm); ddiff is the difference between the maximum and minimum depth of the sample (cm); atr.cond is a binary variable indicating the atrazine degradation potential of the soil. It takes a value of 1 if known atrazine degrading bacteria were present in the soil sample or if atrazine is known to have been applied at the sampling location previously and a value of 0 otherwise. The exponents in eq 1 are unknown constants to be estimated from the data. Equation 1 allows for non-monotonic dependence of k on those predictors for which an intermediate optimal value has been suggested (T, pH, and corg), and assumes monotonic relationships with the other predictors. Although an optimum region has also been hypothesized for water content, it is expected to lie above the range of our data (the majority of samples exhibit
