HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (7) Citing Articles via Google Scholar Google Scholar Articles by Raupach, M.R. Articles by Edge, V.E. Search for Related Content PubMed PubMed Citation Articles by Raupach, M.R. Articles by Edge, V.E. Agricola Articles by Raupach, M.R. Articles by Edge, V.E. Related Collections Water Quality Cotton Agricultural Pesticides Water Pollution

SPECIAL SUBMISSIONS
Minimizing the Impact of Pesticides on the Riverine Environment in Australia

Endosulfan Transport

I. Integrative Assessment of Airborne and Waterborne Pathways

^a CSIRO Land and Water, GPO Box 1666, Canberra, ACT 2601, Australia
^b Dep. of Land and Water Conservation, P.O. Box 3720, Parramatta, NSW 2150, Australia
^c NSW Agriculture, Locked Bag 21, Orange, NSW 2800, Australia

Corresponding author (mike.raupach{at}cbr.clw.csiro.au)

Received for publication October 8, 1999.

	ABSTRACT

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX A: Physical and... APPENDIX B: The "Steady-Drizzle"... APPENDIX C: Soil Water... REFERENCES

 
To reduce endosulfan (C₉H₆O₃Cl₆S; 6,7,8,9,10,10-hexachloro-1,5,5a,6,9,9a-hexahydro-6,9-methano-2,4,3-benzodioxathiepin 3-oxide) contamination in rivers and waterways, it is important to know the relative significances of airborne transport pathways (including spray drift, vapor transport, and dust transport) and waterborne transport pathways (including overland and stream runoff). This work uses an integrated modeling approach to assess the absolute and relative contributions of these pathways to riverine endosulfan concentrations. The modeling framework involves two parts: a set of simple models for each transport pathway, and a model for the physical and chemical processes acting on endosulfan in river water. An averaging process is used to calculate the effects of transport pathways at the regional scale. The results show that spray drift, vapor transport, and runoff are all significant pathways. Dust transport is found to be insignificant. Spray drift and vapor transport both contribute low-level but nearly continuous inputs to the riverine endosulfan load during spraying season in a large cotton (Gossypium hirsutum L.)-growing area, whereas runoff provides occasional but higher inputs. These findings are supported by broad agreement between model predictions and observed typical riverine endosulfan concentrations in two rivers.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX A: Physical and... APPENDIX B: The "Steady-Drizzle"... APPENDIX C: Soil Water... REFERENCES

 
COTTON has been one of the success stories of Australian rural industries, currently being worth $1.5 billion per year. However, this growth has had environmental effects, including the movement of agricultural chemicals into rivers. In the irrigated cotton region in central and northern New South Wales, Australia, a major monitoring program (the Central and North West Regions Water Quality Program) has been conducted by the New South Wales Department of Land and Water Conservation from the early 1990s onward (Cooper, 1996; Muschal, 1997, 1998). This has revealed the presence of several pesticides in rivers near and downstream of cotton-growing areas during the growing season. In particular, spot-sampled riverine concentrations of the insecticide endosulfan are broadly in the range 0.02 to 0.2 µg L^-1 during the growing season (November to March). These values significantly exceed environmental guidelines for protection of ecosystems, currently 0.01 µg L^-1 (Australian and New Zealand Environment and Conservation Council, 1992).

Such findings led to the multidisciplinary research program "Minimising the Impact of Pesticides in the Riverine Environment", funded by several agencies (see Acknowledgments), which aimed to determine the transport and fate of pesticides applied to cotton, assess the ecological effects, and develop improved management practices. The present work, a contribution to that program, is focused on transport pathways for endosulfan. Knowledge of the relative magnitudes and behavior patterns of exposure pathways is needed for effective management and to avoid expensive efforts aimed at closing insignificant pathways.

Transport pathways can be broadly distinguished as airborne or waterborne. When observations of high riverine endosulfan concentrations in northern New South Wales first began to appear in the early 1990s, it was widely believed that waterborne transport was mainly responsible. However, subsequent observations demonstrated that waterborne pathways alone could not account for many observed features of the contamination patterns—in particular for their regularly seasonal temporal behavior (Cooper, 1996; Muschal, 1997, 1998), which contrasts with the typically episodic, event-driven nature of runoff into rivers in the region. This pattern prompted attention on three possible airborne pathways. The most obvious is spray drift onto off-target receptor surfaces. At the same time, it was suggested on theoretical grounds that vapor transport might be significant (Raupach et al., 1996). This pathway involves volatilization of endosulfan into the atmosphere followed by dispersal and deposition of vapor. Field experiments in 1995 (Edge et al., 1998) confirmed the existence of the vapor transport pathway, but did not quantify its magnitude relative to other pathways. A third possible airborne pathway is dust transport, the windborne movement of contaminated dust, though early work (Raupach and Briggs, 1996) suggested that it is not likely to be significant. Waterborne pathways include runoff from farms to rivers during storms, subsurface percolation, and deliberate discharge of tailwaters from holding ponds into rivers. Of these three, runoff can transport endosulfan either in dissolved form or attached to suspended sediment, and has always been a primary concern. Percolation over long pathways is not likely to be significant because of the tendency for endosulfan to bind to soil particles and hence become immobilized for long enough to allow degradation to occur (Kennedy et al., 1998). Deliberate discharge is a management issue: it was probably partly responsible for high riverine concentrations in northern New South Wales during the first year of monitoring in 1991–1992 (Cooper, 1996; Muschal, 1997, 1998), but has been greatly reduced since then by the development and implementation of guidelines for best management practices in the cotton industry (Schofield, 1998; Anthony, 1998).

This brief survey indicates that significant questions remain about the relative significance in the New South Wales riverine environment of the spray drift, vapor transport, dust transport, and runoff pathways. Of particular concern is vapor transport, because it was novel when first suggested in this context and because very little information has hitherto been available to quantify it.

There are a number of existing models relevant to the broad problem of determining the concentrations and transport pathways of contaminants in water bodies. Overviews have been provided by Donigian and Huber (1991) and Singh (1995). Prominent examples of models for waterborne transport pathways are the HSPF model (Donigian et al., 1995), which simulates hydrology and water quality at the catchment scale for both nutrients and toxic organic pollutants, and the CREAMS and GLEAMS models (Knisel, 1980; Leonard et al., 1987), which provide similar capabilities at the scale of field plots. The GLEAMS model has been applied elsewhere in the "Minimising the Impact of Pesticides in the Riverine Environment" research program to study the effects of tillage practices on endosulfan transport by runoff (Connelly et al., 1998).

Despite the existence of a number of partly relevant models, we decided to use a different approach in the present work for four reasons. First, a major focus is the exploration of processes (vapor transport and dust transport) not represented at all in the suite of existing models. New model development is needed to simulate these processes. Second, we are concerned with regional rather than local transport. This demands extensions to existing approaches for some pathways, especially spray drift, where extant models are concerned mainly with local drift from particular sprayed fields. Third, effective comparisons between different pathways require a common modeling approach as far as possible. For example, all airborne pathways involve the three stages of mobilization from a source, atmospheric transport, and deposition to a receptor, of which atmospheric transport and some aspects of the deposition stage are common to all airborne pathways. Finally, we wished to avoid the problem of becoming hostage to the assumptions of an existing model. The key assumptions are not usually in the representation of biophysical processes, but in the choice of which processes to represent.

The present work is an integrated modeling investigation into the relative magnitudes and properties of the spray, vapor, dust, and runoff pathways for endosulfan transport. Our emphasis is on the airborne (spray, vapor, and dust) pathways, runoff being included to the extent necessary to relate its typical magnitude and properties to the other pathways. The work is reported in a pair of papers of which this is the first. Here, process models and data are combined to assess the contributions of each major transport pathway to riverine endosulfan concentrations. We concentrate on data from two rivers, Pian Creek and the Namoi River, in the (mainly) irrigated cotton-growing area of northern New South Wales (see Fig. 1 for a map). The second paper (Raupach et al., 2001; hereafter Paper II) describes the details of the process representations for the airborne pathways.

View larger version (22K): [in this window] [in a new window]   Fig. 1. Location of the focus area, showing the cotton-growing areas and major waterways in the mid and lower reaches of the Namoi River catchment, northern New South Wales, Australia. Cotton areas and catchment boundary from Peasley (1996) provided by Department of Land and Water Conservation, Gunnedah, NSW.

	THEORY

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX A: Physical and... APPENDIX B: The "Steady-Drizzle"... APPENDIX C: Soil Water... REFERENCES

[1]

where C(X,t) is the contaminant concentration (kg m^-3), X is streamwise distance along the river (m), t is time (s), U is the flow velocity in the river (m s^-1), H is the water depth (m), F(i) is the flux density (flux per unit horizontal surface area of the river) entering the water column by pathway i (kg m^-2 s^-1), and S(j) is the chemical source strength (or sink strength if negative) arising from transformation process j (kg m^-3 s^-1). The quantities C, U, and S(j) are assumed to be averaged over the breadth and depth of the river, and F(i) is averaged over the river breadth. We consider four flux terms F(i), representing spray drift, vapor transport, dust transport, and runoff, and three chemical transformation terms S(j), representing oxidation, hydrolysis, and particle–water exchange (as described below). The transformation terms S(j) are functions of the riverine concentration C, whereas the flux terms F(i) are usually independent of C and represent the external forcing on the system, which is highly variable in both space and time.

The modeling framework involves solving Eq. [1] with suitable specifications of F(i) and S(j). However, it is necessary to simplify the enormous space–time complexity in the fluxes F(i) arising from variability in the external forcing. This is done by defining scenarios for each transport pathway to characterize the typical magnitude and time dependence of the flux, with space and time averaging over details of the external forcings, which do not affect the eventual solutions of Eq. [1]. The scenarios are constructed with the aid of more detailed process-based models for each pathway.

In the rest of this section, the physical chemistry of endosulfan is briefly reviewed; the models and scenarios for the fluxes F(i) through the spray, vapor, dust, and runoff pathways are outlined; and the transformation processes S(j) are described.

Two processes of possible significance are explicitly omitted: deliberate discharge and sediment–water exchanges (though each could be included by an extension to the general framework). Deliberate discharge is not included for reasons given in the Introduction. Sediment–water exchanges do not introduce extra endosulfan into the system, but cause a temporal smoothing by exchange of endosulfan between the water column and a group of bottom sediment pools with slow turnover times. There is also the possibility that chemical degradation processes in the sediment happen differently (for example, with modified rate constants) than the equivalent processes in the water column. Though sediment–water exchanges are not treated explicitly, we discuss their likely effects later.

Properties of Endosulfan in Riverine Systems
Endosulfan exists as and ß isomers, occurring in the ratio 2:1 in the technical endosulfan applied to cotton crops in spray form. Once in the natural environment, endosulfan is subject to several chemical and physical transformations: first, the and ß isomers hydrolyze in water to endosulfan diol, which is much less toxic and can be regarded as a sink for endosulfan from the standpoint of toxicity. Second, both the and ß isomers oxidize to endosulfan sulfate in the presence of biotic material. This occurs on leaf surfaces, in the soil, and in natural water bodies because of their typical biotic content, but it does not occur in sterile water (Kennedy et al., 1998). The , ß, and sulfate species are all of comparable toxicity. Third, endosulfan in aqueous solution is adsorbed onto and desorbed from sediment particles in a rapid, effectively instantaneous two-way physical process. Finally, endosulfan in aqueous solution is exchanged with the atmosphere, in another two-way physical process with a time scale of hours to days, depending on the water depth (see Paper II). This air–water exchange is treated as a flux term rather than a transformation term in Eq. [1]. Figure 2 indicates schematically the relationships between the three toxic endosulfan species (, ß, and sulfate); the three phases in which endosulfan exists in the riverine environment (in air as vapor, and in water both in dissolved form and adsorbed form on sediment particles); and the four main transformations and exchange processes (oxidation, hydrolysis, particle–water exchange, and air–water exchange).

View larger version (27K): [in this window] [in a new window]   Fig. 2. Schematic diagram of the physical and chemical interactions included in the endosulfan chemistry model, with reaction rates expressed as time scales (order of magnitude only). Ca, Cw, and Cp are respectively the endosulfan concentrations in air, in aqueous solution, and adsorbed on particles in the river (all in kg m-3).

The transfer of endosulfan between the three phases (vapor in air, dissolved in water, and adsorbed on particles) occurs by reversible exchange processes described by equilibrium partition coefficients. First, the equilibrium between the dissolved and adsorbed phases is specified by the particle–water partition coefficient K^s_p (with units m³ per kg particles), defined for each species as the ratio of r_p to C_w at equilibrium:

[2]

Second, the equilibrium between the dissolved and air concentrations is specified by the dimensionless water–air partition coefficient:

[3]

This is the thermodynamic equilibrium ratio of water to air concentrations in a closed vessel containing water, air, and endosulfan of species s. It is of order 10³ for -endosulfan, 10⁴ for ß-endosulfan, and 10⁵ for endosulfan sulfate, decreasing roughly by a factor of two for each 6°C increase in temperature (see Table 1 and Fig. 3a, with further details in Appendix A). Because A^s is generally around 10³ or larger, the equilibrium concentration in water is at least 1000 times larger than the concentration in air.

View larger version (24K): [in this window] [in a new window]   Fig. 3. Temperature dependence of the (a) water–air partition coefficient and (b) saturation vapor pressure of endosulfan , endosulfan ß, and endosulfan sulfate.

[4]

[5]

for dissolved and adsorbed endosulfan species, respectively. Subscripts and superscripts on the flux and transformation terms denote the phase and species, as for the concentrations.

Fluxes through the Spray Drift Pathway
During a spraying operation, some spray drifts off-target. The deposition of this drift onto a downwind receptor point depends on the extent of the sprayed area, the release height of the point source, the droplet size distribution, the meteorological conditions (wind speed, wind direction, and atmospheric stability), and the depositional characteristics of the receptor surface. If the sprayed area is a uniform field, extensive in the cross-wind direction, then the deposition on any downwind receptor point (at a distance x from the downwind edge of the field) can be expressed as a drift deposition fraction, f_drift(x), a dimensionless quantity equal to the deposition D(x) on the receptor at the downwind point x (kg m^-2 [receptor]) divided by the intended deposition or nominal dose D₀ over the sprayed field (kg m^-2 [sprayed field]). Paper II describes a model for f_drift(x) in terms of the above independent variables, based on (i) a settling Gaussian-puff algorithm for spray dispersion and (ii) a single-layer description of particle deposition to the underlying surface, which incorporates gravitational settling, impaction, and diffusive transport. The model has been tested against independent data sets, with adequate results (Paper II). A predicted value of f_drift for conditions typical of spraying operations in New South Wales cotton fields (a median droplet size of 80 µm, a wind speed of 4 m s^-1, neutral atmospheric stability, and a receptor point at r₀ = 500 m downwind of a large sprayed field) is f_drift = 0.03. Values for other conditions are shown graphically in Paper II.

The present requirement is to estimate the average spray drift into a reach of river passing through a cotton-growing area, in which the spraying pattern and meteorological conditions (especially wind direction) are highly variable in both space and time. Information about the specific details of spray applications in space and time is not available, but average characteristics of spray operations at a regional level (dose per spray, extent and area fraction of sprayed fields, mean time between sprays) are known reasonably well. Therefore, we use a statistical approach set out in detail in Appendix B. The key result is this: Suppose that sprayed fields occur in a homogeneous random pattern covering an area fraction f_area of a region, except in a buffer zone of radius r₀ around a receptor, where no spraying occurs. If a spray dose D₀ (kg m^-2) is applied to each sprayed field, the average deposition to the receptor is the same as the deposition at the setback distance r₀ from a large uniform field subject to a uniform "steady drizzle" of spray at the spatially averaged dose D₀f_area. This holds for any distribution of wind directions. The averaging can either be in time or in space; in the spatial case, the receptor is regarded as a slug of water moving along a river through a cotton-growing area in which frequent (though randomly distributed) spraying occurs outside of a setback distance r₀.

This "steady-drizzle" idealization is not perfectly met, since the buffer zone around the river is a meandering strip rather than circular. However, the only practical consequence is that the effective value of r₀ is a little larger than the half-width of the strip. Neglecting this discrepancy, the average flux into the river through the spray drift pathway during a time interval T is estimated from the steady-drizzle idealization to be:

[6]

where D^s₀ is the dose of endosulfan species s on each sprayed field (kg m^-2 per spray), N is the mean number of sprays per field during the time interval t, T_spray = t/N is the mean time interval between sprays on any one field, and f_drift(r₀) is the drift deposition fraction at a setback distance r₀ from a large uniform field.

The model for f_drift in Paper II does not account for spray droplet evaporation. However, the usual method of spray application in New South Wales cotton-growing areas is ULV (ultra-low-volume) aerial application of a mixture of endosulfan and an inert oil-based adjuvant with a low evaporation rate in a ratio of 1:3, in droplets with a median diameter around 80 µm. Both constituents of this mixture have low vapor pressures, especially the endosulfan itself (Table A1), so neglect of spray droplet evaporation is a reasonable approximation. Such droplet evaporation as does occur will transfer endosulfan from the spray drift to the vapor pathway.

[7]

where V^s_d is the deposition velocity (m s^-1) for endosulfan vapor of species s to a water surface. Vapor transport is therefore driven by the difference between the concentrations in air and water (weighted by the water–air partition coefficient A^s) and is a bidirectional process: endosulfan dissolves from air into water when the air concentration (C_a) is high and revolatilizes back to air when C_a is low. The response time for this process is A^sH/V^s_d, where H is the water depth [this characterizes the response of C^s_w to a change in C^s_a under the equation C^s_w/t = F^s_w/H].

In this work, post-spray volatilization is described by a simple, exponentially decaying release rate constrained to conserve mass, with a time constant set by observations reported in Paper II. Dispersion is modeled with a Gaussian-plume algorithm (see Paper II). Together, these determine C^s_a. Deposition to water is then determined by the Deacon model (Paper II), which determines V^s_d in Eq. [7]. Model results and observations given in Paper II show that the period of downward air-to-water flux typically extends for a day or two after spraying, and that typical values of the total air concentration C_a during this period, at a distance of order 500 m from a sprayed field, are around 0.3 to 1 µg m^-3 (though higher values are observed during stable conditions at night).

Flux through the Dust Transport Pathway
The dust transport pathway operates by the windblown movement of endosulfan-bearing dust. Mechanisms for on-farm dust generation include dust uplift during wind erosion events, uplift by vehicular traffic on unpaved roads, and uplift by agricultural operations. The dust flux of endosulfan to a receptor is determined by the dust deposition flux F_dust (kg dust m^-2) and the mass fraction r^s_p of endosulfan on the dust (kg endosulfan per kg dust). From field measurements on New South Wales cotton farms, Leys et al. (1998) found r_p 1 x 10^-6 for road dust uplifted by vehicles, and r_p 1.8 x 10^-6 for dust uplifted from fields by cultivation (summed for all species). For dust uplifted by regional-scale wind erosion, r_p would be lower because of dilution by uncontaminated dust. The regional dust deposition rate F_dust, at distances of more than a few hundred meters from a cotton farm, was measured by Leys (unpublished data, 1997) to be around 2 g dust m^-2 mo^-1, incorporating dust from all sources arriving via dry deposition. Higher dust deposition rates are observed close to sources on-farm, but this short-range dust transport relocates endosulfan on-farm rather than moving it over sufficient distances to reach the riverine environment. To make an upper-limit estimate of the dust transport, we assume that all of the monthly regional dust deposition occurs in a single event in each month, and that all of it is contaminated at an average endosulfan mass fraction of 1.5 x 10^-6. The implied endosulfan input to the river for this event is 3 x 10^-9 kg m^-2, nearly a factor of 1000 lower than the typical endosulfan deposition from a single spray drift event, estimated below as 2.16 x 10^-6 kg m^-2. Hence, dust transport is not a significant pathway.

Flux through the Runoff Pathway
The runoff pathway as defined here is actually a collection of pathways involving the transport of endosulfan in water by overland or stream flow, either in dissolved or particle-bound forms. These are grouped together for the present purpose. The aim is not to construct a full model of the runoff pathway, but simply to use mass balance considerations to identify its order-of-magnitude properties for comparison with the airborne pathways.

Mass Balance
Consider a river of depth H and width Y, with flow velocity U, containing a contaminant with mean concentration C (kg m^-3). The discharge from tributaries into a length X of river is Q_trib (m³ s^-1), carrying contaminant at concentration C_trib. In the limit X 0, the contaminant mass balance is:

[8]

where F_air (kg m^-2 s^-1) is the net flux into the river across the air–water interface due to all airborne pathways, and S (kg m^-3 s^-1) is the net source strength. The quantity Q_trib/X is the discharge from tributaries per unit length of river. Equation [8] is of the form of Eq. [1], [4], and [5], with the runoff flux explicitly identified as the last term on the right hand side. Hence, the runoff flux density from the tributary discharge Q_trib/X per unit length of river is:

[9]

for dissolved species, as in Eq. [4]. For compatibility with other flux densities, this is expressed as a flux density per unit surface area of river. A similar expression determines the flux density F^s_p(runoff) for adsorbed species in Eq. [5]. The behavior of the adsorbed species also depends on the suspended sediment concentration _p, which obeys:

[10]

This is a conservation equation similar to Eq. [8], in which F_ent is the net sediment entrainment from the river bottom and sides. We do not parameterize this term.

In the case of a single-point junction between a river (inflow Q₀, outflow Q₁) and a tributary (inflow Q_trib), the water mass balance is Q₁ = Q₀ + Q_trib and the contaminant mass balance is C₁Q₁ = C₀Q₀ + C_tribQ_trib. [The latter equation neglects storage changes at the junction and contributions from airborne fluxes and transformation processes, so (CV)/t = 0, F_air = 0 and S = 0.] Defining the dilution factor f_dil = Q_trib/Q₁, the contaminant mass balance becomes:

[11]

If a tributary with dilution factor f_dil enters a river, then Eq. [9] shows that the average flux density (normalized per unit surface area of river over length X of river) is:

[12]

Scenario for the Runoff Pathway
From Eq. [12], the flux through the runoff pathway depends on the concentration difference between runoff and river water and the amount of runoff relative to river discharge. A third factor, the frequency of runoff events, determines when Eq. [12] is applicable. We now estimate each of these factors for the region studied here.

The concentrations in runoff water are usually much higher than those in the river and hence approximately equal to the runoff-river concentration difference. Endosulfan concentrations in tailwater dams can be as high as 50 µg L^-1 (M. Silburn and D. Connelly, personal communication, 1997). However, values around 2 to 10 µg L^-1 are more typical of concentrations measured during flood events in off-farm waterways and small creeks (Cooper, 1996; Muschal, 1997, 1998). The difference between these figures implies an attrition of endosulfan in overland flow, caused both by adsorption and sedimentation.

The amount of local runoff is influenced by on-farm retention of water and local topography. Retention systems to recycle water from irrigation and rainfall are used on many New South Wales cotton farms, so overflows into local rivers occur only when runoff exceeds the capacity of on-farm storage. O'Brien (1996) estimated that 69% of Upper Namoi growers and 97% of Lower Namoi growers can contain 25 mm of rain on-farm (1995–1996 data). Local topography has also been influenced by silt accumulation over geological time on the Lower Namoi floodplain and similar systems, causing a substantial fraction of runoff to flow away from the river and lessening the amount that would otherwise reach the river.

To examine the frequency (as well as the amount) of runoff, a simple soil water balance model for a cotton field has been constructed using available daily rainfall and pan evaporation data. The model predicts the runoff, evapotranspiration, and irrigation terms in the water balance. Details of the model formulation (which is quite conventional) are given in Appendix C, and sample results are shown in Table 2 and Fig. 4. The predicted mean annual runoff is around 100 mm yr^-1 in the lower Namoi Valley, occurring in only a few events per year (Fig. 4). With 15 mm of on-farm water storage there are typically three or four events per year. The results are not strongly sensitive to key parameters in the water balance model (Table 2). In particular, an on-farm water storage of zero has little effect on the average annual runoff, although it does increase the number of runoff events to about 10 per year. It is concluded that the probable interval between major off-farm runoff events is of the order of 1 to 3 mo, and not more than one or two such events are likely to occur in any one growing season (November to January).

View larger version (27K): [in this window] [in a new window]   Fig. 4. Estimated local surface runoff at Myall Vale for a sample period (1988 to 1991) using a simple water-balance model (Appendix C), observed precipitation and evaporation data, and predicted irrigation scheduling based on modeled root-zone soil moisture storage. Model parameters are for the central case (Appendix C).

Transformation Processes
The main transformation processes are oxidation, hydrolysis, and particle water exchange (recalling that air–water exchange is treated as a flux). Oxidation transforms the dissolved and ß isomers into sulfate, and so is a sink for and ß and a source for sulfate. Hydrolysis acts as a sink by transforming dissolved and ß into the much less toxic endosulfan diol. Adsorption to sediments is a sink for dissolved endosulfan of all three species and a corresponding source in the adsorbed phase. Hence, the transformation terms for the dissolved species (, ß, and sulfate) are:

[13]

where k_p and k^ß_o are the oxidation rate constants (time^-1) for the conversion of endosulfan and ß to sulfate; k_h, k^ß_h, and k_h are the hydrolysis rate constants for the conversion of endosulfan , ß, and sulfate to diol; and E_p, E^ß_p, E_p are the transformation terms for particle–water exchange, which are sinks for dissolved species and sources for adsorbed species. These terms take the form:

[14]

where K^s_p is the particle–water partition coefficient defined in Eq. [2], and k_p is the equilibration rate constant for particle–water exchange. For adsorbed endosulfan of all species (s = , ß, or ), the only transformation terms considered are for particle–water exchange, so:

[15]

We neglect oxidation and hydrolysis in the adsorbed phase, a reasonable approximation given the rapidity of particle–water exchange and the dominance of the dissolved phase in a typical river.

The values adopted for all partition coefficients and rate constants are summarized in Table 1, based on detail given in Appendix A. The sensitivities of the model to these values are explored in the Results and Discussion section. We note that the rate constants for oxidation and hydrolysis in the natural environment are very uncertain, depending particularly on the presence of biotic material (Peterson and Batley, 1991; Kennedy et al., 1998). In particular, the oxidation time constants (the inverses of the rate constants k_o, k^ß_o) are likely to be very long (100 d or more) in the absence of biotic material, but may be much shorter (perhaps as low as a few days) when biotic material is present. Here, 100 d is assumed.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX A: Physical and... APPENDIX B: The "Steady-Drizzle"... APPENDIX C: Soil Water... REFERENCES

 
Model Specification
The modeling framework has been used to simulate endosulfan concentrations in two riverine environments where irrigated cotton is a major land use: Pian Creek and the Namoi River in the Lower Namoi Valley near Narrabri, NSW (Fig. 1). These rivers differ in depth and flow speed: Pian Creek is shallower and slower-flowing than the Namoi River (Table 3). The simulations are based on the following assumptions:

(ii) As justified in Appendix B, the spray-drift flux into the river is modeled with the "steady-drizzle" idealization using Eq. [6]. Survey data (B. Peasley, personal communication, 1996) show that cotton in the study area is typically grown on 50% of land in a strip extending 5 km on either side of each river, except in a buffer strip of width 500 m on either side of the river (Fig. 5). Hence in Eq. [6] we take f_area = 0.5 and r₀ = 500 m. The interval between sprays is taken to be T_spray = 13 d, a typical value for the area. These choices imply that an average of four sprays per day occur in each spatial computational cell, of length X = 10 km. The spray dose is D₀ = 0.72 x 10^-4 kg m^-2 of endosulfan (partitioned 2:1:0 between , ß, and sulfate). The drift deposition fraction at the typical setback distance r₀ is f_drift(r₀) = 0.03, based on model results (Paper II) for particles of median diameter 80 µm. These values imply an endosulfan deposition from a single spray, D(x) = f_drift(x)D₀, of 2.16 x 10^-6 kg m^-2 on a receptor at x = 500 m.

View larger version (31K): [in this window] [in a new window]   Fig. 5. Source and receptor geometry for integrated modeling of each 10 km stretch of river (of a total of 100 km). On either side of the river is a 500-m buffer zone, beyond which are 50 cotton fields of area 1 km2 (25 on each side of the river) alternating with unsprayed fields.

(iv) For dust transport, we assume a steady dust flux into the river of 2 g m^-2 mo^-1, carrying an endosulfan mass fraction r_p = 1.5 x 10^-6, based on the measurements of Leys et al. (1998).

(v) Because of the infrequency of runoff, with only one or two events likely during a spraying season, we examine the effects of a single large runoff event in the 40-d simulation period. The event is assumed to occur from tributary inflow lasting for 2 d (Days 8 and 9 of the 40-d simulation) centred at X = 35 km (implying that the inflow Q_trib/X is uniform between 30 km and 40 km, because of the step length of 10 km). Following the previous section, the endosulfan concentration in the tributary is taken to be C_trib = 5 µg L^-1 and the dilution factor f_dil is assumed to be 0.05 for the Namoi River and 0.2 for the slower-flowing Pian Creek.

(vi) The initial and boundary conditions are that the river starts clean and is clean upstream, so C = 0 at t = 0 for all X, and C = 0 at X = 0 for all t.

The model consists of seven coupled ordinary differential equations, for the water concentrations C_w, C^ß_w, and C_w (Eq. [4]), particle-bound concentrations C_p, C^ß_p, and C_p (Eq. [5]) and sediment concentration _p (Eq. [10]), together with specifications for the transformation terms and the flux terms F(spray), F(vapor), F(dust), and F(runoff). The equations are solved numerically by the Runge–Kutta method. Table 3 summarizes all parameter choices. Sensitivities to these choices are explored below.

Qualitative Behavior of the System
The behavior of the system, determined by the characteristic time scales for the major physical and chemical reactions, can be discerned qualitatively from the form of the equations. For air–water exchange the time scale is A^sH/V^s_d, around 10 h when the water depth H = 0.1 m and 100 h when H = 1 m. The adsorbed and water concentrations equilibrate much more rapidly, in a time of order 1 h. The chemical source–sink time scales are around 100 h for hydrolysis, 1 h for adsorption to colloids, and uncertain for oxidation (100 d is assumed). When chemical decay is a much slower process than air–water exchange (true for shallow water bodies of depth around 0.1 m or less), the basic behaviour of the system is that C^s_w tracks and slightly lags A^sC^s_a (around 10³ C^s_a for the isomer). Deposition from air to water occurs when C^s_w < A^sC^s_a, and revolatilization from water to air when C^s_w > A^sC^s_a.

Numerical Results
Figures 6a and b show (for the Namoi River and Pian Creek, respectively) the variation of the total concentration C (= C_w + C_p) with t, at X = 10, 40, and 100 km. Likewise, Fig. 7a and b show the variation of C with X, at t = 2, 10, and 40 d. In the upper panels, each figure also shows an indicator of the endosulfan species composition, the fraction of and ß in the dissolved endosulfan /C_w.

View larger version (28K): [in this window] [in a new window]   Fig. 6. Modeled variation over time of + ß fraction and total riverine endosulfan concentration, integrating all transport pathways (spray + vapor + dust + runoff), for the (a) Namoi River and (b) Pian Creek, using model parameters in Table 3. Time traces are given for X = 10, 40, and 100 km downstream. A major runoff input occurs on Days 8 and 9 at X = 35 km.

View larger version (26K): [in this window] [in a new window]   Fig. 7. Modeled variation with downstream distance of + ß fraction and total riverine endosulfan concentration, integrating all transport pathways (spray + vapor + dust + runoff), for the (a) Namoi River and (b) Pian Creek, using model parameters in Table 3. Plots with distance are given for times t = 2, 10, and 40 d after the start of spraying. A major runoff input occurs on Days 8 and 9 at X = 35 km. In (a) the 2-d line is beneath the 40-d line.

The steady background concentrations in Fig. 6 are the result of airborne transport. These pathways cause the riverine concentration to rise within a few days to an equilibrium level at which the inputs from airborne transport pathways are balanced by losses, mainly from volatilization, hydrolysis, and advection (downstream transport in the river). The background concentrations appear as steady because of our simplifying "steady-drizzle" idealization for the airborne fluxes into the river, and in reality would be subject to short-term variation around the mean levels predicted here. However, Fig. 6 captures a realistic picture at a spatial resolution of 10 km (the computational step length X) because airborne transport events are frequent in both space and time.

Figure 6 also shows that the slug of water injected by runoff has a much lower + ß fraction than the water in the rest of the river, because the endosulfan composition in the runoff water is strongly weighted toward sulfate, whereas the airborne pathways only transport and ß endosulfan into the river.

A similar picture emerges from Fig. 7, showing the spatial variation of C at 2, 10, and 40 d. At 10 d, just after the end of the runoff event, there is a large peak in C just downstream of the runoff entry point at X = 35 km. By 40 d the contaminated water introduced by the runoff event has long since flowed out of the region, leaving a background concentration due to airborne transport, which rises slowly with X as a result of continuous inputs from airborne transport along the river. Equilibrium (C steady in X) is not reached over the 100-km length of river simulated here, because the time required for the physical and chemical processes to reach equilibrium under steady inputs into a moving slug of water is several days, which is substantially longer than the travel time of the slug through the 100-km domain.

For the Namoi River and Pian Creek, the roles of different transport pathways are shown in Fig. 8, which breaks the total concentration C (at X = 50 km) into contributions from spray drift, vapor transport, dust transport, and runoff. From the leftmost frames of Fig. 8 (at t = 10 d, just after the runoff event), it is evident that runoff is the dominant transport pathway on the infrequent occasions when significant runoff occurs. However, most of the time the situation is more like the rightmost frames (at t = 40 d), where the entire riverine concentration is due to airborne transport. Of the three airborne pathways, dust transport is negligible but vapor transport and spray drift are of the same order of magnitude.

View larger version (35K): [in this window] [in a new window]   Fig. 8. Modeled contributions to the total riverine endosulfan concentration at X = 50 km by each contributing transport pathway for the Namoi River and Pian Creek at t = 10 d and t = 40 d.

Though we have not treated sediment–water exchange explicitly, it is possible to assess its qualitative effect. Since this process introduces no new endosulfan but shifts endosulfan reversibly between pools in the water column and the bottom sediment, its effect is to smooth the temporal peaks due to runoff events in Fig. 6. Likewise, it delays the return of riverine concentrations to near-zero levels at the end of the spraying season as the river is flushed with clean water. Sediment–water exchange may also be involved with downstream transport of endosulfan attached to sediment particles, but the above consideration of the ratio C_p/C for suspended (not bottom) sediment suggests that this riverine flux is small compared with the flux of endosulfan in dissolved form, except during floods that transport significant quantities of sediment. Such events are rare, because of low bed slopes and the infrequency of surface runoff.

Figure 9 shows a section of the data for endosulfan concentrations recorded by the New South Wales Department of Land and Water Conservation under its Central and North West Regions Water Quality Program (Cooper 1996; Muschal, 1997, 1998). The data shown are for the Namoi River at Bugilbone and Pian Creek at Rossmore, stations representative of the conditions for present simulations. These data are consistent with the model predictions in two respects. First, the overall levels during the spraying season are observed to be about 0.05 µg L^-1 (Namoi) and 0.1 µg L^-1 (Pian Creek), values similar to the concentrations predicted in Fig. 8 to arise from airborne fluxes. Second, the observed concentrations have a regular, steady character through the spraying season (with some variability) rather than a spiky, intermittent character, consistent with the conclusion that airborne pathways rather than runoff dominate the endosulfan inputs to the river except when infrequent runoff events occur.

View larger version (15K): [in this window] [in a new window]   Fig. 9. Measurements of total ( + ß + sulfate) waterborne endosulfan concentration for the Namoi River at Bugilbone and Pian Creek at Rossmore, September 1991 to March 1997, by the New South Wales Department of Land and Water Conservation (Muschal, 1998).

[16]

This is independent of the fractional change P for small P, as shown by Taylor expansion of Q. Sensitivities of +0.5 and -0.5 (for example) mean that a 10% increase in the parameter P causes a 5% increase and a 5% decrease in Q, respectively. Our choice for the model output Q (among many possibilities) is the total riverine concentration C, at two locations (X,t) that respectively maximize and minimize the contribution of runoff to the total endosulfan load. In the first all pathways case, Q is C(X,t) at X = 40 km and t = 10 d, just after and downstream of the simulated runoff event, so the contribution of runoff is large (see Fig. 6 and 7). In the second air pathways only case, Q is C(X,t) at X = 100 km and t = 40 d, and runoff makes a negligible contribution to C. For all significant parameters, we then calculate the sensitivity defined by Eq. [16], using a 10% perturbation in the parameter. The only exception is river water temperature T, where a perturbation of 1 K is used and the resulting sensitivity is (Q/Q)/T, with units K^-1.

Table 4 shows the sensitivities for Pian Creek and the Namoi River, with the central case (about which perturbations are made) defined by Table 3. To assist in interpreting the sensitivities, Table 3 also includes a rough, necessarily subjective assessment of the uncertainty or variability in each parameter. Several features emerge from Table 4: first, the sensitivity of total concentration C to individual parameters depends on the contribution of the process described by that parameter to C. Thus, in the all pathways case, parameters controlling the runoff pathway, such as f_dil and C_w(trib), have sensitivities close to 1 and the sensitivities for parameters controlling airborne pathways are small, whereas in the air pathways only case the situation is reversed and major pathway parameters are those controlling spray drift (such as D₀ and f_drift) and vapor transport (such as C_air and V_d). Second, sensitivity to river flow speed is negative: greater flow speeds lead to lower concentrations. Third, sensitivity to water temperature is also negative: higher temperatures lead to lower C values, by favoring revolatilization from the river water back into the air. Fourth, the most important chemical properties of endosulfan are the oxidation rate constants describing the conversion of and ß to sulfate, because this process is the major riverine sink for endosulfan along with revolatilization back to the air (mainly a sink for endosulfan because of its relatively low water–air partition coefficient, as shown in Table 1). Finally, sensitivities to parameters apart from those just discussed are low.

View this table: [in this window] [in a new window]   Table 4. Sensitivities of total riverine endosulfan concentration C to small (10%) perturbations in parameters. The sensitivity to a parameter P is defined as (C/C)/(P/P). In columns headed "all pathways", C is at X = 40 km and t = 10 d, just after and just downstream of the runoff event where the contribution of runoff to C is greatest. In columns headed "air pathways only", C is at X = 100 km and t = 40 d, where runoff makes a negligible contribution to C.

	SUMMARY AND CONCLUSIONS

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX A: Physical and... APPENDIX B: The "Steady-Drizzle"... APPENDIX C: Soil Water... REFERENCES

(i) Runoff pathway events are large and infrequent. The infrequency of runoff is shown by a water balance model (Appendix C). The large concentrations in these infrequent runoff events are indicated by Fig. 8.

(ii) Airborne pathway events are smaller in magnitude than runoff events but act quasi-continuously, resembling a "steady drizzle". The steady-drizzle idealization (Appendix B) is that, if sprayed fields occur in a homogeneous random pattern covering an area fraction f_area of a region, except in a circular buffer zone of radius r₀ around a receptor where no spraying occurs, then the average deposition to the receptor is the same as the deposition at the setback distance r₀ from a large uniform field subject to a uniform spray at the spatial average of the spray doses applied to the random array of fields. The average is across the ensemble of wind directions, which can be arbitrary. This idealization provides a good characterization for spray drift into a river passing through a cotton-growing area containing numerous sprayed fields.

(iii) Of the airborne pathways, spray drift and vapor transport are both significant but dust transport is negligible. This finding (Fig. 8) rests on the simple physical and mass-balance considerations embodied in the models for each pathway.

(iv) Most of the routinely observed riverine endosulfan is transported by airborne routes. This conclusion rests on modeling and observational evidence. First, modeled concentrations from airborne pathways are around 0.05 µg m^-3 for the Namoi River and 0.1 µg m^-3 for Pian Creek (Fig. 8), comparable with growing-season observations (Fig. 9). Therefore, airborne transport is sufficient to account for the quasi-continuous background concentrations observed in the data. Second, runoff events can cause large concentrations but they occur only infrequently (a few times per year, from Appendix C) and are flushed from the system relatively quickly (from Fig. 6, though sediment–water exchange provides a smoothing mechanism not accounted for in this figure).

The results of this work have implications for efforts to limit the transport of endosulfan into the riverine environment. First, and not surprisingly, they provide support for efforts to reduce spray drift and rain-induced runoff as laid down in current guidelines for best management practices. In particular, it is important to abide by all best management practice guidelines for avoiding catastrophic spray drift events such as oversprays of waterways or other sensitive areas, applications in inappropriate conditions when the wind is blowing directly from a sprayed field onto a nearby sensitive area, or applications in very stable conditions. Such spray events can lead to riverine concentrations large enough to cause fish kills, unlike vapor transport, which causes a slower but steadier leakage of endosulfan into the river.

Secondly, and more controversially, our results suggest that there is an irreducible minimum level of riverine contamination associated with vapor transport, ranging from less than 0.02 to up to 0.04 µg m^-3, depending on the depth and flow in the river and its proximity to sources. This cannot be significantly reduced with buffer strips or other practices used to reduce spray drift. The only means of significantly reducing vapor transport is to reduce the use of sprays through techniques such as integrated pest management and the selective use of genetically modified cotton resistant to insect attack.

	APPENDIX A: Physical and Chemical Properties of Endosulfan

TOP ABSTRACT INTRODUCTION THEORY RESULTS AND DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX A: Physical and... APPENDIX B: The "Steady-Drizzle"... APPENDIX C: Soil Water... REFERENCES

 
Saturation vapor pressure: The saturation vapor pressure P^s (with s = , ß, or ) depends on temperature according to the Clausius–Clapeyron equation P^s = 100 exp[c₁ - (c₂/T)], where P^s is in Pascals, T is absolute temperature, and the coefficients c₁ and c₂ are given in Table A1 (Hoechst Aktiengesellschaft, 1993a, b). The temperature dependence of P^s is shown in Fig. 3b.

Aqueous solubility: Based on the review of Suntio et al. (1988), covering data from several authors, the saturation aqueous solubility of endosulfan is taken to be 0.15 mg L^-1, identical for the , ß, and sulfate species. Temperature dependence is ignored in the absence of information.

Water–air partition coefficient A^s: From Eq. [3], A^s = _eq. In an equilibrium system in which endosulfan is saturated in both air and water, the