Pesticide extraction from soil into runoff in North American and Australian croplands

Conceptual framework

Silburn (2023) outlined a conceptual framework for pesticide runoff, which is summarised here. Transfer of chemicals from the soil into runoff during a rainfall event is the outcome of the following processes. During rainfall, chemicals leach downwards into the soil, depending on the infiltration amount and sorption properties of the chemical and the soil. Next, extraction from the soil into runoff occurs, in the water phase and by sediment entrainment. There is a positive correlation between water and sediment phase extraction (Sharpley 1985; Ahuja 1986). The amount of chemical available for runoff extraction and movement downwards depends on the concentration in the soil surface, in a layer termed the runoff-mixing layer. An ‘effective depth of interaction’ can be defined, which is only 2–3 mm thick (Ahuja 1986) but in practice 5–25 mm soil layers are sampled. The chemical distribution with depth in the soil surface layer is important but unknown. For the sediment-sorbed component, runoff concentrations of strongly sorbed chemicals are generally 1–10 times the soil concentrations (Leonard et al. 1979; Willis et al. 1983). This is because of size-selective erosion (enrichment) and differences in the concentrations in the depth of the soil sampled and the eroded depth. Pesticide concentrations extracted into runoff are also affected by the concentration profile within the soil surface layer (Baker 1980), which will change through time after application.

Leaching through the runoff-mixing layer continues throughout a rainfall event. Chemicals that are weakly sorbed have exponentially decreasing concentrations (C) with time (t), i.e. C(t) = C₀ exp[−(flow/volume)t] (Baker 1980; Ahuja and Lehman 1983). This equation can be formulated to account for sorption. Truman et al. (1998) confirmed the exponential decline in soil surface concentrations under rainfall experimentally for atrazine. Chemicals that are strongly sorbed to soil will be unaffected by this process.

Where leaching is restricted by moist, compacted subsoil, the soil concentration will be diluted into the rainfall volume and transferred into runoff rather than into the leachate. Runoff concentrations are then 1–2 orders of magnitude higher than where leaching is unrestricted (Ahuja and Lehman 1983). Snyder and Woolhiser (1985) observed significant exfiltration on sloping flumes, which increased dye removal in runoff; dissolved chemicals were contributed as ‘interflow’ [as explained by Loch et al. (1987)] as well as by surface extraction. In brief, interflow is water flowing through the soil surface layers, for example the tilled layer of a cultivated soil or the A horizon of a duplex soil, whereas overland flow is runoff flowing across the soil surface. Indeed, Barnett et al. (1972) found runoff concentrations were highest for all nutrients when interflow occurred rather than overland flow, on steep slopes. Wang et al. (2022) found that interflow was the main loss pathway for nitrate-nitrogen whereas the main loss pathway for more sorbed compounds (e.g. ammonium) was surface runoff. Edwards et al. (1980) observed lower glyphosate runoff concentrations from interflow than overland flow, due to topsoil ‘filtering’ the glyphosate (i.e. adsorption). Thus, chemicals need to be weakly sorbed to be transported in interflow.

In summary, the main processes that affect extraction of chemicals into runoff after application amount and dissipation are leaching and dilution of solutes (increasingly for more weakly sorbed pesticides) and sediment transport (increasingly for strongly sorbed pesticides). Both processes are affected by pesticide sorption and hydrology (including interflow) and the concentration profile in the surface soil.

Runoff extraction ratios

Baker (1980) and Leonard et al. (1979) give non-linear equations relating event runoff pesticide concentrations (C_RO μg L⁻¹) to soil concentrations (C′_SOIL mg kg⁻¹) for poorly sorbed and tightly sorbed chemicals (Silburn 2023) where the slope was E and the exponent P1. Leonard et al. (1979) conceptualised coefficient E as an ‘extraction coefficient’ for poorly sorbed pesticides and an ‘enrichment factor’ for sorbed pesticides. Exponent P1 represents non-linearity and was 1.2 (Leonard et al. 1979) or 1.03 (Baker 1980) for poorly sorbed pesticides, reflecting changing extraction efficiency and distribution in the surface soil with time of contact. Higher soil concentrations were related to more recent applications. The P1 was 0.83 for sorbed pesticides (Leonard et al. 1979), indicating a changing sediment composition with sediment concentration, consistent with size-selective deposition and more clay and organic matter in finer sediment. Silburn (2023) found soil and runoff concentrations were linearly related, because they did not include longer times after application (>50 days); these data are suitable for the highest risk period. Assuming linearity, the runoff extraction (E_RO) follows:

The maximum potential runoff concentration (Silburn 2023) is where the runoff concentration is equal to the pesticide mass in the mixing-layer divided by the rainfall volume. For sediment sorbed chemicals, maximum potential runoff concentration (or no-enrichment runoff concentration) equals the soil concentration multiplied by the sediment concentration. If sediment enrichment occurs, pesticide runoff concentration is greater than this value; if the pesticide is desorbed from the soil/sediment, it is less than this value.

Experimental outline

Data were used for soil pesticide concentrations before rain and event mean concentrations in runoff for Australian rainfall simulation studies (Silburn 2023) and North American rainfall simulator and catchment studies. Properties of the pesticides studied are given in Table 1.

Silburn (2023) studied 12 pesticides with a wide range of properties, from dimethoate [solubility ~40,000 mg L⁻¹, soil sorption coefficient (K_OC) 20 L kg⁻¹] and pyrithiobac sodium (solubility ~700,000 mg L⁻¹, K_OC 9–21 L kg⁻¹) to DDE with a low water solubility (0.003 mg L⁻¹) and high K_OC (380,000–880,000 L kg⁻¹) (Table 1). The Australian rainfall simulation studies (Silburn 2023) were at Gatton (Lockyer Valley), Emerald (Central Highlands) and Jondaryan on the Darling Downs (Table 2). The Gatton soil has a dark clay loam to light clay (crusting or cloddy) surface, is a Black Dermosol (Isbell and National Committee on Soil and Terrain 2021) used for cropping. The Emerald site was on an irrigated cotton farm west of Emerald, Queensland, on a Black Vertosol (Isbell and National Committee on Soil and Terrain 2021) and is strongly self-mulching and cracking. The site was used to grow irrigated cotton for >20 years. The Jondaryan site is a gently undulating alluvial plain of Oakey Creek. The soil is a Haplic self-mulching Black Vertosol (Isbell and National Committee on Soil and Terrain 2021) used for growing irrigated cotton and winter cereals. Studies occurred on row-crop layouts with 1 m wide rows except at Gatton (0.75 m wide rows), hills were 0.25 m high, with linear 50% side-slopes 0.4 m long. The furrows were used for irrigation and wheel traffic. Downfield slopes were 0.2–1.5%. Studies occurred early in the cotton season between planting and the first irrigation. All sites had low cover (<5%), except where cover was applied at Emerald as cotton trash (6.8 t ha⁻¹ biomass with 47–56% cover) or wheat stubble (1.8 t ha⁻¹ biomass with 15–25% cover or 2.8 t ha⁻¹ biomass with 30–35% cover). Surface soils (0–50 mm) were loose and at air-dry moisture content; the surface had a strong crust at Jondaryan. There were no cracks as the sites had been pre-irrigated or fallowed. All sites had several centimetres of loose soil in the furrow over firm moist subsoil, compacted to varying degrees. Gatton had no crop and wheel traffic was random. The Emerald and Jondaryan sites had a wheel track and non-wheel track as separate plots under the simulator.

The North American studies involved 13 herbicides for rainfed studies and 11 herbicides and insecticides for irrigated cropping (Table 3). Table 3 includes the pesticides studied, descriptions of the soils, slopes and how soil concentrations were obtained, and whether simulated rainfall was used. Pesticides had a wide range in properties and over four orders of magnitude range in soil concentrations. Sorption varied from low for atrazine and propachlor (K_OC ~ 100 L kg⁻¹), to intermediate for glyphosate (19,600 L kg⁻¹) and high for paraquat (1,000,000 L kg⁻¹) (Hornsby et al. 1996). Soil concentrations were measured for depths of 5–150 mm. Availability of soil concentrations was the main limitation determining which studies were used here. Further descriptions are provided with the results.

Statistical analysis

Initial data analyses demonstrated the data contained heterogeneity of variances and non-normal distributions of multiple pesticides (Shapiro Wilk test) and consequently a base-10 logarithm transformation was required to achieve normality and homogeneity. The pesticide dataset was divided into slope classes: level to very gently inclined (0–3%), gently inclined (3–10%) and moderately inclined (10–32%). Following this, one-way analysis of variance (ANOVA) test for differences between group means was undertaken. If significant F-tests were obtained from ANOVA at P < 0.01 then a means comparisons report for all pairs, Tukey HSD test and connecting letters report were produced. Pesticides not connected by the same letter significantly differ. Statistical analysis was conducted using JMP®, Ver. 15.1.0. (SAS Institute Inc., Cary, NC, 1989–2021).

All data that were normally distributed and had equal variances for level to very gently inclined (0–3%) slope were pooled if a significant F-test was obtained from ANOVAs at P < 0.01, then a means comparisons report, Tukey HSD test and connecting letters report were produced. Pesticides not connected by the same letter significantly differ. The pesticides atrazine, dimethoate, diphenamid, glyphosate and metolachlor were not included due to non-normality, while 2,4-D was normally distributed but failed the equal variance criteria.

Statistical comparisons

Runoff extraction was not significantly different for 15 pesticides but was significantly lower for trifluralin (Table 4). Runoff extraction ratios were not significantly different for total, alpha, beta and sulphate-endosulfan for non-transformed analysis. Thus, runoff extraction for endosulfan was not significantly different. Metolachlor was less sorbed (K_oc = 200) than the other pesticides, which explains why it had lower runoff extraction. Trifluralin is tightly sorbed (K_oc = 15,800 c.f. <600 for most herbicides) but it is also volatile and was applied <50 days prior to simulated rainfall. Volatilisation would have led to lower concentrations in the surface of the runoff-mixing layer, leading to lower runoff extraction. Moderate slopes did not cause significantly greater runoff extraction for atrazine, but runoff extractions of glyphosate and metolachlor were significantly lower (Table 4). As shown in Table 4, the runoff extraction ratios for North American catchments (Leonard et al. 1979) were not significantly greater than those for the Australian rainfall simulator plots. One exception was trifluralin runoff extraction for the rainfall simulator plots, which was significantly lower.

Log-transformed E_RO values from the cyanazine data of Leonard et al. (1979) and the Australian rainfall simulator plots (Emerald, Gatton and Darling Downs; Silburn 2023) are shown in Table 5. The E_RO was significantly lower for trifluralin, pendimethalin and DDE (all more tightly sorbed compounds) from the rainfall simulator studies, but was was significantly higher for the cyanazine data from Leonard et al. (1979). The E_RO values for beta, total and sulfate endosulfan were somewhat lower. However, E_RO values for most of the pesticides from the rainfall simulator data were not significantly different, once the more sorbed trifluralin, pendimethalin and DDE were excluded. This lack of difference in E_RO values for many pesticides is consistent with the analysis in Table 4.

Leonard et al. (1979)

Leonard et al. (1979) had data from four small cropping terraced catchments in Georgia USA (~3% slope, sandy loam soils), over 4 years of annual herbicide applications to summer crops (data from Smith et al. 1978). They measured runoff, sediment and herbicide concentrations in runoff from the catchments. The annual average sediment concentration was 5.6 g L⁻¹ (range 1.8–8.8 g L⁻¹). They investigated relationships between concentrations in runoff and in the surface soil (0–10 mm depth). For four herbicides of similar low sorption (atrazine, cyanazine, diphenamid and propazine, K_OC 100–210), dominantly transported in the water phase, they found a strong relationship (R = 0.93), with no difference due to herbicide, watershed or year (Fig. 1). The rainfall simulator data of Silburn (2023) coincided with these data over much of the range (Fig. 1) and runoff extraction of the herbicides and most other pesticides were not significantly different (Table 4).

Fig. 1.

Pesticide soil and runoff concentrations from rainfall simulator plots (C_ro = 30 × C′soil; Silburn 2023) compared with herbicides (Leonard et al. 1979) from catchments (C_ro = 50 × C′soil^1.2), and potential maximum concentrations for 50 mm rainfall. Rainfall simulator sites: DD, Jondaryan; EM, Emerald, Gatton; Py_Na, pyrithiobac sodium.

Leonard et al. (1979) derived the relationship between runoff concentrations C_RO (μg L⁻¹) and 0–10 mm soil C′_SOIL (mg kg⁻¹). The data extended over four orders of magnitude and was significantly non-linear. This indicated that extraction was more efficient at higher soil concentrations and less efficient with longer time of contact with soil. This relationship and that of Silburn (2023) coincide at a soil concentration of around 0.1 mg kg⁻¹ (Fig. 1). The catchment runoff concentrations were twice those of Silburn (2023) at a soil concentration of 5 mg kg⁻¹. Higher concentrations were associated with runoff soon after spraying and presumably higher concentrations in the top few millimetres of soil. For the simulator data, lower soil concentrations were associated with pesticides with low applications rates (e.g. pyrithiobac sodium and organophosphates). Leonard et al. (1979) considered the non-linearity reflected less efficient extraction with time after spraying, and ‘inadequacy in using the 0–10 mm zone as a reference throughout the growing season’. As time progresses, 0–10 mm soil over-estimates surface concentrations extracted into runoff, with concentrations lower at the soil surface and increasing with depth. Leonard et al. (1979) provides some of the few data on this issue. At 13 days and 20 mm of rain after spraying, atrazine concentrations were lower in 0–10 mm than 10–20 mm soil.

The Leonard et al. (1979) and rainfall simulator data were about an order of magnitude lower than the maximum potential runoff concentrations (Fig. 1), which had extraction ratios of 300 and 500, for 0–10 and 0–25 mm soil depth, respectively (for 50 mm of rain). Thus, both natural and simulated rainfall extracted <10% of the pesticide mass from these soil layers, consistent with observed pesticide losses as a percent of applied amounts (Wauchope 1978) and with a runoff-mixing layer shallower than 10 mm (Baker 1980; Ahuja 1986). Potential runoff concentrations correspond to observed runoff with complete extraction of a soil depth of <2 mm (Fig. 1).

Leonard et al. (1979) also studied paraquat, a compound highly sorbed to clay surfaces and dominantly transported in sediment. They found a good relationship between runoff concentrations and 0–10 mm soil concentrations multiplied by sediment concentration (SC) (Fig. 2): C_RO (μg L⁻¹) = 1000 × 2.16 × [SC(g L⁻¹) × C′_SOIL(mg/kg)/1000]^0.83 (R = 0.97). Multiplying soil concentrations by sediment concentrations gives an estimate of runoff concentrations where the sediment mass was eroded with a concentration equal to the soil concentration, i.e. with no enrichment or desorption (‘no-enrichment’ line in Fig. 2). Paraquat concentrations in sediments were greater than those measured in the soil. Leonard et al. (1979) identified the coefficient 2.16 as an enrichment factor, due to preferential transport of finer sediments. Similar enrichment ratios for clay, organic matter and sediment specific surface compared to catchment soil were also found. The non-linearity results in an enrichment ratio of 7.0 at low values of SC × C′_SOIL and declining to 1.5 at the highest values of SC × C′_SOIL (Fig. 2).

Fig. 2.

Pesticide runoff concentrations related to soil concentration times sediment concentration (SC), from rainfall simulator plots (Silburn 2023), and catchments for paraquat (P1–4, Leonard et al. (1979), and enrichment ratios for paraquat. Rainfall simulator sites: DD, Jondaryan; EM, Emerald, and Gatton; Py_Na, pyrithiobac sodium.

The rainfall simulator plots gave runoff concentrations that were lower than the data for paraquat at the same soil concentration (Fig. 2). At the lowest paraquat concentrations, paraquat concentrations were almost an order of magnitude greater than rainfall simulator concentrations, because of this seven-fold enrichment. Pesticides from rainfall simulator plots were less enriched, with concentrations in sediment (mg kg⁻¹) closer to the soil concentration. Enrichment ratios for paraquat and their relationship to sediment concentration are like relationships found for sediment-sorbed chemical transport from lighter textured soils (Rose and Dalal 1988; Smith et al. 1993).

Baker (1980) (data from Baker and Laflen 1979)

Baker and Laflen (1979) studied runoff of atrazine, propachlor and alachlor (K_OC = 100–200), applied to the surface on wheel tracks and non-wheel tracks, or incorporated into the soil. They used a rainfall simulator to apply 70 mm h⁻¹ of rainfall 7–11 h after spraying (Table 3). The study was typical of many older simulator studies, with rain applied soon after spraying at a high intensity. Baker (1980) found the relationship between event average concentration in runoff and 0–10 mm soil was C_RO (μg L⁻¹) = 45 × [C′_SOIL (mg kg⁻¹)]^1.03 (R = 0.81). The agreement with the Leonard et al. (1979) data for water-transported herbicides was surprisingly good. Indeed, the data from the two studies were not significantly different in the mean E_RO (P < 0.001) (Fig. 3). Both studies had slightly greater runoff concentrations than Australian rainfall simulator plots (Silburn 2023) at higher soil concentrations. Simulated rain on small plots (Baker and Laflen 1979) provided a useful estimate of herbicide runoff from catchments, albeit at different locations. A direct comparison of simulator and catchment data at the same location (White et al. 1976) is given below.

Fig. 3.

Pesticide soil and runoff concentrations from North American rainfall simulator and catchment studies, compared with Leonard et al. (1979) (C_ro = 50 × C′soil^1.2), Baker (1980) and Silburn (2023) (C_ro = 30 × C′soil), and lines approximating limits of the data.

Other North American studies in crop land

Data for runoff and soil concentrations are shown in Fig. 3 for catchment studies of Ritter et al. (1974) for atrazine and propachlor, White et al. (1976) for 2,4-D, Edwards et al. (1980) for glyphosate and Buttle (1990) for metolachlor, and rainfall simulator plots of Trichell et al. (1968) for three acidic herbicides. In the interests of brevity these studies are described in Table 3. The data of Baker (1980) and Leonard et al. (1979) and best-fit lines from Leonard et al. (1979) and Silburn (2023) are also included.

The herbicide data from these five studies show less agreement with the other data (Fig. 3). Runoff extraction for atrazine (Ritter et al. 1974; Baker and Laflen 1979) was significantly greater while runoff extraction was significantly lower for glyphosate (Edwards et al. 1980) and metolachlor (Buttle 1990) (Table 4). Surface-applied metolachlor gave higher soil (0–25 mm) and runoff concentrations than when incorporated (Buttle 1990). A two-sample t-test showed that the surface applied and incorporated treatments were significantly different. Mean E_RO for surface applied was 4.81, significantly greater than the mean E_RO of 2.61 for incorporated. The mean average extraction ratio for the data of Ritter et al. (1974) was 167, five times those from Silburn (2023). Runoff concentrations for all five studies approached the maximum potential extraction of 200 for events soon after spraying (Fig. 3). Extraction ratios were highest soon after spraying and declined with time. The studies of Ritter et al. (1974) and Buttle (1990) were under highly erosive conditions and experienced large pesticide losses. Ritter et al. (1974) recorded a storm 7 days after application where 15% of applied atrazine was lost in runoff. Buttle (1990) measured 31–46% of applied metolachlor lost in sediment. This is higher than generally expected for this soluble herbicide because sediment concentrations were high, averaging >100 g L⁻¹ and above 200 g L⁻¹ in some events, compared with 6 g L⁻¹ on average for Leonard et al. (1979).

Glyphosate data (Edwards et al. 1980) fitted the relationship C_RO (μg L⁻¹) = 56.3 × [C′_SOIL (mg kg⁻¹)]^1.6, steeper than the relationships from Leonard et al. (1979) and Baker (1980). The catchment had high cover (sprayed-out meadow), and soil concentrations included glyphosate on cover and runoff included washoff from cover. Aslam et al. (2013) found that glyphosate was more easily desorbed from decomposed maize residues than were other herbicides (e.g. metolachlor). Rampoldi et al. (2011) found limited and reversible glyphosate adsorption on soybean and maize crop residues. Thus, the higher runoff extraction of glyphosate was probably explained by its greater desorption from the sprayed-out meadow grass. High runoff losses from meadow cover were like higher glyphosate runoff losses from cane trash compared to bare soil found by Silburn et al. (2023). The 2,4-D (White et al. 1976) had greater runoff concentrations than the other data (Leonard/Silburn) soon after spraying but fitted with the other data at longer times after spraying. High runoff concentrations for acidic herbicides (Trichell et al. 1968) were related to high soil concentrations and were not due to inherently greater runoff extraction.

White et al. (1976)

White et al. (1976) measured runoff of 2,4-D from a small, cultivated catchment in Georgia (annual corn, conventional till) (Table 3). They present maximum and minimum concentrations for storm runoff and annual maxima over 3 years and soil concentrations (0–5 mm) for 1 year. Event average concentrations were not given, so means of range values were used. Also, simulated rainfall was applied to three 30.2 m² sub-plots at 1, 8 and 35 days after 2,4-D application, at 165 mm h⁻¹ for 30 min. Soil concentrations for storms and simulator plots were determined from the soil concentration-days since spraying data. Two aspects are of interest here, namely, comparison of rainfall simulator with catchment data, and comparison of soil and runoff concentrations.

Mean 2,4-D runoff concentrations from the rainfall simulator and catchment fit the same relationship with days after spraying well (Fig. 4, R² = 0.906), indicating that the rainfall simulator provides a good estimate of catchment pesticide runoff. Maximum runoff concentrations for simulator plots agreed closely with maximum concentrations from the catchment, as a function of days after application (Fig. 4, R² = 0.937).

Fig. 4.

Comparison of 2,4-D runoff concentrations from rainfall simulator plots and catchment, and soil concentrations through time after application (White et al. 1976). Concentrations are means (symbols), maximum and minimum (error bars), and maximum annual (‘max’).

Mean runoff concentrations from the rainfall simulator and catchment relate well to soil concentrations (R² = 0.94) and the equations of Leonard et al. (1979) and Silburn (2023) (Fig. 5), apart from at 1 day after spraying (the largest value). This point had a higher runoff concentration, which may be related to less leaching from the runoff-mixing layer and less contribution in sediment, as the partitioning coefficient was lower than at early times (White et al. 1976). Maximum runoff concentrations for the rainfall simulator and catchment were well correlated and also correlated with soil concentrations (R² = 0.98), with an extraction coefficient three times the event average.

Fig. 5.

Mean and maximum 2,4-D runoff concentrations from rainfall simulator plots and catchment, related to soil concentrations (0−5 mm) (White et al. 1976), compared with Leonard et al. (1979) (C_ro = 50 × C′soil^1.2) and Silburn (2023) (C_ro = 28 × C′soil). Concentrations are means (symbols), maximum and minimum (error bars), and maximum annual (‘max’).

Irrigated cropland – Spencer et al. (1985)

Spencer et al. (1985) measured runoff, sediment and concentrations of 11 pesticides (herbicides and insecticides) in runoff from 10 furrow irrigated fields (28–60 ha) in Imperial Valley, California, for 1–3 years each (on silty clay or silty clay loam soils, 40–50% clay, OC = 0.4–0.7%) (Table 3). Event average (flow-weighted) pesticide concentrations were given. Sediment concentrations were low, with an average across fields of ~0.7 g L⁻¹ (0.2–1.2 g L⁻¹) or 0.2–2 g L⁻¹ for individual irrigations. Before some irrigations, 0–10 mm soil samples were analysed for pesticides. These data were used by Spencer et al. (1985) to investigate relationships between pesticides in runoff and in soil prior to irrigation. The pesticides studied are listed in Table 1, and have a range in K_OC of 72 (methomyl) to 11,500 (endosulfan).

Runoff concentrations relate well to soil concentrations for each pesticide or pesticide group (Fig. 6, R² = 0.59–0.95), though there was variation between some events. This was not surprising given data are from a variety of fields, years, crop stages and irrigations within seasons. Prometryn and methomyl data coincide well with the relationship of Silburn (2023). Runoff extraction ratios for the prometryn data of Spencer et al. (1985) significantly differed to those of Silburn (2023). In contrast, the runoff extraction ratios of trifluralin were the same (n.s.d. P = 0.007) for the data of Spencer et al. (1985) and Silburn (2023). Runoff extraction ratios varied by a factor of seven and were highest for the less sorbed prometryn (31.3) and methomyl (21.7) and decreased to 4–7 for more sorbed compounds.

Fig. 6.

Mean runoff concentrations from furrow irrigation related to furrow soil concentrations (0−10 mm), Imperial Valley, California (Spencer et al. 1985), compared with C_ro = 27.7 × C′soil (Silburn 2023). Herbicides DCPA and dinitramine. The OPs were chlorpyrifos, diazinon, ethyl-parathion, malathion and methyl-parathion. K_OCWauchope et al. (1992).

The E_RO values for each pesticide or group (Fig. 6) are plotted against partition coefficients in Fig. 7. Decreasing runoff extraction ratio with greater partition coefficient was consistent where partition coefficients were measured in irrigation runoff by Spencer et al. (1985) (Fig. 7). Measured partition coefficients were about five times ‘selected’ soil sorption coefficients (Wauchope et al. 1992) for chlorpyrifos, endosulfan and trifluralin, 1.6 times for DCPA and similar for prometryn (see also Silburn 2003; Silburn et al. 2023).

Fig. 7.

Runoff extraction ratios (slopes of equations fitted in Fig. 6) related to soil sorption coefficients (K_D) from K_OC (Wauchope et al. 1992) and OC = 0.5%, or partition coefficients (K_P) in irrigation runoff (Spencer et al. 1985). OPs: chlorpyrifos, diazinon, ethyl-parathion, malathion and methyl-parathion.

Lower runoff extraction for more sorbed pesticides was related to low sediment concentrations in runoff. The contribution of sediment phase pesticides to the total concentration was low (e.g. <20%), even for highly sorbed pesticides (‘Fsed SC = 0.7 g L⁻¹’ in Fig. 8a). For tightly sorbed pesticides, little was transported in the water phase, so without sediment movement, little was transported in total.

Fig. 8.

(a) Pesticide sediment phase fraction (F_SED) related to K_P, for the actual sediment concentration in runoff (~0.7 g L⁻¹) and for sediment concentrations of 25−50 g L⁻¹. E_RO–K_P shown to indicate the K_P for each pesticide. (b) Runoff extraction ratios for measured sediment concentration and calculated for sediment concentration of 25 g L⁻¹, using measured K_p values.

Had sediment concentration in runoff been greater, the sediment phase proportion would have increased for more sorbed pesticides (e.g. 80–100% for partition coefficients K_P > 100, Fig. 8a) and runoff extraction ratios would have been higher (E_RO = 25–35), similar to prometryn (Fig. 8b). Even for DCPA with a measured K_P of 40, sediment phase proportion increased from almost nothing with sediment concentration of 0.7 g L⁻¹ to about 50% with sediment concentration of 25 g L⁻¹, doubling the E_RO to 34. Increased sediment concentration did not affect runoff extraction of prometryn, because little was transported on sediment (K_P = 1.5). Interpretation of the Spencer et al. (1985) data indicates that runoff extraction ratios are like those of Silburn (2023), once differences in sediment concentrations are taken into account.

What differentiates runoff extraction between pesticides?

Differences in runoff extraction between pesticides for bare soil occur in the following cases:

Factors such as cover and slope impact runoff extraction to the extent they affect sediment concentrations and hydrology. Cover reduces erosion and enhances infiltration and thus reduces runoff extraction if interflow does not occur. Greater slope leads to greater sediment concentrations and increase runoff extraction of more sorbed pesticides.

Leaching from the runoff-mixing zone

Silburn (2023) calculated effects of infiltration on pesticide leaching for a 25 mm deep soil layer using the advection equation¹ (Leonard et al. 1987) for a range of soil sorption coefficients (K_D) and indicative pesticides. Soil concentrations were barely affected by leaching for pesticides with K_D greater than 100 (e.g. endosulfan, trifluralin and DDE), even with 100 mm of infiltration. This is consistent with endosulfan being found mostly in the surface few centimetres of soil months after application (Kennedy et al. 2001). Leaching only reduced concentrations by 10% during 50 mm of infiltration for K_D = 14 (e.g. prometryn). With K_D = 1.4 (e.g. atrazine or fluometuron), soil concentrations were reduced by ~25% with 30 mm of infiltration (at the start of runoff), were halved with 50 mm and reduced by 20% with 100 mm. The K_D needs to be ~0.2 for soil concentration to be halved with 30 mm of infiltration. Thus, K_Dmust be low (<2) for leaching to significantly reduce concentrations in 0–25 mm soil during a large rainfall event. This is consistent with Briggs (1981) who found that K_D < 1 was a critical divide for leaching. While many pesticides have K_D > 2, many herbicides and a few insecticides will have values around or less than two (Briggs 1981).

For other Australian rainfall simulator sites (Silburn 2023), infiltration amounts were about half those at Emerald (46–53 mm), only 17–25 mm. Thus, a considerable proportion of pesticides were still in the mixing-layer and available to mix with runoff. In contrast, where infiltration was highest (65 mm, Emerald covered plots), nitrate-nitrogen runoff indicated that leaching was restricted by subsoil compaction, especially on wheel-track plots (Silburn et al. 2002), causing interflow and more complete mixing. This is consistent with much greater runoff bromide ion (Br⁻) concentrations where leaching was restricted (Ahuja and Lehman 1983). However, for Emerald plots, runoff concentrations of prometryn, the least sorbed pesticide studied, only increased slightly on covered wheel-track plots, probably because some leached prometryn was adsorbed. Thus, pesticides must be very weakly sorbed to runoff, similarly to nitrate-nitrogenor Br⁻. A case where leaching probably reduced runoff extraction was dimethoate (poorly sorbed) at Gatton where runoff extraction was more than halved (Silburn 2023). This is supported by soil concentrations measured after rain which were only 5% of concentrations before rain.

Thus, reduction in runoff concentrations due to leaching was not influential in the rainfall simulator studies; infiltration was too small and soil sorption too high. The lower infiltration for the Australian rainfall simulator plots is not unusual during storm rainfall on bare cultivated clay soils used for grain and cotton cropping in eastern Australia (Silburn 2023). Restricted infiltration through compacted subsoil is common (Silburn and Connolly 1995). The similarity of runoff extraction for the Australian studies and most of North American runoff studies indicates that the same processes were probably occurring in both regions.

Why pesticides have similar runoff extraction ratios

Several factors push runoff extraction towards similarity rather than difference between pesticides, so long as leaching is restricted. An important factor is that the mass of soil and volume of water involved in mixing are the same for all pesticides on a plot. The consistency of runoff extraction indicates that this is also similar between plots and catchments. Also, many factors that increase solute extraction also increase detachment of sediment (e.g. increased rainfall intensity, greater slope and low cover) (Ahuja 1986, 1990), and cause similar extraction of poorly sorbed and tightly sorbed pesticides. Most pesticides studied had some contribution in the water and sediment phases. Also, as sediment concentrations decreases, physical enrichment of chemicals in sediment increases (Rose and Dalal 1988), so the sediment phase concentration does not decrease in linear proportion with the sediment concentration. The net effect is to moderate expected effects of sediment concentration on runoff extraction, except at very low or high sediment concentrations. Thus, in the Gatton simulator study of Silburn (2023), runoff extraction was similar between plots with sediment concentrations from 20 to 60 g L⁻¹.

Runoff concentration will be linearly related to soil concentrations where large differences in days after application (>50 days) do not occur. Longer time after spraying also means that runoff extraction is lower, as partition coefficients increase (Wauchope et al. 2002; Silburn 2003) and the proportion in the surface of the mixing-layer decreases. Thus, a non-linear relationship between runoff concentrations and soil concentrations is likely when long times after application are included (Leonard et al. 1979).

Climate change may change the intensity of rainfall across northern Australia. Fraser et al. (2011) developed a model to estimate 15-min rainfall intensities (I₁₅) across Australia. This used minimum temperature, daily rainfall and diurnal temperature range to estimate I₁₅. The model was calibrated on one long-term set of pluviometer data and validated on another. The resulting model, while simple, was considered an improvement on the previous model. The model was considered a robust model of short duration rainfall intensity across Australia, particularly northern Australia. Analysis of trends in observed I₁₅ showed that I₁₅ generally increased over the period of data (average record length 37 years), with a greater rate of increase over more recent years (1990–2004). In principle, increased rainfall intensities will lead to increased risk of runoff occurring and thus would increase the risk of pesticide runoff losses.

Knowledge gaps

Major knowledge gaps include understanding runoff extraction for a wider range of pesticides including more recently introduced pesticides and understanding how runoff extraction is related to pesticide properties. There are also major knowledge gaps around pesticide half-lives in soil and on crop residues in tropical and sub-tropical environments. Understanding the toxicity and comparative runoff risk of pesticides is a major knowledge gap.