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


     

Published online 7 November 2005
Published in J Environ Qual 34:2197-2207 (2005)
DOI: 10.2134/jeq2004.0474
© 2005 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 ISI Web of Science
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 ISI Web of Science (4)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Cryer, S. A.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Cryer, S. A.
Agricola
Right arrow Articles by Cryer, S. A.
Related Collections
Right arrow Pesticides
Right arrow Best Management Practices
Right arrow Software

TECHNICAL REPORTS

Atmospheric Pollutants and Trace Gases

Predicting Soil Fumigant Air Concentrations under Regional and Diverse Agronomic Conditions

Steven A. Cryer*

Dow AgroSciences, LLC, 9330 Zionsville Road, Indianapolis, IN 46268

* Corresponding author (sacryer{at}dow.com)

Received for publication December 15, 2004.

   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
SOFEA (SOil Fumigant Exposure Assessment system; Dow AgroSciences, Indianapolis, IN) is a new stochastic numerical modeling tool for evaluating and managing human inhalation exposure potential associated with the use of soil fumigants. SOFEA calculates fumigant concentrations in air arising from volatility losses from treated fields for large agricultural regions using multiple transient source terms (treated fields), geographical information systems (GIS) information, agronomic specific variables, user-specified buffer zones, and field reentry intervals. A modified version of the USEPA Industrial Source Complex Short Term model (ISCST3) is used for air dispersion calculations. Weather information, field size, application date, application rate, application type, soil incorporation depth, pesticide degradation rates in air, tarp presence, field retreatment, and other sensitive parameters are varied stochastically using Monte Carlo techniques to mimic region and crop specific agronomic practices. Regional land cover, elevation, and population information can be used to refine source placement (treated fields), dispersion calculations, and risk assessments. This paper describes the technical algorithms of SOFEA and offers comparisons of simulation predictions for the soil fumigant 1,3-dichloropropene (1,3-D) to actual regional air monitoring measurements from Kern, California. Comparison of simulation results to daily air monitoring observations is remarkable over the entire concentration distribution (average percent deviation of 44% and model efficiency of 0.98), especially considering numerous inputs such as meteorological conditions for SOFEA were unavailable and approximated by neighboring regions. Both current and anticipated and/or forecasted fumigant scenarios can be simulated using SOFEA to provide risk managers and product stewards the necessary information to make sound regulatory decisions regarding the use of soil fumigants in agriculture.

Abbreviations: ARB, Air Resources Board of the California Environmental Protection Agency • 1,3-D, 1,3-dichloropropene • GIS, geographical information systems • ISCST3, Industrial Source Complex Short Term air dispersion model • PDF, probability density function • PUR, California pesticide use records data


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
S OIL FUMIGANTS, used to control soil born pests such as nematodes, typically have high vapor pressures and thus the propensity to volatilize from soil into the atmosphere to create an inhalation potential for resident populations in agricultural regions. Recent work has explored inhalation potential of agricultural pesticides in California (Dowling and Seiber, 2002; Lee et al., 2002). California is one of the few states that monitors air for pesticides. Results for recent monitoring years can be found at the California Department of Pesticide Regulations website (http://www.cdpr.ca.gov/docs/empm/pubs/tac/tacstdys.htm; verified 14 June 2004). California also follows pesticide applications using a unique system known as pesticide use reporting (PUR). Simple attempts have been made to correlate populations with PUR records to qualitatively infer the risk associated with pesticide use (Clary and Ritz, 2003).

What is lacking in historical assessments for pesticide risk is the physical quantification of volatility loss from soil and subsequent transport once in air (i.e., exposure). The impacts of different management practices to reduce inhalation exposure and/or risk to volatile pesticides can be explored if physical processes are sufficiently quantified. Proper characterization of fumigant exposure is an important stewardship issue since future use of a variety of different fumigants will likely increase as the soil fumigant methyl bromide (MeBr) is phased out as mandated by the Montreal Protocol of 1987 (United Nations Environment Program, 1995).

Recent efforts have been made to simulate transient air concentration surrounding a single fumigant-treated field (Sullivan et al., 2004; Cryer et al., 2003). Also, several attempts to numerically simulate the impact of airborne soil fumigants on a large scale (multiple sources) have been undertaken (Honaganahalli and Seiber, 2000; Cryer and van Wesenbeeck, 2001). This paper discusses a modeling system (SOFEA) designed specifically for agricultural uses of soil fumigants that can simulate multiple transient source terms (to address neighboring and regional drift issues), couple relevant GIS data layers for spatial and temporal use data, and provide different mechanisms for agronomic practices and crop types, all while providing an easy-to-use front end (graphical user interface) for nonskilled users. Transient-air concentrations via a numerical model can be used in exposure procedures for inhalation risk characterization of unlimited numbers of use scenarios.

SOFEA was extensively reviewed by the Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA) Scientific Advisory Panel (SAP) to the USEPA in 2004 (Scientific Advisory Panel, 2004). Several areas of concern were highlighted, namely the limitations associating with ISCST3 during calm conditions and the scaling of field observed flux profiles to estimate variability. Both of these limitations are discussed in more detail in subsequent sections.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
California has surveyed land into 9654- x 9654-m (6- x 6-mile) townships. Within each township, the area is further subdivided into 36 equal subareas 1609 x 1609 m called sections (1 x 1 mile). Chronic exposure for the soil fumigant 1,3-dichloropropene (1,3-D) is managed in part by limiting the total amount of 1,3-D mass that can be applied per year in a given township. This mass limit is known as the township allocation and is mandated by the California Department of Pesticide Regulation. When the current township allocation for 1,3-D is reached, no further applications of 1,3-D are allowed. However, not all townships use 1,3-D at the current allowable allocation level. In addition, cities, mountains, and oceans constrain where fumigant treated fields (source terms) occur and thus these variables must be accounted for within a given region.

Air Dispersion Model
The Industrial Source Complex Short Term model (ISCST3, 1995) is a Gaussian plume model useful for estimating air quality surrounding contaminant release sites that was developed by USEPA as a regulatory tool for predicting concentrations of air contaminants in diverse air sheds. A limitation for all Gaussian plume models is the singularity that arises when the wind speed is zero (i.e., calm conditions). If these conditions are encountered, either the wind speed is set to some small nonzero value or the model will ignore the particular time step. Limitations and potential impact are discussed elsewhere (Scientific Advisory Panel, 2004; Coulter and Eckhoff, 1998).

Examples of ISCST3 use include vehicle exhausts in urban areas (Hao et al., 1999), industrial sulfur dioxide emissions (Kumar et al., 1999), methyl bromide concentrations resulting from soil fumigation in rural areas (Honaganahalli and Seiber, 2000), single field fumigant source terms (Sullivan et al., 2004; Cryer et al., 2003), and 1,3-D township-wide air concentrations for multiple transient agricultural sources within a California township (Cryer and van Wesenbeeck, 2001).

Modifications to ISCST3 deal with buffer zones and reentry periods (B. Johnson, California Department of Pesticide Regulation, personal communication, 2001). The user can now specify a buffer zone around source terms (treated fields). Any receptors within the field or within the buffer zone are excluded from analysis until the user supplied reentry period (e.g., 7 d) has expired, at which point the receptors are reactivated. However, these same receptors will continue to receive contributions from other fields for which the given receptors are outside of the other field's buffer zones.

Parameter Representation
Stochastic Portrayal
Air concentrations resulting from transient agricultural source terms are dependent on meteorological conditions, application timing, and so forth. A mechanism was required that could propagate parametric uncertainty in sensitive model inputs to air concentration predictions. Monte Carlo methods provide a straightforward technique to propagate such uncertainty in independent parameters to dependent output variables (Rubinstein, 1981; Yakowitz, 1977). Variability in input is described by probability density functions (PDFs) that are randomly sampled to generate input parameter sequences. If the number of randomly generated input parameter sequences is large enough, then the entire parameter space can be statistically mapped out. Output predictions are no longer single valued, but rather a discrete distribution is generated from which exceedence probabilities and return frequencies can be calculated (e.g., 1-in-100-yr exposure potential, and so on). Stochastic variables for SOFEA include the pesticide application rate, application date, hour of day an application is initiated, depth of incorporation, tarp presence, shank or drip application, field size, weather year, and pesticide degradation coefficient in air.

Crop Selection
Soil fumigants are used on a variety of agricultural commodities. Each commodity or crop is potentially unique, with different application, agronomic, and management practices. The crops chosen can be based on current or future forecasted fumigant uses, and currently up to five different crop types can be considered. Predominant crops include tree and vine, field crops, nursery crops, strawberries, and post-plant vines. Contributions to air quality from each crop are easily extractable by keeping the crop types and parameters unique during simulation. This aids in determining appropriate best management practices (BMPs) by crop type.

Simulation Domain
The simulation domain is defined as a series of contiguous townships. The central township of interest is represented as the central township within the simulation domain. This is typically the township where the largest amount of soil fumigant is used and/or a region of increased sensitivity or susceptibility. This central township surrounded by a township on all sides is referred to as the central 3 x 3 township domain (Fig. 1) . Additional townships can be considered by SOFEA up to a domain of 23 x 23 townships.



View larger version (43K):
[in this window]
[in a new window]
  		Fig. 1. Simulation domain based on contiguous township representation.

 
Receptors
Receptors are specific (x,y,z) locations in the simulation domain where air concentrations are calculated. Receptors are uniformly spaced as dictated by the user or can be calculated based on specific buffer setbacks useful for acute exposure analysis (Fig. 2) . For simulations reported here, there are 36 equally spaced receptors per township section that yields 1296 receptors per township (11664 receptors within a nine-township simulation domain). Receptor height was 1.5 m above the soil surface to mimic the breathing height of an adult. The user can specify other receptor heights to address exposure issues for children or heights associated with different floors of a neighboring apartment complex, etc.



View larger version (60K):
[in this window]
[in a new window]
  		Fig. 2. Example of field locations (dark squares) and receptor placement for a single township for setback distances of 30.5, 61.0, 122, 244, and 457 m (100, 200, 400, 800, and 1500 ft).

 
GIS Data Layers
Much of the required data for SOFEA is georeferenced and amenable to GIS overlays and extraction. The complex terrain algorithms of ISCST3 can take advantage of elevation changes within specific regions, and land cover can be used to define locations of agricultural land for field placement. Land cover information is obtained by Landsat Thematic Mapper images (30-m resolution) that contains 21 unique land classifications (available from the National Land Cover Data [NLCD] database, http://landcover.usgs.gov/natllandcover.html; verified 14 June 2005). Elevation information is obtained from the USGS Digital Elevation Models (DEM) data at 1:24000 scale. Population information is given by census blocks and populated with data from the 2000 U.S. Census.

Meteorological Data
Meteorological data required by ISCST3 include hourly air stability class, wind speed, air temperature, wind direction, and mixing and ceiling height for the region. A single weather location is used for the simulation domain. Multiyear meteorological data for a variety of locations can be found at the USEPA Support Center for Regulatory Air Models (SCRAM; USEPA, 2001) website or through state specific organizations such as the California Irrigation Management Information System (CIMIS; California Department of Water Resources, 2005). A weather library is created and each weather year within the library has a user supplied probability of being sampled (typically a uniform distribution is assumed).

Source Placement
Transient source terms for simulation are soil fumigant treated fields where pesticide volatility can occur. Source strength and location are based on experimental observations, management practices, agricultural-capable (ag-capable) land, and historical information. Source terms can be placed external to the central 3 x 3 townships up to a domain of 23 x 23 townships (19000 mi2). The user need only specify the annual mass applied to any township within a 23 x 23 township domain. Townships external to the central 3 x 3 only contribute potential drift to the central 3 x 3 or center township.

The total number of source terms selected for a given year of simulation is a function of the field size, application rate, and the total amount of pesticide mass allowed in a given township (e.g., township allocation). No further source terms are allowed once the township allocation is met. In addition, once a tree-and-vine field location has been selected, this area is restricted from use for all additional years of simulation since perennial crops' (such as vines or orchards) productivity life span is large. A fumigant reapplication is typically not performed until after the orchard or vineyard is destroyed.

Random Placement
Sources within a township can be placed randomly or weighted to specific township locations. Randomly placed fields within a township have a uniform probability of being placed within any ag-capable land found in the township. Ag-capable land is defined as all land excluding urban areas, water bodies, barren, rock, quarries, and wetlands.

Section Weighting
Certain sections within a township (1/36 township area) historically apply larger quantities of soil fumigant than in other township sections. Receptors in such sections will register higher chronic soil fumigant air concentrations due to the spatial intensity of fumigant use. For numerical implementation of this observation, the user specifies the probability of each section receiving source terms. Section-weighting probabilities can be based on historical records or expert judgment. Fields are placed randomly within the appropriate section at frequencies governed by the section probability (but are still constrained by ag-capable land). This presents a "worst-case" scenario for each year of simulation such that field locations can be quite dense in a single section. No other pesticide is rotated throughout the simulation cycle (i.e., all fields are always treated with the same fumigant) for each consecutive year of the simulation.

Overflow of Source Terms to Surrounding Sections
It is possible the number of treated fields can exceed the usable land area in a given township section for a given simulation year. This can occur if the section has a large percentage of non-ag-capable land, a majority of the township mass goes into relatively few sections, and/or when the total amount of pesticide mass (township allocation) is large. Once a field is placed, this area is unusable for additional applications of pesticide during that year since a crop is now growing (most soil fumigant applications are made pre-plant). Thus, a "cookie cutter" scenario arises as different field sizes are randomly placed within a section (Fig. 3) . There may be the possibility that a sufficiently large field cannot be placed in a given section, although the overall remaining ag-capable land is of sufficient area. In these cases, a spill-over/overflow algorithm was developed. The algorithm would first try to place the field in the user-defined section for up to 100000 trials before any overflows to neighboring sections occur. This maximizes the number of treated fields in a user specified section. Details of the overflow and other methodologies of SOFEA are found elsewhere (Cryer, unpublished data, 2004; van Wesenbeeck and Cryer, unpublished data, 2004).



View larger version (16K):
[in this window]
[in a new window]
  		Fig. 3. Illustration of field placement within a section or township where not all land can be utilized. PDF = probability density function.

 
Township Allocation of Fumigant Mass
The amount of fumigant mass applied to each township within the simulation domain is an input parameter specified by the user. Each township is assigned an allocation based on a user-supplied fraction of a reference amount. This fraction can be any scalar ≥ 0.0. The reference allocation for the total amount (mass) of soil fumigant can be the legally mandated maximum allowable allocation or a specific value for a unique region. For example, a value of 2.0 represents a township that receives 2.0 times (2.0x) the reference amount.

Figure 4 illustrates SOFEA field placement results for both random and section weighting for a 3 x 3-township simulation domain near Ventura, CA. A raster grid of 10 x 10 per township was assumed for land cover attributes. Each small dot represents a source term for different crop types. Field sizes are determined by sampling an appropriate user specified PDF for each crop. Clustering of fields is due to township sections given a larger probability of receiving treated fields.



View larger version (61K):
[in this window]
[in a new window]
  		Fig. 4. Example random and section weighted field placement (with overflow) in a 3 x 3 township domain for a 10 x 10 township grid for Ventura, CA.

 
Source Strength
Soil Volatility Flux Patterns
Fumigant volatilization from soil can be estimated by field and laboratory measurements or numerical predictions. Historical fumigant research has focused on field and laboratory measurements (a large time and financial resource commitment). However, numerical models have been used to predict fumigant volatilization. These include the two-dimensional USDA model CHAIN_2D (Simunek and van Genuchten, 1994; Wang et al., 1998, 2000), the one-dimensional models LEACHV (Chen et al., 1995) and PRZM3 (Carsel et al., 1995; Cryer et al., 2003), and linear approximations (Woodrow et al., 2001). Volatility loss from soil predicted by these models (or field measurement) can be specified as transient source terms for air dispersion modeling.

Application Scaling Factor
Measured flux rates, specific for the conditions at the time of the study, are adjusted based on depth of incorporation and time of year in an attempt to represent the complete flux response surface. Volatilization losses for soil fumigants are sensitive to temperature and depth of soil incorporation (Cryer et al., 2003). A simple procedure to account for seasonal and incorporation depth variability was developed for specific experimental (or numerical) observations selected as references for fumigant volatility loss. The transient flux loss used in the simulations for each field is given by Eq. [1]:

[1]
where fluxi = appropriately scaled hourly flux loss for hour i based on observations of field trial (kg ha–1 h–1), R = pesticide application rate (kg ha–1), Fri = experimentally observed flux rate (reference profile, scaled by experimental application rate for hour i [h–1]), Sincorp = scaling factor for depth of incorporation (dimensionless), and Syr = scaling factor for time of year (dimensionless).

Depth of Incorporation (Sincorp)
Volatility losses from a treated field decrease as the soil incorporation depth for the fumigant is increased. Multiple field and/or flux chamber studies using the soil fumigant 1,3-D have been performed (Gan et al., 1998; Knuteson and Petty, 1995; Knuteson et al., 1998; Wang et al., 2001). However, only the experiments of Gan et al. (1998) were designed specifically to investigate the impact of incorporation depth with cumulative flux losses. Data summarized by Gan et al. are represented in Fig. 5 , along with independent numerical predictions using the USEPA model PRZM3 and the USDA model CHAIN_2D for 1,3-D volatility losses as a function of incorporation depth. PRZM3 simulations assumed a Metz sandy loam (sandy, mixed, thermic Typic Xerofluvents), while CHAIN_2D soil properties were for a Myakka sand (sandy, siliceous, hyperthermic Aeric Alaquods). Numerical results corroborate the nonlinear behavior for volatilization losses with incorporation depth. Models such as CHAIN_2D typically have less than 100% mass loss for surface applications due to the wetting of soil from the application and additional physical processes such as pesticide partitioning, degradation, and transport through soil.



View larger version (23K):
[in this window]
[in a new window]
  		Fig. 5. Functional nonlinear dependence of cumulative volatilization losses for the fumigant 1,3-D as a function of application depth.

 
Volatility losses for tarped soils are indirectly incorporated through the percent mass loss input parameter for surface applied applications. Virtually impermeable films (VIF) will have near zero losses, while a more permeable film such as polyethylene (PE) will have higher loses. If no tarp is present at the soil surface, than 100% mass loss is assumed for surface applications. If a polyethylene plastic tarp is present, then a default value of 64% of applied is assumed unless otherwise specified (Cryer et al., 2003). This value can be adjusted for different tarp materials if appropriate. Either linear or nonlinear scaling with depth can be specified in SOFEA and both algorithms are dependant on if a tarp is present at the soil surface.

Temporal Representation (Syr)
Diffusion coefficients are often strong functions of temperature and temporal scaling for California is broken down into a warm or cool season to account for the greater potential mass loss during warm seasons. The scaling of cumulative mass loss between cool (22 September–21 June) and warm (22 June–21 September) season emission rates was assigned a factor of 1.6 for the soil fumigant 1,3-D (B. Johnson, California Department of Pesticide Regulation, personal communication, 2001). Thus, Syr is used to indirectly account for gross seasonal temperature effects. A continuous function (sinusoidal) based on the day of the year can also be selected should the discrete nature of the California Department of Pesticide Regulation methodology be rejected. Here, the amplitude and frequency are driven by daily average air temperatures. Temperature phase shifts in soil can be approximated using analytical or numerical soil models.

Source Constraints
Field Size Optimization
A consequence of the Monte Carlo analysis is that different field sizes and application rates will be selected for each crop type and for each year of simulation. A method was required to keep the fumigant mass applied to any given township (user specified) as a constant under any condition. Optimization procedures are used such that the township allocation is achieved but constrained by the user-supplied percentages for each crop type found within the township being simulated. User-defined field size PDFs are initially sampled to obtain starting values for field sizes for each crop type. The number of fields are determined such that the total mass of 1,3-D for all source terms is constrained at the user-specified township allocation and the residual for the crop percent cover is minimized. This results in a Mixed Integer Linear Program (MILP) problem whose optimal solution is obtained using a modified flexible-polygon search procedure (Himmelblau, 1972). The objective function requiring minimization is {Psi}:

[2]
where {gamma} = weighting variable (100 or 1000) based on order of magnitude analysis so optimization procedure executes properly under a variety of diverse conditions (by adjusting {gamma}, one can emphasize the crop percent residual, township allocation residual, or both), Tpdf i = percent of township ag-capable land that is specifically for crop i (from PDF), Talloc = township allocation for 1,3-D (kg), Ni = integer number of fields for crop i (initially unknown), Aij = area of a field for crop i (ha), Rij = application rate for crop i (kg ha–1), i = counter for the five different crop types that can be present, and AFij = California Department of Pesticide Regulation 1,3-D application factor (a dimensionless scalar between 1.0 and 2.3).

The first and second terms in Eq. [2] represent the sum of residuals for cropping area percentages and for the township allocation, respectively. Thus, Eq. [2] is a function of two user constraints: (i) the township allocation and (ii) the percentages of user defined crop areas within the township. The numbers of fields for each crop type (Ni) are constrained as integers ≥ 0. Once the number of fields for a given crop type are known, then the PDF field sizes (Ai) are adjusted (slightly stretched or shrunk) to meet the township allocation constraint. Details are found elsewhere (Cryer, unpublished data, 2004).

Annual Field Retreatment
It is possible that a farmer can repeatedly treat a field with a soil fumigant for several consecutive growing seasons. The percent of fields retreated from year to year can be specified by the user. For the first year of rotation, fields are placed appropriately (random or section weighted as defined by the user). If a 50% field retreatment is requested, then 50% of the fields from the previous year are randomly selected and marked as fields that will be retreated the following year. Fields that are to be retreated for the new year of simulation are added with new fields such that properties of retreated fields and new fields meet optimization constraints. This process is repeated for each year of simulation. Retreated fields are not stretched or shrunk during the optimization procedure.

Forecasting—Temporal Parameter Changes
Agronomic practices can and do change over time. These practices can include such things as the percentage of retreated fields from year to year, crop rotation, field size changes, the amount and type of pesticide used, application rates, incorporation depths, and loss of ag-capable land as cities grow. In addition, the ability to explore heuristic rules that may mitigate fumigant exposure is desirable (such as a field cannot be treated 3 yr in a row, staggering of sources between township sections in alternating years, and so on). SOFEA offers this flexibility by allowing up to five different parameter change regimes for a multiyear simulation interval. Parameters can included both scalar and PDF values.

Model Output Characterization
Summarized SOFEA output includes 24-h maximum and annual average receptor concentrations. Post-processing routines were written and additional averaging periods can be specified by the user (i.e., 15 d, 60 d, and so forth) if subchronic exposure levels are desired.

Population-Based Risk Assessments
Georeferenced population data can be superimposed on the air concentration data generated by SOFEA to address population-based risk and exposure scenarios. Each receptor of the uniform grid within a township is assigned a population density (if this information is supplied as an input). An example for a Monterey County chronic exposure simulation is given by Fig. 6 where it is evident that the lowest air concentrations occur in or near urban areas where population densities are the greatest. The graphic on the left indicates urban areas (hatched polygons), the location of the southwest corner of a treated field, and the township allocation fraction of 1,3-D for a 3 x 3 township domain. The graphic on the right is a surface plot for chronic air concentrations.



View larger version (34K):
[in this window]
[in a new window]
  		Fig. 6. Air concentration results for Monterey 3 x 3 (central Township 15S04E) represented as a three-dimensional mesh plot with exaggerated z axes.

 
Model Evaluation Criteria
The percentage of difference (%Di) and the model efficiency (EF; Wu et al., 1999; Bakhsh et al., 2004) were used to quantify model prediction capability against air monitoring observations:

[3]

[4]
where n = number of observations, Pi = predicted value for observation point i, Oi = observed value for observation point i, = average of the observed data, and i = the number of observations ranging from 1 to n.

Model efficiency is a measure of the deviation between model predictions and observations relative to the scattering of the observed data.

Sensitivity Analysis
Determining input parameters that create the largest variance in model output is an important consideration for quantifying parametric sources of variability. It is known a priori that both the amount of 1,3-D mass applied to a township (township allocation) and the proximity of a treated field to a monitoring location are important variables. These two sensitive inputs were eliminated by performing a sensitivity analysis on a single treated field and allowing parameters associated with the field to be altered. Input parameter sensitivity was determined by computing the rank correlation coefficients between every input PDF and dependent variable (i.e., air concentration at a specific buffer setback). Probability density functions having a high correlation to the dependant variable indicate significant impact. Sensitivity results are represented as an approximate percent variance by squaring the rank correlation coefficients and normalizing the coefficients to 100%.

Monitoring of 1,3-D in Kern County
The aerodynamic flux and flux chamber methods have been used to measure 1,3-D volatilization from treated fields. Figure 7 represents experimental observations for 1,3-D volatility losses for a California field-scale study near Salinas performed by Dow AgroSciences using the aerodynamic flux method (Knuteson and Petty, 1995; Knuteson et al., 1998). Cumulative loss was approximately 25.0% of applied for the shank application (bare soil, 45.7-cm incorporation depth). This observation is similar to numerical predictions reported elsewhere (Cryer et al., 2003).



View larger version (13K):
[in this window]
[in a new window]
  		Fig. 7. Field observations of transient volatility losses of 1,3-D for shank injection application (application rate of 137 kg ha–1 [122 lb acre–1]).

 
The measured flux profile for 1,3-D given in Fig. 7 is used as the reference flux profile for SOFEA simulations presented here. This can be a severe limitation for regions having dramatically different soil type, agronomic practices, and meteorological conditions. However, the Salinas flux loss profile yielded the highest cumulative mass loss observed in the three different California shank field trials performed by Dow AgroSciences. Accounting for soil volatility variability is beyond the scope of this work. It is assumed that the reference flux profile yields correct order of magnitude predictions for other regions of California when updated for temporal and spatial (depth of incorporation) considerations.

The Air Resources Board (ARB) of the California Environmental Protection Agency has been monitoring air for several soil fumigants, including 1,3-D, for the past decade. Seven monitoring locations in Kern County had daily air samples taken and analyzed for 1,3-D concentrations during the high use periods in the summers of 2000 and 2001. Table 1 summarizes the sampling locations of ARB. Monitoring in 2000 was from 19 July–31 August, while in 2001, the monitoring interval was from 30 June–30 August. Details and summaries of monitoring results are found elsewhere (Air Resources Board, 2001, 2002)


View this table:
[in this window]
[in a new window]
  		Table 1. Descriptor–location and IDs for air monitoring locations in Kern County.

 
Figure 8 represents the application date (day of year) histogram for all 1,3-D applications made in Kern County in 2000–2001 as collated by the private service called California Data Management System (CDMS). The CDMS data are extracted from PUR and subject to a quality control check. Clearly, the monitoring window of ARB only captured a single mode of the tri-modal distribution. Thus, SOFEA was only executed over the ARB monitoring interval using PDFs generated from CDMS data occurring during this time frame.



View larger version (30K):
[in this window]
[in a new window]
  		Fig. 8. Histogram for 1,3-D application dates occurring in Kern County during 2000–2001 and over the monitoring interval of the California Department of Environmental Protection Air Resources Board (ARB).

 
Data by crop types in Kern County for the 2000–2001 year illustrate the different agronomic practices that occur (Fig. 9) . A total of three crops were assumed (carrots, potatoes, other [peppers, onions, roses]) for the Kern simulation over the ARB time window. The agronomic practice PDFs for input parameters (field size, application rate, date, depth, tarp presence) for each of these three crop types were determined from data summarized in the CDMS database for this county and calendar years.



View larger version (32K):
[in this window]
[in a new window]
  		Fig. 9. Pie chart showing Kern County crop percentages treated with 1,3-D for 2000–2001.

 

   RESULTS AND DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
CONCLUSIONS
 REFERENCES
 
Kern Monitoring Comparison
Township M31S29E within Kern County had the highest use of 1,3-D from 2000–2001 and also the highest reported air concentrations where two of the ARB sampling stations are located (ARV, VSD). The VSD station is used for comparison because the largest 1,3-D air concentrations were observed at VSD. Ten years of simulations were performed, and the air concentrations at each unique receptor on the same day of the year were averaged. In addition, the minimum and maximum predicted 24-h air concentrations for each specific receptor in the simulation domain were tabulated. All townships in Kern county that had reported 1,3-D usage were used by SOFEA to prescribe source terms for M31S29E and all surrounding townships.

Figure 10 represents the predicted 24-h average air concentrations for the Kern Township M31S29E. Both simulation and monitoring results are expressed as an exceedence percentile in a log-linear scale due to the order of magnitude difference between low and high values. An exceedence plot represents how often a concentration of a certain magnitude is exceeded. Only at the 100th percentile did SOFEA dramatically overpredict experimental observations.



View larger version (17K):
[in this window]
[in a new window]
  		Fig. 10. Observed and simulated 24-h 1,3-D air concentrations in Kern County (M31S29E) over the 2001 California EPA Air Resources Board (ARB) monitoring interval (30 June–30 August).

 
Correct order-of-magnitude simulation results, when compared to monitoring data, suggest the simulation tool adequately captures the appropriate physics and important process parameters to yield meaningful acute and chronic concentration predictions. A percent deviation average and model efficiency values (EF) for the average daily predicted and measured air concentrations represented in Fig. 10 were 44% and 0.98, respectively, over the several order of magnitudes that the concentrations spanned. The EF relates deviations between predictions and measurements relative to the scattering of the field observations. A perfect fit between simulation predictions and field observations would yield values for and EF of 0.0% and 1.0, respectively. If the 100th percentile point (where dramatic over prediction occurred) is excluded, then values for become 15.6%.

Figure 11 represents the percent difference (Eq. [3]) between model predictions and ARB observations at various exceedence percentiles. Of interest are the relatively small percent differences (<25%) over a large exceedence range (approximately 40–90 percentile). SOFEA overpredicts at the highest and lower percentiles, and slightly underpredicts at percentiles in between. SOFEA dramatically overestimates air concentrations at the 100th percentile (267 µg m–3 vs. 36.1 µg m–3, yielding a %Di = 640%). Overprediction is likely due to one of the approximately 50 treated fields per year being placed closer to the monitoring location than an actual treated field during the ARB monitoring period. The exact locations of treated fields were not known or quantified by ARB. Thus, SOFEA was used to place fields in ag-capable land within Kern County with total 1,3-D use per township given by PUR.



View larger version (13K):
[in this window]
[in a new window]
  		Fig. 11. The percent differences (Eq. [3]) between simulated and observed Kern Township air concentrations for different exceedence percentiles.

 
Figure 12 represents several receptors whose daily air concentration predictions follow field observations. This could be indicative of the spatiotemporal placement of fields and application timings to the receptor associated with the VSD monitoring location. Perhaps future monitoring preformed by California ARB can capture these inputs in addition to spatially specific air concentrations such that a full validation of SOFEA or other regional dispersion models can be addressed.



View larger version (16K):
[in this window]
[in a new window]
  		Fig. 12. Example of simulated 1,3-D air concentrations for two neighboring receptors in the Kern simulation that mimic field observations for the Vineland School District–Sunset School monitoring location (VSD).

 
Limitations
The use of SOFEA to determine air concentrations for Kern Township M31S29E is not an exact validation exercise since the proximity of treated fields to the ARB monitoring location was not known, and actual historical weather conditions at the monitoring site over the sampling interval were unavailable. Weather information obtained from the California Irrigation Management Information System for 1993–1997 for Merced, CA, was used as a nearby surrogate and field information followed PUR data, with source terms randomly placed in ag-capable land. Hourly meteorological records indicate calm conditions occurred 10.8% of the time over the ARB monitoring window. Nevertheless, of interest is the similar order of magnitude comparison that exists between predictions and observations when appropriate township allocations, best available data for land and crop type, representative meteorological information, and agronomic practices have been specified as input parameters.

Good empirical correlations between monitoring information and air concentrations are difficult to obtain (Li et al., 2005). Predictive tools such as SOFEA can potentially estimate temporal and spatial pesticide source contributions and resulting air concentrations under different management strategies. However, a current limitation for any parametric deterministic model is the lack of complete input data sets to perform a comprehensive model validation. The California Department of Water Resources (DWR) has land use information where field polygons have been digitized from aerial photographs. Ideally, farmers' fields can someday be assigned a unique spatial ID that is coupled to the DWR field polygon and the PUR. Alternatively, ground truthing and farmer surveys can be established to garner this information for unique regions (Cryer et al., 2001). In this way, the exact proximity of treated fields to points of interest can be determined. Also, an on-site weather station should be at or near the monitoring site to provide the necessary meteorological conditions required by air dispersion modeling. The validation exercise reported here used the most complete data set currently available. Numerical tools such as SOFEA can be further validated, modified, and refined as more comprehensive data sets become available.

Additional limitations deal with the adequacy of characterizing uncertainty in fumigant volatility losses and potentially in how calm conditions are handled by Gaussian plume models. A simple attempt to address emission variability using actual field observed volatility profiles and two known sensitive input parameters (depth of incorporation and soil temperature) were made. Additional sources of variability in emission losses can include soil type, mechanical alterations in soil bulk density by application equipment, coupling of meteorological conditions to volatility losses from soil, agronomic practices such as banding and soil molded into beds and furrows, use of agricultural films, and use of a Gaussian plume model for dispersion transport. Dispersion coefficients in ISCST3 can be adjusted for specific California scenarios if tracer experiments are conducted. However, ISCST3 default values for dispersion coefficients are apparently of the correct magnitude since discrepancies between predicted and measured air concentrations are minimal. These additional contributions to fumigant emission variability were not investigated in this work.

Sensitivity Analysis
Identical agronomic input PDFs and the 5-yr weather library used in the validation exercise were used in the sensitivity analysis (i.e., crop percentage, field size, application rate and date, depth of incorporation, 1,3-D degradation coefficient in air, indirect mass transfer coefficient for tarp, and various algorithms for scaling emission losses). The dependant variable endpoint was the 15-d directional average air concentration at 30.5 m (100 ft) from the field edge (i.e., 30.5 m is the minimal buffer setback required when using 1,3-D). A 15-d air concentration spans the off-gassing time interval observed in experimental field trials (Knuteson and Petty, 1995; Knuteson et al., 1998).

A 2000-iteration simulation for a single source term was executed, where the inputs for each iteration were obtained by Monte Carlo sampling of the Kern input parameter PDFs. All iterations simulated a 60-d time interval that spanned the ARB sampling period. In addition, the algorithms for determining temporal and depth scaling factors for emission losses (i.e., Syr and Sincorp) that include a tarp parameter were added to the sensitivity analysis. Almost all of the 1,3-D applications made in Kern were with bare soil. It should be emphasized that regionally specific choices for input parameters can alter both the ranking and sensitivity of parameters (Cryer et al., 2003). Thus, sensitive parameters are specific for Kern County and those PDF attributes used to characterize agronomic practices.

Sensitive model inputs for agronomic parameters characterizing Kern County in 2001 are summarized in Fig. 13 . The crop percentage was the most sensitive parameter, accounting for 44% of the variance contribution. Crop types for the Kern analysis include carrots, sweet potatoes, and other (peppers, onions, roses). Crop percentage was followed by several sensitive crop parameters, the most dominant being the application rate. Weather had a minor impact, having a contribution to the overall variance of approximately 1.8%. Thus, the lack of having actual Kern weather in the validation exercise is probably minor as long as the surrogate weather for Merced is similar to Kern. Analysis results indicate detailed records for agronomic parameters must be available for simulation purposes (i.e., PUR or other similar databases) since factors such as crop percentage, application rates, and dates are important parameters in terms of variance in model predictions for Kern County.



View larger version (21K):
[in this window]
[in a new window]
  		Fig. 13. Results of sensitivity analysis represented as an approximate contribution to variance in model output (15-d directional average 1,3-D air concentration at 30.5 m [100 ft] from field edge and 1.5 m from the soil surface) for the Kern validation simulation.

 
Correct order of magnitude predictions for 24-h air concentrations following 1,3-D usage in Kern county over a two-month interval illustrate the ability of SOFEA to generate similar transient concentration profiles as seen in air monitoring observations. This result gives credence for the ability of SOFEA to be used as a predictive management tool for understanding, managing, and mitigating population exposure potential to soil fumigants. Soil fumigant air concentrations and the proximity to populated areas is an important topic for regions in California having high usage of fumigant products.


   CONCLUSIONS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
REFERENCES
 
Simulations used to predict soil fumigant air concentrations should reflect actual or projected use as accurately as possible to avoid overly conservative exposure predictions that would inhibit or restrict the availability of fumigant tools for growers. A comprehensive numerical tool, SOFEA, now exists to explore the ramifications of temporal changes in fumigant use in the agricultural regions of California and throughout the United States. Advances in database compilation, GIS systems, CPU processor speed, and the use of scientific programming now allow exposure resolution and refinement to be made on a level that has been historically difficult to achieve.

Comparison of 1,3-D air monitoring measurements with simulation results illustrates correct trends and appropriate order of magnitudes are attainable. SOFEA is a powerful numerical tool that is capable of simulating regional agronomic practices and resulting air concentrations for soil fumigants. A current limitation in validating SOFEA is the lack of comprehensive data sets for specific agronomic regions where the California ARB performs air monitoring.

Having a predictive tool is especially appealing for regulatory risk managers and product stewards who often must make decisions when only small amounts of information are available. Adjustments of the amount of fumigant used in a given region, both spatially and temporally, can be simulated to calculate the exposure endpoint and provide the necessary concentration distribution for performing a population-based risk assessment. Thus, the exposure calculation system outlined in this paper can be coupled into a formalized risk assessment procedure where risk to the human population can be addressed. Proper forecasting techniques, anticipated market adjustments, and sales projections will be the focus of further research to explore the viability of alternative soil fumigants to fully replace MeBr. Understanding how various agronomic best management practices affect acute, subchronic, and chronic exposure potential is a mandatory requirement for proper stewardship for all fumigants. Future work will entail coupling a soil physics model and appropriate GIS data layers to quantify spatial and regional variability in soil emission losses for the source terms used by SOFEA.


   ACKNOWLEDGMENTS
 
The author wishes to thank the following individuals for shaping the outcome of this research. Dr. Bruce Johnson of the California Department of Pesticide Regulation was instrumental in offering suggestions regarding SOFEA functionality and for his intense review of the system. Dow AgroSciences colleagues Dr. Ian van Wesenbeeck, Dr. Pat Havens, and Bruce Houtman are appropriately acknowledged for tests of earlier versions, generation of GIS graphics, and management support. In addition, Dr. Karl Schnelle and JEQ anonymous reviewers and editors were instrumental in offering constructive advice regarding this manuscript.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 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 ISI Web of Science
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 ISI Web of Science (4)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Cryer, S. A.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Cryer, S. A.
Agricola
Right arrow Articles by Cryer, S. A.
Related Collections
Right arrow Pesticides
Right arrow Best Management Practices
Right arrow Software

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Vadose Zone Journal