Journal of Exposure Science and Environmental Epidemiology (2009) 19, 187–200; doi:10.1038/jes.2008.12; published online 12 March 2008

Assessing exposure to allied ground troops in the Vietnam War: A comparison of AgDRIFT and Exposure Opportunity Index models

Michael E Ginevana, John H Rossb and Deborah K Watkinsa

  1. aM.E. Ginevan & Associates, 307 Hamilton Avenue, Silver Spring, Maryland 20901, USA
  2. Inc., 5233 Marimoore Way, Carmichael, California 95608, USA

Correspondence: Dr Michael E. Ginevan, M.E. Ginevan & Associates, 307 Hamilton Avenue, Silver Spring, MD 20901, USA. Tel.: +301 585 4951; Fax: +301 585 1350; E-mail:

Received 10 August 2007; Accepted 7 December 2007; Published online 12 March 2008.



The AgDRIFT aerial dispersion model is well validated and closely related to the AGDISP model developed by the USDA Forest Service to determine on- and off-target deposition and penetration of aerially applied pesticide through foliage of trees. The Exposure Opportunity Index (EOI) model was developed to estimate relative exposure of ground troops in Vietnam to aerially applied herbicides. We compared the output of the two models to determine whether their predictions were in substantial agreement, but found a total lack of concordance. While the AgDRIFT model estimated that ground-level deposition through foliage was reduced more than 20 orders of magnitude at less than 1km from the flight line, the EOI model predicted deposition declines less than one order of magnitude 4km from the flight line. Interestingly the EOI model predicts a four-fold variability in EOI on the flight line, where exposure should be essentially invariant because the spray apparatus is designed to apply herbicide at a constant rate. We believe that the EOI model cannot be used to provide individual exposure estimates for the purpose of conducting epidemiologic studies. Moreover, evaluation of the position data for both herbicide spray swaths and troop locations, together with the actual patterns of spray deposition predicted by the AgDRIFT model, suggests that precise individual-level exposure assessments for ground troops in Vietnam are impossible. However, we suggest that well-validated tools like AgDRIFT can be used to estimate exposure to groups of individuals.


AgDRIFT, Agent Orange, exposure assessment, Exposure Opportunity Index, herbicide, Vietnam War



The aerial application of herbicides by the military during the Vietnam War began in 1962 and reached a peak during the period 1967–1969. Agent Orange (AO), a 50:50 mixture by weight of the n-butyl esters of two phenoxy acids, 2,4-dichlorophenoxyacetic acid (2,4-D) and 2,4,5-trichlorophenoxyacetic acid (2,4,5-T), was the most extensively used herbicide during the war. Military use of AO ended in April 1970 following a scientific report that concluded that one of the primary chemicals in the herbicide could cause birth defects in laboratory animals (Gough, 1986, 2003; IOM, 1994; Butler, 2005). The health concerns stemmed from the fact that 2,4,5-T was found to be contaminated with 2,3,7,8- tetrachlorodibenzo-p-dioxin (TCDD), which had been found to cause some health effects, such as chloracne in humans, and which had also been shown to cause a number of adverse health effects in animal studies.

By the mid-1970s, a growing number of Vietnam veterans began to question the possible link between their conditions and/or diseases—ranging from cancer to diabetes to birth defects in their offspring—and exposure to herbicides, mainly AO, in Vietnam (Gough, 1986). By the end of the decade, Vietnam veterans, skeptical of their treatment from the Department of Veterans’ Affairs and other government agencies took their concerns to Congress (IOM, 1994; Gough, 2003)

Congressional response to the veterans resulted in numerous public hearings and several pieces of legislation. In 1983, P.L. 98–181 appropriated $57.4 million to the Centers for Disease Control (CDC) to conduct research on the health risks for Vietnam veterans exposed to AO. The Veterans’ Dioxin and Radiation Exposure Compensation Standards Act of 1984 (P.L. 98–542) was passed to address the issue of compensation for disabilities that might have resulted from exposure to AO in Vietnam. The Agent Orange Act of 1991 (P.L. 102–104) directed the Secretary of Veterans Affairs to request the National Academy of Sciences to conduct a “comprehensive review and evaluation of available scientific and medical information regarding the health effects of exposure to Agent Orange, other herbicides used in Vietnam, and their components, including dioxin” (IOM, 1994).

The first of these comprehensive reviews was published in 1994. In that review, the Committee made several recommendations (as mandated under the Agent Orange Act) concerning the need for additional scientific studies. Noting that exposure assessment had been a “weak” aspect of most epidemiologic studies of Vietnam veterans, the committee recommended the development of exposure reconstruction models by an “independent, nongovernmental scientific panel” and if a “valid” exposure reconstruction model is feasible, the “Department of Veterans Affairs and other government agencies should facilitate additional epidemiologic studies of veterans” (IOM, 1994).

In 1996, at the request of the Department of Veterans Affairs, the Institute of Medicine (IOM) formed the Committee on the Assessment of Wartime Exposure to Herbicides in Vietnam. This committee was charged with overseeing the development and evaluation of models of herbicide exposure for use in studies of Vietnam veterans. The Committee noted that the intent of the research was to (1) develop and document a detailed methodology for retrospectively characterizing the exposure of Vietnam veterans to the major herbicides used by the military in Vietnam and (2) demonstrate the feasibility and appropriateness of the proposed methodology in sufficient detail to permit the assessment of its potential for use in the conduct of epidemiologic studies (IOM, 1997).

In 2003, investigators from Columbia University published the first of several descriptions of an “Exposure Opportunity Index (EOI) Model” designed to characterize exposure to AO and other herbicides in Vietnam (Stellman et al., 2003; Stellman and Stellman, 2004, 2005). This EOI model relies on proximity to spray missions to estimate “exposure opportunity” but does not estimate actual exposure. To date, no one has published any quantitative estimates of the relationship between EOI, as estimated by this model, and actual exposure.

Independent of the issues related to herbicide exposure in Vietnam, another relevant model-building activity began in the 1960s. Models, such as those described by Yates et al. (1966), focused on drift of airborne pesticides beyond the intended area during aerial applications. This drift is a source of concern due to potential human health impacts, damage to nontarget crops and livestock, and ecological impacts. This model development effort has culminated in a model for spray deposition from aerial application called AgDRIFT®. The general model is discussed in Teske et al. (2002), the kinds of data used in developing the model are discussed in Hewitt et al. (2002), and the model's algorithms are discussed and evaluated in Bird et al. (2002) and presented in detail in Teske et al. (1993). The important difference between the EOI model and the AgDRIFT model is that AgDRIFT is explicitly concerned with quantitative dispersion and deposition estimates, whereas the EOI model is a more qualitative measure of potential exposure, although it has been presented as a quantitative model useful for epidemiology. This paper compares and contrasts the performance of AgDRIFT used in conjunction with accepted pesticide exposure assessment methodology (Franklin and Worgan, 2005) and the performance of the EOI model for predicting exposure potential of US ground troops to herbicides in Vietnam.

Aerial Application Module of AgDRIFT

AgDRIFT has three basic levels of operation: Tier I, II (agricultural use only) and Tier III (agricultural and forestry module). Each successive level requires fewer default assumptions and more input from the user. The ultimate goal of the input variables is to achieve an estimate of particle size distribution from the point of emission (nozzles) and the subsequent deposition of these particles. Airspeed, nozzle diameter, surface tension and viscosity are the minimum variables necessary to estimate spray particle size distribution, as these factors have been used in wind tunnels to empirically determine distribution as a function of these variables. The mathematics of the dispersion model is described in detail by Teske et al. (1993). Lagrangian mathematics describes the “near-field” particle movement to approximately 800m, while Gaussian distributions can be used for particles further from the aircraft (in the AGDISP model). AgDRIFT simulates individual particle movement as a function of time and space, and predicts deposition mass of particulates as a function of distance. It allows both ground and foliar deposition estimates.

First, AgDRIFT is a mechanistic model (Teske et al., 2002), because it uses detailed mathematical formulae to describe dispersion of spray droplets near an aircraft as influenced by type of propulsion and aerodynamic surfaces. Other factors incorporated by AgDRIFT to estimate spatial concentration distribution of deposited spray particles include aircraft speed, type and orientation of spray nozzles, spray pressure, droplet size distribution from specific nozzles and influence of spray mixture physicochemical properties, and droplet evaporation. AgDRIFT can also model interception of spray droplets by forest canopy. AgDRIFT can accommodate at least 30 different variables to estimate spray deposition (Teske et al., 2002). Key variables include spray boom length-to-wingspan ratio, air speed, altitude of spray over canopy, canopy density, crosswind speed and volatility of the formulation being sprayed. The current AgDRIFT model is used by US EPA, the Pest Management Regulatory Agency in Canada, and the Australian National Occupational Health and Safety Commission. Although AgDRIFT is a proprietary model that is compensable for pesticide registration, it is publicly available for other uses at Moreover, virtually identical results can be obtained for aerial application from its predecessor, AGDISP, that is in the public domain, because it has the same computational engine for estimating aerial dispersion (Bird et al., 2002).

Second, a large body of data has been used to develop and validate AgDRIFT (Hewitt et al., 2002). Overall, AgDRIFT was supported by empirical measurements of pesticide deposition from approximately 180 aerial applications, as well as the dozens of applications used to create its predecessor model, AGDISP (Teske et al., 1998).

Finally, pesticide deposition predictions from AgDRIFT agree well with empirical data (Bird et al., 2002), although it tends to slightly underestimate deposition at near release distances. Pertinently, the aerial application module of AgDRIFT has been validated by empirical measurements independent of any data used to develop the model, as would be expected by a mechanistic model that predicts the entire spray deposition process. Nevertheless, model results tend to overestimate far-field dispersion compared with actual measurements (Bird et al., 2002), an observation of particular relevance to subsequent discussions. AgDRIFT is particularly suitable for estimating exposures from aerial application of herbicides in Vietnam. Herbicide application missions were frequently made from Fairchild Industries Provider UC-123K (or C123) fixed-wing aircraft. This model aircraft is specifically included in the aircraft library of AgDRIFT, and, therefore, application parameters as affected by aircraft characteristics are easily modeled.

AgDRIFT predicts the effective spray swath of a C123 as 50 to 100m. Spray swath can be defined operationally as the area of biologically effective (herbicidal) deposition. Modeling results also show that there is little spray deposition outside the swath (less than 2% of the ground deposition through foliage occurs outside the spray swath). These modeling results are similar to the empirical results from Air Force studies reported by Harrigan (1970), where swath width using the C123 plane and spraying system produced a swath width on bareground of 157ft in a crosswind. Thus, the fact that AgDRIFT, using the physical characteristics of the aircraft-spray apparatus system, predicts the observed results obtained in the earlier Air Force studies provides independent validation of the model results.

Pesticide Exposure Assessment Practice

Although AgDRIFT provides estimates of the levels and spatial dispersion of pesticide residue, turning these estimates into actual exposure estimates requires exposure models. There are two types of exposure to be modeled: exposure to pesticides from direct deposition and exposure to pesticides following contact with the treated environment (post-application exposure). A very large body of empirical exposure monitoring data exists relating deposition estimates to both exposure from direct aerial application and exposure due to post-application contact with pesticides. Examples of direct application exposure include ground personnel directing aerial application (Ramsey et al., 1979; Lavy et al., 1980; Newton and Norris, 1981), persons attending a ball game (Dong et al., 1994) or running under an aerial application of malathion (Ross et al., 1991), and the aerial flagger category in the Pesticide Handlers Exposure Database (EPA, 1998).

These extensive data allow one to characterize factors that modify exposure potential resulting from aerial application. These factors include penetration of spray particles through a leafy canopy, clothing penetration, “body shading” and dermal absorption of the material reaching the skin. Each of these factors has been discussed in detailed publications, but it is worthwhile summarizing them as they pertain to AO application. Clothing penetration is a physical factor that determines how much of a pesticide gets from outside a shirt, for instance, to the inside. The arithmetic mean value for single-layer clothing penetration is 10% for relatively nonvolatile chemicals (Driver et al., 2007) with a range of 1–50% inversely proportional to the deposition density. Body shading is the phenomenon, where the head and shoulders projected over the lower portion of the body protect it from deposition of aerial residues coming from overhead. This phenomenon essentially results in a nonuniform deposition over the body producing approximately two-fold reduction from an assumed uniform whole-body deposition (Williams et al., 2001). Finally, dermal absorption represents the effectiveness of the skin in reducing systemic uptake from dermally applied chemical. Dermal absorption for 2,4-D and 2,4,5-T has been measured in humans or nonhuman primates at 5.7% and 0.6%, respectively (Newton and Norris, 1981; Ross et al., 2005). TCDD dermal penetration has only been measured in vivo in rats, but rats may overestimate human dermal absorption by an average of five-fold (Ross et al., 2001). Application of this average human/rat factor to TCDD dermal absorption for adult rats (22–40%; Anderson et al., 1993) yields estimates of absorption by humans similar to those observed for the phenoxy herbicides.

Of all the factors influencing exposure on the ground following aerial application, the most important is penetration of spray particles through the forest canopy. The emission droplet size spectrum is extremely important in affecting this penetration or deposition. Droplet size was determined empirically using AO sprayed from C123s onto Kromkote cards, and a spread factor was used to estimate spherical size of droplets at the point of impact (Klein and Harrigan, 1969; Harrigan, 1970). Although this is not the most accurate method, it was the method in common use at the time. Current methods would employ laser sizing of spray droplets produced in a wind tunnel. Three factors significantly reduce deposition of aerial spray at ground level when trees are sprayed. One is “canopy roughness” meaning the differences in height of the foliage moving laterally from the spray zone. Another is canopy density and is frequently expressed as leaf area index or LAI (the area of leaves on trees divided by the area covered by the trees). A third factor is the non-perpendicular movement of particles through leaves increasing path length to the ground. None of these are accounted for in the EOI model, yet these factors account for several orders of magnitude reduction in potential exposure. More specifically, these factors result in ground deposition lower than one might predict based strictly on the LAI. For instance, in a forest where the nominal LAI is 4.1, one might expect particles sprayed directly over the canopy to be intercepted by foliage and the ground deposition fraction would be 2−4.1 or approximately 6% of a direct spray. In fact, the AgDRIFT model predicts, and measurements confirm, that at ground level within the spray swath the ground deposition is 1–2% under such a canopy. Although estimates on the order of 2–6% canopy penetration can be found in the literature, these are based largely on penetration of light. A more appropriate estimate is about 1% based on AgDRIFT modeling because it also takes into account canopy roughness and non-perpendicular foliar penetration. A study on canopy penetration measurements was conducted in a rain forest in Puerto Rico and savanna forest in Texas using aerially applied herbicide (Tschirley, 1968). Results from that study indicated that spray penetration through forest canopies was inversely related to canopy density and linear through the range of 0–100% canopy density. Interestingly, the lowest penetration measured (~3%) occurred in 5-foot-tall wild roses.

Post-application exposure to pesticides has been investigated quantitatively for over 50 years (Ross et al., 2006). Algorithms based on whole-body human exposure monitoring have been developed that relate exposure rate to dislodgeable foliar residues of pesticides on leaf surfaces. US EPA has summarized a number of the post-application worker monitoring studies in a policy document (EPA, 2000), and regulatory agencies worldwide recognize the method (Whitmyre et al., 2005). Transfer coefficients are empirical measures of the rate of contact with foliage as a function of work task and can be used to estimate dermal exposure as shown in Eq. (1) below (EPA, 1997).

Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact or the author

where, DE=dermal exposure in units of μg pesticide on skin per kg body weight

TC=transfer coefficient in units of cm2/h

DFR=dislodgeable foliar residue in units of μg/cm2

t=time in h/day that contact with foliage occurs

BW=body weight in kg.

Transfer coefficients range from a low of 100cm2/h for “scouting” crops to a high of 17,000cm2/h for work tasks such as detasseling sweet corn, where the person's body is totally enveloped by foliage (EPA, 2000). A key variable in this calculation of post-application exposure is the DFR. This is defined as “That part of a residue of a chemical deposited on a surface which may be transferred by direct contact to human skin or clothing” (Franklin and Worgan, 2005, page 394). That is, DFR is the fraction of the deposition that can be removed by dermal contact. In the absence of data, it is common to conservatively assume that 20% of the amount deposited on foliage is available for transfer in the form of DFR (EPA, 1997). Since DFR decays as a function of time due to factors such as absorption, volatilization, photolysis and rain/dew events, the time after application that foliar contact occurs is a significant determinant of post-application exposure. We note that DFR typically declines more rapidly than total residues. This is important because dioxin sprayed on foliage is often bound to substrates like leaf waxes, and, thus, while it may persist, its bioavailability rapidly decreases (Karch et al., 2004). Decline in DFR is typically a first-order exponential process, although it may be biphasic for some chemicals (Whitmyre et al., 2003). TCDD dissipation from a surface (primarily due to photolysis) has a half-life 1–3 days (Karch et al., 2004).

Although soil dust has been postulated as a significant determinant of exposure in very dry soils during low atmospheric humidity, it has become recognized that this is a minor source of post-application inhalation exposure under most conditions (Gunther et al., 1977). This has been reiterated specifically for polychlorinated dibenzodioxins for individuals living in dusty areas, where inhalation, dermal and ingestion exposure from the environmental source accounted for ~1% of total exposure (Nadal et al., 2004). There are several reasons for the relatively minor exposure from soil. These include very low concentration on soil immediately after application (except for spills), dilution with untreated soil and the reduced dermal bioavailability of chemicals bound to soil (Driver et al., 1989). Thus, direct deposition and dermal contact with treated foliage are the main pathways of interest to exposure assessment for airborne application of pesticides.

Information for the Vietnam Exposure Scenario

To properly evaluate any exposure assessment model one must understand the process resulting in exposure. In this context, the historical record is relevant to the issue of exposure assessment modeling for two reasons. First, it shows that there was an elaborate planning process for herbicide spray missions, and locations and timing were determined well in advance of actual herbicide application. Moreover, there was a significant effort expended to make sure that friendly forces were not in the area to be sprayed. To insure that friendly forces had left the designated spray area, an individual called the “forward air controller” or FAC would fly over the area in a small plane and if evidence of friendly forces in the area to be sprayed was observed the mission would be canceled (Young et al., 2004a).

Young et al. (2004a) also note that ground fire from enemy or “hostile forces”, directed at spray planes was a regular occurrence and resulted in fighter aircraft strafing and bombing the area to be sprayed before the spray planes made their run. If American and Allied ground troops were in the area to be sprayed immediately before the spray mission, the likelihood of casualties from friendly fire seems high. Young et al. (2004a) found no reports of such casualties.

Another important historical point is the existence of an antimalaria mosquito-control program called “Operation Flyswatter” (Young et al., 2004a). This program sprayed the insecticide malathion in and around areas occupied by American and Allied troops, and used the same type of aircraft (and in some cases, the same aircraft with different spray apparatus) as was used in the “Ranch Hand” herbicide spray program. The authors suggest that this may be the source of eyewitness accounts of troops being directly sprayed by fixed-wing aircraft.

The historical record also gives us the Herbicide Reporting System (HERBS) database, which is the source of data on the location of defoliation missions. The geographic coordinates given by HERBS are imprecise. That is, Vietnam era operations predate the advent of GPS systems so the assigned mission coordinates were estimated using maps and landmarks. Young et al. (2004a) discuss both the development of the HERBS database and the problem with precision of locations. Moreover, only a small number of coordinates are used to locate each mission. For many missions, one has only a start coordinate and an end coordinate, whereas for others one may have one or more turn coordinates marking major changes in flight path. Thus, mission paths are necessarily approximated as a series of straight-line segments that is also imprecise in the sense that mission paths were in fact often curved to follow the terrain or the defoliation target such as jungle along a valley, road or river. See Figure 1 for illustration of this point.

Figure 1.
Figure 1 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact or the author

An illustration of the idea of the nominal (straight line) versus actual spray path (curved).

Full figure and legend (33K)

A last point is that troop location data give general locations of units but individual-level location data are not available. The general imprecision in the available data is acknowledged by the authors of the EOI model. Stellman and Stellman (2004) state that Another potential limitation is the fact that the GIS is structured so that exposure at any location within each 1.2km2 cell is assigned the exposure of the centroid. We justify this by noting that while the nominal precision of the HERBS coordinates is 100m, their accuracy must be closer to 500m. This exceeds the accuracy with which troop locations are known from military records, so that any attempt to use more “exact” distances could be considered an exercise in false precision. Thus, the question is, are meaningful individual exposure assessments possible given the imprecision inherent in the data?

The Stellman EOI Model

The Stellman EOI model was developed in response to the request for proposals (RFP) issued by the 1996 IOM Committee on the Assessment of Wartime Exposure to Herbicides in Vietnam and has been described in several publications (Stellman et al., 2003; Stellman and Stellman, 2004, 2005). The model consists of two components. Stellman and Stellman (2004) describe these as We partition exposure opportunity into two components, one representing direct and the other indirect exposure. For every direct hit, we compute an additional E4 term in which the residence time extends from the date of spraying until 3 days later, with a half-life of 1 year, and multiply it by a correction factor to represent the approximate ratio of dermal to respiratory absorption. The result is added to that obtained using the actual residence time and postulated half-life. This correction avoids the situation whereby an individual who was not directly sprayed could accumulate a larger EOI score than someone who was, simply by staying in a once-sprayed location for a very long time. The actual exposure opportunity score is an inverse weighted average of the distance from the flight line modified by the amount of herbicide sprayed. Stellman and Stellman (2004) give the following equation for EOI:

Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact or the author


Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact or the author

Here, f is the EOI score, which the authors term the E4 score, λ is the spray density (gallons of spray divided by length of the spray path), (X, Y) are the coordinates of the receptor or “exposed unit” in the common Cartesian coordinate system, D(xs) is the distance along the spray path, m is the “slope” of the spray path relative to the grid to which exposure is assigned and b is its intercept.

The EOI score is also modified by time, and is reduced as a first order exponential decay process with a default half-life of 30 days, but which can be reduced to as little as 1 day or increased at the investigator's discretion. Here, time is time post spraying.

There are some ambiguities associated with this model formulation. First exposure is weighted by the reciprocal of distance from the spray path (Stellman and Stellman, 2004). The problem with a strict inverse distance weighting is that as distance tends toward zero its reciprocal tends toward infinity. One fix for this is to define some maximum distance, Dmax, within which the weight is always one and then scale distances greater than Dmax by dividing them by Dmax. The problem is that the rate at which weights decline as one moves away from the flight line depends on Dmax. That is, if Dmax=0.1km then the weight at 4km is 1/40 but if Dmax=1km then the weight is 1/4. The writings by Stellman et al. (Stellman et al., 2003; Stellman and Stellman, 2004) suggest that Dmax may be 0.5, but this issue is not explicitly addressed. Another fix is to scale distances greater than Dmax by subtracting Dmax and adding 1 to the result to obtain a new scaled distance, D*. That is,

Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact or the author

here, when D=Dmax, D*=1 and 1/(D*) declines smoothly from 1 toward its limiting value 0 (when D*=∞). Here, the problem is units. If D* is given in km, the weight for 4km would be 1/4 but if D* is given in m, the weight at 4km is 1/4000. We will return to this point in our discussion.

A second issue is that E4 scores are defined over a large area. The total area where there is “exposure opportunity” according to the Stellman model is about 218.5km2, assuming a spray mission track length of 14km. This is developed as:

  • A main track area of 14km and a hit zone 10km wide (5km on either side of the mission path), which gives 140km2 (14km × 10km=140km2).
  • Plus two half circles 5km in radius. The area of a circle is the radius squared (25km2) times π (about 3.1416). So two half circles equal one whole circle with a total are 78.54km2 (25km2 × 3.1416).
  • Adding the main track box plus the two end half circles gives us 218.5km2.

Note that this is the geometry described in Figure 1 of Stellman and Stellman (2004). Our work with AgDRIFT, described below, suggests that actual area where any meaningful exposure could occur is much smaller. The central question is how is EOI distributed over such a large exposure area? For example, what percentage of a “full exposure score” is assigned to a point 4km from the flight path? Likewise, does the exposure score at a given distance from the flight path vary with distance along the flight path? We will address this issue empirically in the following section.



Empirical Tests of the Stellman Model

The Stellman model uses Eq. 2 plus first-order decay kinetics to assign exposures to points on a grid with approximately 1.2km spacing. If more than one mission affects a grid point, the EOI scores are summed for this grid point (Stellman and Stellman, 2004). Thus, if a given point in space and time is specified, its EOI score will often be affected by several missions, and the spatial distribution EOI score for a particular mission will be hard to determine. However, spray missions that are “isolated” in time and space can be identified and one can determine the E4 scores for “locations of interest” around these spray missions. Because the Stellman model does not calculate EOIs for locations beyond 5km of a spray mission, we first require that any potential “isolated mission” must be separated from all contemporaneous missions by at least 12km (5km for each mission plus a 2km safety factor). In defining contemporaneous, we assume that missions at least 4 months before or 1 month after would have little impact on the EOI calculations for a given mission. Here, the reasoning is that, using the 30-day half-life that the Stellman model assumes by default means that after 4 months, only one-sixteenth of the original EOI remains. Thus, our criteria for an isolated mission are ones, which have no other mission path within 12km of the isolated mission for a period of 4 months before or 1 month after the isolated mission. To identify representative missions, we also eliminated those with a total less than 10km flight path and restricted attention to flights involving fixed-wing aircraft (as opposed to helicopters) focused on defoliation missions (as opposed to crop destruction). The ArcView GIS system (ESRI, 2000) was used in conjunction with the version of the HERBS database provided with the Stellman model (Version 1.0.2) to select spray missions that met these criteria. Our search identified a total of seven missions, which met these criteria. The mission characteristics together with their identifiers are listed in Table 1.

For an illustrative quantitative evaluation, we selected missions 5093, 3108 and 6137. For each of these missions, we calculated the EOI score for the 36 relative locations described in Table 2 and shown in Figure 2. We also calculated EOI for two time periods, the day of application and from days 2 to 3 post application. The results of these calculations are summarized in Table 3. We will defer detailed consideration of these results to our discussion but will comment on one major anomaly of the results. Location 18 of Mission 5093 has an EOI value of 60,791. As this point is beyond 5km, one would expect it to have an either 0 or a very small EOI score, as all other beyond 5km points do have for the three missions. In tracing down the source of the anomaly, we first recalled that the EOI scores are calculated on a 1.2-km2 grid (Stellman and Stellman, 2004). Further inquiry showed us that latitude and longitude location data can only be entered into the model to two decimal places (e.g., the nearest 100th of a degree). South Vietnam is in the latitude range of 8–18. The length of a degree of latitude (measured along the longitude) in this range is approximately 110km. The length of a degree of longitude (measured along the latitude) varies from 105 to 110km. Thus, in the South Vietnam area, limiting decimal degree values to two decimal places creates an environment that can have an accuracy no better than approximately ±1km. As 60,791 is quite close to the value calculated for location 17, which is 4km from the flight line, we believe that the anomalous result is a function of exposure calculated on a 1.2-km2 grid and the model restricting location entry to a two decimal level of precision. The question we will consider is whether a minimal accuracy of more than 1km is adequate for meaningful exposure assessment.

Figure 2.
Figure 2 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact or the author

An illustration of the locations selected for calculation of E4 scores.

Full figure and legend (17K)

Empirical Results from AgDRIFT

For estimating drift exposure potential of AO, the AgDRIFT (version 2.0.5) Tier III aerial forestry module was used. This model allows input of aircraft type from a large library, nozzle diameter and droplet size, height of the spray boom above ground, boom length, swath definition, physicochemical characteristics of the spray material, meteorology and canopy. In Table 4, the specific values used to estimate ground deposition through canopy are shown. A few specifics are noteworthy. First, the plane is assumed to be flying in a 5-m/h crosswind that is 90° from the flight path in our analyses, but wind speed and angle are user definable. This results in spray particles moving predominantly at a right angle to the flight path as they move downwind in our analysis. Without this assumption of crosswind, there would be relatively even distribution of particulates to either side of the flight path. Second, we assumed a particle size distribution consistent with observations made using this particular spraying system and AO that produced a mass median diameter on impact from a 150-foot height of 367μm (Harrigan, 1970). This is equivalent to an ASAE “coarse” particle rating. There is no modern system comparable to the spray system used in Vietnam. This is because the “nozzles” used in the A/A45Y-1 spray system in Vietnam were, in fact, spigots. That is, each of the 42 outlets on the spray bars was a 0.375-inch diameter check valve from Spraying Systems Co. mounted 90° to the flight plane and operated at 30p.s.i. (Harrigan, 1970). Both assumptions used in the model produced a drift pattern that would constitute an upper bound for the distance that particles would move from the point of release.

Figure 3 is a plot of the output from the AgDRIFT model using the conditions specified above obtained from the terrestrial deposition file, noting that the output is referenced to the plane's centerline with a negative swath displacement. Note that zero on the x axis represents the centerline of the flight path. This Figure demonstrates that at approximately 100m from the flight line, there is virtually no spray (<0.2%) reaching the ground downwind (i.e., it has been intercepted by the forest canopy). The deposition values represent the concentration of the phenoxy herbicide droplets reaching the ground level through a canopy with a LAI of 4.1. This is representative of moderately forested regions of Southeast Asia. The deposition on the upwind side is somewhat compressed and the curve is shifted to the right. Had we assumed a lower wind speed, the deposition pattern would be a more symmetrical Gaussian curve with arms centered more toward “0.” Similarly, higher wind speed would shift the curve slightly to the right. In Figure 3, the wind has shifted all but the small amount (not pictured) to the left of the 0 (centerline) by about 22m to the right. Note that peak deposition under canopy is ~0.7% of the rate being sprayed by the airplane.

Figure 3.
Figure 3 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact or the author

Deposition of phenoxy herbicide at ground level through a forest canopy as a function of distance from the flight line.

Full figure and legend (27K)

Three factors that could have varied in applications over Vietnam were tested in the AgDRIFT model for their influence on the deposition at ground level. We varied LAI from 2.5 to 4.1 (the range recorded in the Forest Service Library within AgDRIFT) to determine the effect of canopy on ground deposition, which produced the most significant difference. With less canopy density (LAI=2.5), both ground deposition and distance approximately doubled from that shown in Figure 3. Increased crosswind speed outside the canopy to 10m/h (near maximum allowable for aerial herbicide applications; 10 knots; Young et al., 2004a) decreased ground deposition and increased distance by ~40m. Increasing the release height over canopy from 120 to 160ft resulted in less than doubling the distance of dispersion and slightly reduced maximal ground deposition.



Our AgDRIFT modeling of a C123 spraying of herbicide shows that a single plane will have a spray swath of around 50 to 100m at ground level through the canopy. Moreover, there is little spray deposition outside the swath and deposition decreases very rapidly beyond the nominal spray swath. These modeling results are similar to the empirical results from Air Force studies reported by Young et al. (2004b). As a spray swath is at most 100m per plane (Harrigan, 1970) and the HERBS database included as part of the Stellman model shows that approximately 79% of herbicide spray missions involved four or fewer aircraft, it follows that, because spray missions were usually 14km or less (Young et al., 2004b), the typical spray mission covered less than 5.6km2 (14km long × 0.4km wide).

In contrast, the EOI model, for the same 14-km mission would assign scores that are, at a minimum, within eight-fold of a direct hit value to the entire area of about 218.5km2. The pattern of the magnitude of assigned scores (Table 3) shows that even at 4km (more than 2 miles) perpendicular to the flight path (see point 26 for example) the assigned EOI score is more than 20% of the maximum EOI score (points 11 or 12). We have even seen an EOI score of more than 10% of the maximum EOI score assigned as an apparent error (point 18 of Mission 5093). We also see that points A–F, which are directly on the flight path are often equaled or exceeded by points well away from the flight path (e.g., points 11 and 12). This is truly curious because point 12 is 800m perpendicular to the left of flight direction. The average E4 score of points 800m perpendicular to the centerline (points 6, 8, 10 and 12 in Table 3) is ~90% of the E4 score of points B, C, D and E, respectively, that are on the centerline in Table 3. Our AgDRIFT modeling shows that even for an eight-plane mission, which is quite large (according to the HERBS database, less than 3% of defoliation missions involved eight or more aircraft), exposure at 800m, as measured by ground deposition, is less than 1 × 10−21% of the exposure under the spray swath. Even with a 5-m/h crosswind, <0.001% spray reaches more than 150m beyond the peak of the spray swath through the canopy. Since an eight-plane mission has a swath width of ~800m, the centerline plus 1/2 the swath width plus 150m is only 550m or 250m short of 800m. Across the entire 10km width in Figure 2, the E4 score varies by less than 10-fold. This is a miniscule change compared with AgDRIFT estimates on just the downwind side, where the potential exposure varies over 23 orders of magnitude in only 0.8km.

We also note that the E4 scores for points A–F vary considerably. This is curious because all are directly on the mission path. As the object of the spray system is to achieve a uniform level of spray deposition, it would seem that all should have the same score. That is, if three men were directly under the spray path but at different points along the spray path one would expect them to receive very similar exposures. The pattern observed is expected from Eq. (2) because if one simply applies the inverse distance weighting scheme, points near the ends of the spray path receive most of their exposure from the side toward the center of the spray path, whereas points near the center of the spray path can receive exposure from both sides equally. However, until we obtained the results shown in Table 3, we had assumed that there was a “direct hit zone” for which all exposures were the same. We now know that this is not the case, but it should be. Assigning markedly different (2 × difference in some cases) EOI values to different points along the same mission spray centerline makes no sense.

Another interesting aspect of the E4 scores is that, in all cases, the E4 score for entry after spraying on mission day plus 2 and exit on mission day plus 3 is about 0.4% or 1/250th of the mission day EOI score. Although we would agree that by far the largest source of exposure would be from being directly sprayed (assuming such an event occurred), we question whether it is reasonable to assume that exposure at 1.5km from the mission path on the day of application is actually 100 times higher than being on the mission path 2 days after spraying occurred (consider locations 22 and B).

Unfortunately, there are many such examples in Table 3. A fairly high score is assigned to position 16, which is 1.5km behind the start of the flight path, but given the conditions under which herbicides were applied (Young et al., 2004b) there is no way spray could have reached this location.



Given the enormous numbers of anomalies identified in Table 3, one might ask whether our test as opposed to the Stellman model peculiarities is at fault. First, we reiterate that Eq. 2, which is at the heart of their model, is contrary to a large body of pesticide exposure assessment practice (e.g., Teske et al., 1993; Whitmyre et al., 2005), and would not be expected to produce realistic estimates. Second, we have checked our selection of coordinates for the 36 test points for each mission several times and have plotted these coordinates relative to the mission coordinates to make sure that the distribution looks like that portrayed in Figure 2. We have also checked to make sure that our isolated missions were in fact isolated, but note that this really makes little difference. There is a huge drop in E4 score from mission day score to “in day 3 out day 4” score. Thus, if missions were separated by even a week, the pattern in mission day scores would be little affected. Last, with the exception of point 18 for Mission 5093, all of the scores that should be 0 or nearly so are in fact 0 or nearly so. Thus, we are quite sure that our model test is valid.

We also note that the Stellman and Stellman (2004) have presented some justification for their model in the form of arguments such as “To be of use in epidemiological studies, an exposure score should have reasonable statistical properties, including a distribution broad enough to provide adequate contrast between persons or places with high versus low exposure. There were 112,466 cells that had nonzero E4 scores. These scores ranged over six orders of magnitude, from 0.24 to 4,241,013 with a mean of 161,870 and standard deviation of 322,784, and exhibited approximate log-normality (Figure 4). Table 3, which presents deciles for nonzero exposures, shows that good discrimination is attained among the highly exposed strata” (Stellman and Stellman, 2004). We would note that the E4 score, which defines the 90th percentile in the Table 3 mentioned above is 447,355. Perusal of our Table 3 shows that this could be achieved by being in the general vicinity of (but not exposed by) three missions on the day that spraying occurred. Good statistical properties in an exposure score are nice, but it is even better if the score has some association with actual exposure, which we would suggest the Stellman E4 score does not. In the example calculations we summarized in Table 3 using the EOI model, the difference in the EOI score directly on the flight line varied more than four-fold. This is clearly incorrect because exposure should all be uniform along the flight line. In those same examples, over the entire 5km distance on either side of the flight line, the EOI score varied by only eight-fold. Thus, it appears that over half of the variance in day of application EOI scores may be due to idiosyncrasies of the method and have nothing to do with “reasonable statistical properties” attributed to the model.

Figure 4.
Figure 4 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact or the author

The area to which the Stellman E4 model would assign the EOI scores, assuming a 14-km mission. The 0.5-km “high-exposure” area and the actual sprayed area for a four-plane mission are included for comparison.

Full figure and legend (59K)

Stellman et al. (2003) also suggest that their methodology is in some sense equivalent to a job-exposure matrix: “The utility of the GIS as an exposure reconstruction instrument depends, of course, on how accurately the E4 and hit scores can act as proxies for true toxicologic exposure. Proximity to an environmental insult is a widely used concept and various schema such as job-exposure matrices rely on this conceptual premise. In the case of exposure to herbicides in Vietnam, which began 40 years ago and ended 30 years ago, no other reliable measure is available for large-scale epidemiologic studies.” We would disagree. Job-exposure matrices apply to long-term exposures in a relatively homogeneous industrial process. Herbicide exposure of ground troops, when it did occur, was a rare event having more in common with an industrial accident. Moreover, job-exposure matrices must be calibrated by something like industrial hygiene data. There is ample evidence that no such calibration contributed to the development of the E4 score.

Another question that arises is “even if the Stellman EOI model is faulty could one not develop a correct model for assessing exposure to ground troops?” Here, the issue is that the area where exposure could actually occur is represented by a narrow ribbon whose width is determined by the number of planes × 100m (see Figure 4 for an example). Thus, a two-plane mission has a swath of 200m and a four-plane mission has a swath of 400m. In all, 79% of missions involved four or fewer aircraft, so most missions had a quite narrow swath. It is also true that the flight coordinates in HERBS are probably not precise to more than about 500m and the actual flight path is complicated by factors such as those illustrated in Figure 1. Also as Young et al. (2004b) point out, a single plane spraying its entire 1000 gallon payload could cover a spray swath no more than 14km in length. Consider Mission 5093 in Table 1. Three planes sprayed 3000 gallons, so each plane sprayed 1000 gallons, which is possible, but the mission length is 16.74km, which is 20% longer than continuous spray application would allow. This is by no means the longest mission in HERBS, so one must assume that the spray was sometimes turned on and off. The problem is we do not know what parts of the nominal mission path were actually sprayed in those cases. Moreover, troop locations at the individual level are clearly less precise than flight coordinates. Thus, absolute error levels in terms of co-location of spray missions and individual troops are probably on the order of a few kilometers.

One should also recall that, as discussed above, there were programs in place to prevent friendly forces from being sprayed (Young et al., 2004a) and that “Operation Flyswatter” sprayed the insecticide malathion in and around areas occupied by American and Allied troops, and used the same type of aircraft (Young et al., 2004a). Thus, apparent overlaps between troop locations and spray missions are quite likely false-positive reports and self-reports of being directly sprayed could well be malathion, not herbicide. We also note that one could observe herbicide spray missions from a distance and still be at no risk of actual exposure, so even reports of being near spray missions would not be very helpful. As for reentry into recently sprayed areas, we reiterate that the half-life of dislodgeable dioxin on foliage is arguably quite short (Karch et al., 2004). Recently sprayed areas were often occupied by enemy troops and also posed a threat in the form of unexploded ordinance from the fighter missions that frequently preceded herbicide spray missions. Thus, there were powerful disincentives for reentering sprayed areas shortly after herbicide application (Young et al., 2004a). This reasoning makes us pessimistic about our ability to develop individual level exposure estimates for ground troops in the Vietnam War.

In summary, we believe that our analysis shows that the Stellman E4 model, which has been proposed as an exposure metric on which epidemiologic studies of Vietnam veterans could be based (Stellman et al., 2003; Stellman and Stellman, 2004, 2005), is badly flawed and should not be used as a basis for epidemiologic studies. It seems clear that high exposures could occur within at most a few hundred meters of the flight path; as for example, the nominal width of a four-plane mission. Even if we assume a potential exposure zone 1.25 times this width, the potential width of the exposure zone is only 0.5km. If we take 14km as a maximum length for the active spray portion of a mission (Young et al., 2004b), the potential direct exposure area is no more than 7km2. We note that while some herbicides did disperse outside this area, 98% of the droplets were 100μm or greater in diameter (Young et al., 2004b) and would impact within the area. Smaller droplets could disperse beyond the 7km2 area, but even for smaller droplet sizes, many would impact within the area. It is also true that a small percentage of the herbicide could volatilize and be transported to long distances, but if herbicide and, in particular, dioxin stayed aloft for an extended period it would be subject to substantial photodegradation (Young et al., 2004b). Thus, the amount of herbicide or dioxin that was available to cause exposure outside of the nominal spray swath is a small fraction of the amount of material deposited within the nominal spray swath.

As noted earlier, the Stellman model assigns E4 scores within a factor of 8 of maximum to an area of some 218.5km2, and this area is independent of the number of planes in the mission. Moreover, there was an active program in place to exclude Allied Ground Troops from areas where spraying was taking place, and there is substantial evidence that this program was effective (Young et al., 2004a). The inescapable conclusion is that essentially all high EOI scores are assigned to troops who had essentially no herbicide exposure. Thus, if one did epidemiologic studies based on the EOI scores, actual exposure in the high EOI group would differ little, if at all, from that in the low EOI group.

Nonetheless, such hypothetical studies might show health effects. It is certainly true that herbicide spray missions were an indicator of hostile activity and thus quite possibly an indicator of combat stress. Alternatively, destruction of jungle habitats might bring troops into contact with naturally occurring toxins and pathogens that they would not ordinarily encounter. In either case, an EOI-based epidemiologic study might show clear evidence of health effects, which, in fact, had nothing to do with herbicide exposure.

For reasons just discussed, we also are pessimistic about anyone's ability to develop individual-level exposure assessments for Vietnam veterans. One area where we believe progress is possible is to use modern pesticide exposure assessment techniques (e.g., Franklin and Worgan, 2005) to quantify exposure for a number of general exposure scenarios. Such studies have value because they can tell us how dioxin exposures in ground troops compare with those of well-studied groups such as the veterans who were involved in the Ranch Hand Program, which was the major herbicide application program for the Vietnam War. These soldiers had well-documented exposure to herbicides, and, in some cases, their exposure is verified by elevated dioxin body burdens. However, evidence of adverse health effects in this cohort is not compelling. That is, despite careful scrutiny, little in the way of elevated disease risk was observed (AFHS, 2005). Getting insight as to how hypothetical highly exposed ground troops compare with Ranch Hand veterans would provide some basis for evaluating the extent to which health problems in the general Vietnam veteran population could be a result of dioxin or herbicide exposure.



  1. Air Force Health Study, Final Report. An epidemiologic investigation of health effects in Air Force personnel following exposure to herbicides. 2002 follow-up examination results. Contract number F41624-01-C-1012. 31 March 2005.
  2. Anderson Y.B., Jackson J.A., and Birnbaum L.S. Maturational changes in dermal absorption of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) in Fischer 344 rats. Toxicol Appl Pharmacol 1993: 119(2): 214–220. | Article | PubMed | ChemPort |
  3. Bird S.L., Perry S.G., Ray S.I., and Teske M.E. Evaluation of the AGDISP aerial spray algorithms in the AgDRIFT model. Environ Toxicol Chem 2002: 21: 672–681. | Article | PubMed | ChemPort |
  4. Butler D.A. Connections: the early history of scientific and medical research on "Agent Orange". J Law Policy 2005: XIII(2): 527–552.
  5. Dong M.H., Draper W.M., Papanek P.J., Ross J.H., Woloshin K.A., and Stephens R.D. Estimating malathion doses in California medfly eradication campaign using a physiologically based pharmacokinetic model. In: Draper W.M. (Ed.). Environmental Epidemiology: Effects of Environmental Chemicals on Human Health. American Chemical Society, Washington DC, 1994, 189–208.
  6. Driver J., Ross J., Mihlan G., Lunchick C., and Landenberger B. Derivation of single-layer clothing penetration factors from the pesticide handlers exposure database. Regul Toxicol Pharmacol 2007: 49(2): 125–137. | Article | PubMed | ChemPort |
  7. Driver J.H., Whitmyre G.K., Konz J.J., Roy T.A., Yang J.J., and Krueger A.J. Dermal exposure assessment for toxicants in soil: soil adherence to skin and dermal bioavailability. Proceedings of the EPA & A&WMA Specialty Conference "Total Exposure Assessment Methodology: A New Horizon", Las Vegas, NV, 1989. pp. 597–618.
  8. EPA. Series 875 — Occupational and Residential Exposure Test Guidelines, Group B: Post Application Exposure Monitoring Test Guidelines, U.S. Environmental Protection Agency, Office of Prevention, Pesticides, and Toxic Substances, Washington, DC, 1997.
  9. EPA. PHED Surrogate Exposure Guide: Estimates of Worker Exposure from the Pesticide Handler Exposure Database Version 1.1, EPA, Office of Pesticide Programs, Washington, DC, 1998.
  10. EPA. Policy 3.10 — Agricultural Transfer Coefficients. revised 7 August. Office of Pesticide Programs, Science Advisory Council for Exposure, Washington, DC, 2000.
  11. ESRI. ArcView GIS 3.2a. Environmental Systems Research Institute Inc., Redlands, CA, 2000.
  12. Franklin C.A., and Worgan J.P., (Eds.). Occupational and Residential Exposure for Pesticides. John Wiley and Sons Ltd., New York, 2005.
  13. Gough M. Dioxin, Agent Orange: The Facts. Plenum Press, New York, 1986.
  14. Gough M., (Ed.). The political science of Agent Orange and dioxin. Politicizing Science: The Alchemy of Policymaking. Hoover Press, Stanford, CA, 2003.
  15. Gunther F.A., Iwata Y., Carman G., and Smith C.A. The citrus reentry problem: research on its causes and effects, and approaches to its minimization. Residue Revs 1977: 67: 1–132. | ChemPort |
  16. Harrigan E.T. Calibration test of the UC-123K/A/A 45Y-1 spray system. Technical report ADTC-TR-70-36 Armament Development and Test Center, Eglin AFB, Florida, February 1970.
  17. Hewitt A.J., Johnson D.R., Fish J.D., Hermansky C.G., and Valcore D.L. Development of the Spray Drift Task Force Database for aerial applications. Environ Toxicol Chem 2002: 21: 648–658. | Article | PubMed | ChemPort |
  18. Institute of Medicine. Veterans and Agent Orange. Health Effects of Herbicides Used in Vietnam. National Academy Press, Washington, DC, 1994.
  19. Institute of Medicine. Characterizing Exposure of Veterans to Agent Orange and Other Herbicides Used in Vietnam: Scientific Considerations Regarding Request for Proposals for Research. National Academy Press, Washington, DC, 1997.
  20. Karch N.J., Watkins D.K., Young A.L., and Ginevan M.E. Environmental fate of TCDD and Agent Orange and bioavailability to troops in Vietnam. Organohalogen Compd 2004: 66: 3689–3694.
  21. Klein R.E., and Harrigan E.T. Comparison Test of Defoliants. Technical report ADTC-TR-30, Vol. I and II, Directorate of Test and Evaluation, Armament Development and Test Center, Eglin AFB, FL, 1969.
  22. Lavy T.L., Shepard J.S., and Mattice J.D. Exposure measurements of applicators spraying 2,4,5-trichlorophenoxyacetic acid in forests. J Agric Food Chem 1980: 28: 626–630. | Article | PubMed | ChemPort |
  23. Nadal M., Schuhmacher M., and Domingo J.L. Probabilistic human health risk of PCDD/F exposure: a socioeconomic assessment. J Environ Monit 2004: 11: 926–931. | Article | ChemPort |
  24. Newton M., and Norris L.A. Potential exposure of humans to 2,4,5-T and TCDD in the Oregon Coast Range. Fundam Appl Toxicol 1981: 1: 339–346. | Article | PubMed | ChemPort |
  25. Ramsey J.C., Lavy T.L., and Braun W.H. Exposure of forest workers to 2,4,5-T; calculated dose levels. Project Completion Report to the National Forest Products Association. 14 February 1979 (T.L. Lavy, Univ Arkansas, Director) Dow Chemical Co., Midland, Michigan.
  26. Ross J.H., Driver J.H., Cochran R.C., Thongsinthusak T., and Krieger R.I. Could pesticide toxicology studies be more relevant to occupational risk assessment? Ann Occup Hyg 2001: 45(Suppl 1): S5–S17. | Article | PubMed | ChemPort |
  27. Ross J.H., Driver J.H., Harris S.A., and Maibach H.I. Dermal absorption of 2,4-D: a review of species differences. Regul Toxicol Pharmacol 2005: 41: 82–91. | Article | PubMed | ChemPort |
  28. Ross J.H., Driver J.H., Lunchick C., Wible C., and Selman F. Pesticide exposure monitoring databases in applied risk analysis. Rev Environ Contam Toxicol 2006: 186: 107–132. | PubMed | ChemPort |
  29. Ross J.H., Thongsinthusak T., Krieger R.I., Frederickson S., Fong H.R., Taylor S., Begum S., and Dong M. Human clearance of malathion. Worker Health and Safety Branch, California EPA, Department of Pesticide Regulation, Sacramento, CA. HS-1617 1991.
  30. Stellman J.M., and Stellman S.D. Characterization of exposure to Agent Orange in Vietnam veterans as a basis for epidemiological studies. J Law Policy 2005: XIII(2): 505–525.
  31. Stellman J.M., Stellman S.D., Weber T., Tomasallo C., Stellman A.B., and Christian Jr R.A. Geographic information system for characterizing exposure to Agent Orange and other herbicides in Vietnam. Environ Health Perspect 2003: 111: 321–328. | PubMed | ISI |
  32. Stellman S.D., and Stellman J.M. Exposure opportunity models for Agent Orange, dioxin, and other military herbicides used in Vietnam, 1961–1971. J Expo Anal Environ Epidemiol 2004: 14: 354–362. | Article | PubMed | ChemPort |
  33. Teske M.E., Bird S.L., Esteru D.M., Curbishley T.B., Ray S.L., and Perry S.G. AgDRIFT®: a model for estimating near-field spray drift from aerial applications. Environ Toxicol Chem 2002: 21: 659–671. | Article | PubMed | ChemPort |
  34. Teske M.E., Bowers J.E., Rafferty J.E., and Barry J.W. FSCBG: an aerial spray dispersion model for predicting the fate of released material behind aircraft. Environ Toxicol Chem 1993: 12: 453–464. | Article |
  35. Teske M.E., Thistle H.W., and Eav B. New ways to predict aerial spray deposition and drift. J Forestry 1998: 96(6): 25–31.
  36. Tschirley FH Research report: response of tropical and subtropical woody plants to chemical treatments. Technical Report CR-13-67, U.S. Department of Agriculture, Agricultural Research Service, Washington, DC, 1968.
  37. Whitmyre G.K., Ross J.H., Ginevan M.E., and Eberhart D. Development of risk-based restricted entry intervals. In: Franklin CA and Worgan JP (Eds.). Occupational and Residential Exposure for Pesticides. John Wiley and Sons Ltd., New York, 2005.
  38. Whitmyre G.K., Ross J.H., Lunchick C., Volger B., and Singer S. Biphasic dissipation kinetics for dislodgeable foliar residues in estimating postapplication occupational exposures to endosulfan. Arch Environ Contam Toxicol 2003: 46(1): 17–23. | Article | ChemPort |
  39. Williams R.L., Bernard C.E., and Krieger R.I. Child-like mannequin for observation and measurement of spray drift particle deposition for pesticide exposure assessment. Bull Environ Contam Toxicol 2001: 67: 179–186. | Article | PubMed | ChemPort |
  40. Yates W.E., Akesson N.H., and Coutts H.H. Evaluation of drift residues from aerial applications. Trans ASAE 1966: 6: 389–397.
  41. Young A.L., Cecil Sr P.F., and Guilmartin Jr J.F. Assessing possible exposures of ground troops to Agent Orange during the Vietnam War: the use of contemporary military records. Environ Sci Pollut Res Int 2004a: 11: 349–358. | Article | ChemPort |
  42. Young A.L., Giesy J.P., Jones P.D., and Newton M. Environmental fate and bioavailability of Agent Orange and its associated dioxin during the Vietnam War. Environ Sci Pollut Res Int 2004b: 11: 359–370. | Article | ChemPort |


This work was supported by the Dow Chemical Company and Monsanto Company. We also thank Randy O’Boyle of Exponent for his excellent GIS programming work.


These links to content published by NPG are automatically generated.


Ho Chi Minh City conference: Defoliants in Vietnam: the long-term effects

Nature News and Views (17 Mar 1983)

Pesticides and the environment

Nature News and Views (07 Oct 1976)