Ecological effects of pesticide use in the netherlands: Modeled and observed effects in the field ditch

Authors

  • Dick de Zwart

    Corresponding author
    1. National Institute for Public Health and the Environment (RIVM), Laboratory for Ecological Risk Assessment, PO Box 1, NL-3720 BA, Bilthoven, The Netherlands
    • National Institute for Public Health and the Environment (RIVM), Laboratory for Ecological Risk Assessment, PO Box 1, NL-3720 BA, Bilthoven, The Netherlands
    Search for more papers by this author

Abstract

This study addresses the potential risks to the aquatic ecosystem posed by pesticides currently used in The Netherlands. The study used a novel method to predict aquatic exposure to pesticides based on a geographic information system (GIS) map of agricultural land use, comprising 51 crops used in open-canopy areas. Through the application of species-sensitivity distributions for aquatic organisms, in combination with rules for mixture-toxicity calculation, the modeled exposure results were transformed to risk estimates for aquatic species. The majority of the predicted risks were caused by pesticides applied to potato cropland, and approximately 95% of the predicted risk was caused by only 7 of the 261 pesticides currently used in The Netherlands. For risk verification, local toxic-risk estimates were compared with observed species composition in field ditches. The field verification study was not able to draw firm conclusions regarding the predicted impact of pesticide use on overall biodiversity. A toxicity-related shift from sensitive to more tolerant or opportunistic species could be observed for a few species.

INTRODUCTION

Although small, The Netherlands experience a very intensive agricultural practice when compared with other European countries. Of the country's about 32,500 km2 of total dry-surface area, about 22,000 km2 is dedicated to open land use for culturing crops and feeding cattle. In 1998, approximately 6.5 million kg of 261 different active pesticide ingredients was applied to both pasture and crop lands. The elevation of about half the country is at sea level. Without dunes and water barriers, The Netherlands would be flooded. Therefore, the agricultural land is drained by an extensive system of constructed ditches. The ditches are connected to canals from which the excess water is pumped to rivers and, eventually, to the sea. In the lower regions of the country, the ditches form a network with an interdistance between 25 and 100 m.

Because of the close proximity of agriculture and surface water, and depending on the application requirements, land use, and chemical properties of the different pesticides, there may be the potential for unintentional contamination of the aquatic environment. The question is whether unintended releases of pesticides to the environment present a significant risk to the aquatic community.

Local and regional water management in The Netherlands is in the hands of regional Water Boards, responsible for flood control, water quantity, water quality, and the treatment of urban wastewater. Each of the 35 regional Water Boards operate a monitoring network. The concentrations of pollutants and water quality parameters are regularly measured in combination with surveys of the occurrence of aquatic species. However, a certain lack of realism in measured pesticide concentrations may be ascribed to the sampling schemes, which are not adjusted to the regime of the pesticide application. There may also be regional discrepancies between the types of pesticide measured and the types of pesticide used in agricultural lands.

This article introduces a novel, modeling method that uses the distribution of crop types in agricultural areas to map the potential risks associated with predicted pesticide exposure to organisms in aquatic communities. Compared with measured pesticide concentrations, this method has the advantage that modeled effects can be attributed locally to the types of crops grown, as well as to the types of pesticides used. Using this proposed approach enables environmental managers and scientists to conduct scenario studies that can guide policy decisions in land-use planning and in pesticide regulation.

The objectives of this study were to demonstrate that aquatic exposure to pesticides can be predicted with plausible results based on knowledge of the crops grown and crop-specific pesticide-application rates, including several exposure pathways and attenuation processes. Because exposure concentrations in the aquatic environment, without some understanding of pesticide toxicity, do not provide any information on ecological impact, a 2nd objective was to convert predicted concentrations of individual pesticides and mixtures of pesticides into estimates of aquatic risk to potentially affected species. Scaling pesticide concentrations according to toxicity may ultimately provide a valuable asset to risk management and support future environmental legislation because it will enable the ranking of agricultural crops, pesticides, exposure routes, and regional exposure according to their predicted ecological impact. The 3rd objective was to evaluate whether the modeled impact of pesticide use could be demonstrated in biota in field ditches located adjacent to agricultural areas, which are also subjected to the influence of other environmental stress factors.

METHODS

A geographic information system (GIS) was used to estimate pesticide concentrations in field ditches by combining a crop map with crop-specific weekly application rates of pesticides. The main input to the analysis was the 1998 map of The Netherlands specifying the types of 51 different crops grown on agricultural land (including grassland). The GIS map contained about 122,000 grid cells, each representing 500 · 500 m of land area.

Table Table 1.. Land-use classes for agricultural areas in The Netherlands in 1998a
CodeLand-use classSurface area (km2)Number of crops
  1. a LGN4 (www.lgn.nl).

1Pasture12,5103
2Corn/maize2,5903
3Cereals1,8702
4Potatoes1,7905
5Sugar beet1,1401
6Fruit orchards2805
7Flower bulbs2306
8Other crops1,85026
 Total22,26051

Exposure pathways and processes incorporated in the assessment to calculate freely dissolved water concentrations included: (1) aquatic exposure by direct spray drift, (2) runoff and drainage (R&D) from the soil, (3) wet and dry deposition for airborne pesticides, (4) sorption to soil particulates, (5) leaching to deeper groundwater, and (6) degradation in soil and water. By using species-sensitivity distributions (SSD) for aquatic species (de Zwart 2002), together with criteria for mixture-toxicity evaluation (Traas et al. 2002; de Zwart and Posthuma 2005), the calculated concentrations of different pesticides were transformed into risk estimates for an aquatic community. The risk is expressed as the multisubstance potentially affected fraction (msPAF) of species, which is defined as the proportion of species exposed to a mixture of pesticide concentrations exceeding their respective predicted no-effect concentration (PNEC).

Available data

Land use and crops—Geographical land-use data were obtained from a database on land use based on satellite imaging for the years 1999 and 2000 (LGN4, www.lgn.nl, September 2003). The satellite image recognizes 8 types of agricultural land use. The land-use data were aggregated to the scale level of grid cells of 500 · 500 m by calculating the distribution of land-use classes from 400 pixels (25·25 m) on satellite images within each grid cell. The land-use data were combined with the 1998 detailed-inventory data on municipal crop areas from the Agricultural Economics Research Institute (LEI-DLO, www.lei.dlo.nl/home.htm, September 2003). This inventory comprises 540 of 548 municipalities in The Netherlands and quantifies the area of 51 different crops in each municipality. To relate actual crop data to land-use classes, the areas of the 51 crop types were proportionally classified into each of 8 agricultural land-use classes in each grid cell within each municipality. If several crops are grown that belong to the same land-use class, then the relative area of those crops in the municipality was applied to the area of the corresponding land-use class in each grid cell in the municipality. With most land-use classes only a few crops are associated. Using this approach, 26 of 51 different crops were assigned to the land-use class that contains “other agricultural crops.” The national surface area of the 8 agricultural land-use classes and the number of crops are shown in Table 1. The crops and their land-use class are given in Table 2.

Table Table 2.. Agricultural crops grown in each of the land-use classes indicated in Table 1
ClassCrop nameClassCrop nameClassCrop name
1Grass seed6Pear, old8French endive
1Permanent pasture6Pear, young8Green peas
1Temporary pasture7Daffodil8Headed cabbage
2Corn7Gladiolus8Hedge plants
2Corn-cob mix7Hyacinth8Leek
2Fodder maize7Iris8Marrowfat pea
3Summer barley7Lily8Other flowers
3Winter wheat7Tulip8Oyster plant
4Plant potato on clay8Asparagus8Park trees
4Plant potato, other8Brown beans8Perennial garden plants
4Potato on clay8Cabbage, storable8Plant onion
4Potato, other8Chicory root8Rose
4Starch potato8Cocktail onion8Seed onion
5Sugar beet8Conifer8Small carrot
6Apple, old8Dwarf bean8Sprout cabbage
6Apple, young8Field bean8Strawberry
6Other fruit tree8Flowers to be dried8Winter carrot
Table Table 3.. Chemical properties of active pesticide ingredients used in The Netherlandsa
PropertyClarification
  1. a Secondary sources, reported in de Zwart 2003.

  2. b Leistra et al. 2000.

Kom (L/kg)The partitioning coefficient between soil organic material and pore water used to calculate the concentration of pesticide in runoff and drainage water as interpolation input for the metamodel PEARL.b
kW (1/week)The degradation rate constant in water per week, used to calculate the residual concentration from the previous weeks exposure.
kS (1/week)The degradation rate constant in soil per week, used to calculate the concentration of pesticide in runoff and drainage water as interpolation input for the metamodel PEARL.b
VdW (cm/s) for 34 compoundsThe downward, dry-deposition velocity to water in cm per s, used to calculate dry deposit of airborne pesticide per unit area of surface water.
VdS (cm/s) for 34 compoundsThe downward, dry-deposition velocity to soil in cm per s, used to calculate dry deposit of airborne pesticide per unit area of soil.
LC50 (μg/L)Toxicity of the compound for a variety of aquatic species, subdivided by major taxon, used for constructing species-sensitivity distributions that form the basis for toxic risk estimation.
Toxic mode of action (TMoA)An indication of the molecular processes affected by the chemical; 99 different modes of action are recognized, used to make decisions on applying a concentration or a response-addition strategy in calculating the combined risk of exposure to multiple chemicals.

Data on pesticides—For the 51 crops, the national use of 261 pesticides on a weekly basis in 1998 was available from the ISBEST 4.0 database maintained by Alterra, Wageningen, The Netherlands (http://statline.cbs.nl, September 2003; see also www.alterra.nl). Together with the estimated crop areas, both nationwide and for each grid cell, this information generates the estimated weekly use of active ingredients in units of kg/grid-cell/week. For each of the 261 different active pesticide ingredients, it was possible to generate estimates of the chemical properties presented in Table 3 by consulting open literature, as well as by querying publicly available databases on chemical properties. These data can be found in the original RIVM report (de Zwart 2003), which is freely available on the internet (www.rivm.nl). Drift is expressed as the crop-specific percentage of the applied dose (kg/grid-cell/week) transferred to a hectare of surface water. Based on the data describing weekly application rates for the individual pesticide ingredients, per crop and per grid cell, together with chemical properties, a query of ISBEST 4.0 generated data on the average fraction of each pesticide ingredient transferred to the soil. The drift given by ISBEST 4.0 is based on the evaluation of experimental data presented by de Nie (2002).

Air concentrations and deposition—During 2000 and 2001, The Netherlands Organisation for Applied Scientific Research (TNO) operated a monitoring program quantifying the concentration of pesticide ingredients in air and precipitation at 18 stations in The Netherlands (TNO 2002). Thirty-four out of 261 pesticide ingredients were selected for this monitoring program by virtue of their ability to evaporate and become airborne. The concentrations of pesticides in air and precipitation were converted to nationwide, 10-km2 grid cell–based maps by applying kriging interpolation. The concentration in air was expressed in ng·m–3, averaged per 4-week period. The quantity of pesticides in rainwater was expressed in g per ha per 4-week period. Divided by 4,000, this yielded the estimated, weekly load (kg/ha/week) used in the current analysis. The weekly amount of rain (mm) was calculated as one-fourth of the 4-week amount. Daily average temperature data during the period 1991 to 2000 were obtained from the Royal Netherlands Meteorological Institute (www.knmi.nl/product, September 2003). Weekly potential evapotranspiration (PET, in mm) was calculated from the 10-y series of daily average temperature (T) values, using the relationship between T and PET from the 30-y monthly averages given in Table 4.

Pesticide concentration in surface water runoff and drainage water—The PEARL model (Leistra et al. 2000) was used to generate a meta-model table of pesticide concentrations in surface water runoff and drainage water (porewater) as a function of 2 chemical properties for a range of imaginary pesticides, assuming an application of 1 kg per hectare to a standard soil. The chemical properties were the half-life in soil (DT50), ranging between 1 and 200 d, and partitioning between soil organic matter and porewater (Kom fom), ranging between 1 and 200 L/kg. The table was generated assuming that the organic matter content of soil was 4.7%.

Table Table 4.. Monthly average temperature and potential evapotranspiration (PET) rates in The Netherlandsa
MonthTemp (°C)PET (mm/month)
  1. a KNMI (www.knml.nl).

12.88.3
22.915.7
35.632.9
48.156.4
512.585.1
615.090.2
717.295.1
817.183.1
914.250.3
1010.427.8
116.311.5
124.06.5

Grid-based soil properties—From the 500-m2 grid cell–based soil map of The Netherlands (de Vries and Denneboom 1992), relevant soil properties were identified. These data are summarized in Table 5.

Species monitoring data in ditches—Water Boards operate an ecological monitoring network that provided species census data for 257 field ditches in The Netherlands from which empirical data on macrofauna and macrophytes were collected. The 1998 data were retrieved from the database Limnodata Neerlandica (R.A.E. Knoben, Royal Haskoning, 's Hertogenbosch, The Netherlands, personal communication, 17 April 2003). The database comprises counts of 1,007 macrofauna taxa and 291 macrophytes. Removing extremely scarce species, a total of 344 macrofauna taxa and 113 macrophytes were retained for this study. If a station was evaluated 2 or more times during 1998, then the maximum count per taxon was used. Identified taxa range from species to order. For practical purposes, the term “species” will be used for all entities, regardless their taxonomic level. The biological data set was matched with a chemical data set at 212 out of 257 stations. The data set included data on the concentrations of chloride (Cl), total phosphorus (TP), Kjeldahl nitrogen (KN), dissolved oxygen (DO), and pH. If the stations were visited several times, then the average was calculated from the number of observations recorded in 1998.

Exposure assumptions and analysis—Three assumptions were made in the exposure calculations performed to estimate exposure and risk in a field ditch: (1) measurements of a standard field ditch included an overall width of 1 m, a depth of 0.30 m, sides sloping 45°, with 210 L of total water volume in each 1 m2 of the standard field ditch; (2) all fluxes of pesticide input to a field ditch on a weekly basis were assumed to take place at a single moment in time; and (3) the surfacewater in a field ditch was assumed to be completely stagnant despite the input of rain and drainage water.

All calculations, including the risk estimation, were performed 1 grid cell at a time, without considering the influences of adjacent grid cells. This is considered an allowable option by realizing that the particular geomorphology of The Netherlands is rather homogeneous on a regional scale, where geomorphology is linked to crop type. (The algorithms used in this study are available on the Internet at www.rivm.nl.) The steps taken to perform the exposure assessment are summarized in Table 6.

Risk calculation

Species sensitivity distributions—Toxic risk was calculated using the SSD methodology (Posthuma et al. 2002). An SSD curve (Figure 1) is a cumulative density function (CDF) of laboratory-derived toxicity data for a single toxicant. In this study, the SSD curves were assumed to follow a log-normal distribution of no observed-effect concentrations (NOEC). SSD curves are used to derive environmental quality criteria (EQC) and quantify ecotoxicological risk. As an EQC, the hazard concentration for 5% of the species (HC5) predicts an environmental concentration below which only an a priori, acceptably small proportion of species (5%) would be affected. As a risk estimate, the SSD is used to predict the proportion of species exposed to a concentration generating an adverse effect (the PAF), because the concentration exceeded the NOEC of the affected species.

Toxicity data for each pesticide ingredients was obtained from de Zwart (2002). Both acute median effect concentrations (EC50) and chronic NOEC concentrations were 10log-transformed before calculating the average log toxicity (AVG) over major taxonomical groups of organisms and the associated standard deviation (SD). If sufficient chronic toxicity data were available, then the risk evaluation was based on the NOEC. For many of the pesticide ingredients, chronic data were extremely scarce. In those cases, the risk calculations were performed with chronic toxicity data extrapolated from acute observations. An uncertainty factor of 10 was applied to the acute SSD, in other words,

equation image((1))

as indicated in Equation 1 (de Zwart 2002). For the 261 pesticides, a total of 1,143 AVG and SD values were calculated and used as input for the risk calculation applied to each of the 18 major taxonomic group (AVGTax.Grp and SDTax.Grp).

Toxic risk per pesticide ingredient—From the calculated weekly exposure concentrations, the theoretical risk of each pesticide ingredient to each major taxonomical group was calculated by applying the function NORMDIST(10log[Present week's final concentration (μg·L–1)], AVGTax.Grp, SDTax.Grp, 1) found in Microsoft Excel®.

The theoretical risk per ingredient and per major taxon was averaged over the major taxonomic groups. Because the data set generated by this calculation would be too large to store electronically (122,000 · 261 · 52 ≈ 1.7 giga-records), the PAF values per pesticide and per grid cell were averaged over 52 weeks, and only the nonzero values of the yearly average PAF (Avg PAF) were retained for future use in the analysis. The yearly average origin of loading percentages (Old%, Drift%, Dry%, Wet%, and R&D%) for each pesticide ingredient also were retained in the analysis.

Table Table 5.. Key soil properties used to predict releases of pesticides to the aquatic environment
PropertyClarification
  1. a Leistra et al. 2000.

fomThe fraction of organic matter in the soil, used to calculate soil sorption and the concentration of pesticide in runoff and drainage water as interpolation input for the metamodel PEARL.a
fwaterThe fraction of surface-water area per grid cell, used to calculate the additional effect of exposure to the pesticide content in drainage and runoff water.
Soil permeability (m/d)The leeching velocity of water in soil in m per d, used to calculate the proportion of precipitation excess that will end up as runoff and drainage water.
Table Table 6.. Steps taken to perform the exposure calculations used to quantify the concentrations of 261 pesticides potentially released to field ditches in The Netherlands
StepResultMethod of calculation
  1. a TNO = Netherlands Organisation for Applied Scientific Research.

  2. b R&D = runoff and drainage.

  3. c Leistra et al. 2000.

  4. d PET = potential evapotranspiration.

1Rain input to ditches (L/L/week)Actual precipitation data, combined with standard ditch dimensions.
2Rain input to soil (L/m2/week)Actual precipitation data (mm/m2/week).
3Concentration of 34 pesticides in rain water (μg/L/week)Pesticide-loading with rainwater (g/ha/week), combined with precipitation per week (mm/week).
4Spray drift exposure of ditch to 261 pesticides (μg/L/week)Combination of crop area per grid cell in relation to the national crop area and crop-related pesticide use, together with the crop- and pesticide-specific drift transfer percentages and the dimensions of the standard ditch. Because only half of the ditches are located downwind of the application field, the calculated drift input is divided by 2.
5Dry deposition of 34 pesticides to the standard ditch (μg/L/week)Weekly concentration of 34 pesticides in air (ng/m3), combined with the dry-deposition velocity to water (cm/s) and the dimensions of the standard ditch.
6Soil-loading contribution A: Soil loading of 261 pesticides during application (kg/ha/week)Function of the crop area per grid cell in relation to the national area, the national use of related pesticides, and the soil-transfer fraction. The soil surface-area treated is corrected for the untreated area of surface water per grid cell.
7Soil-loading contribution B: Dry-deposition loading of 34 pesticides to soil (kg/ha/week)Merging of weekly maps of air concentrations (ng/m3) with the downward, dry-deposition velocity to soil (cm/s).
8Soil-loading contribution C: Wet loading of 34 pesticides to soil (kg/ha/week)The data have been directly provided by the TNOa data set.
9Total loading of 261 pesticides to soil (kg/ha/week)Calculated as the sum of the 3 soil-loading contributions: A, B, and C.
10Concentration of 261 pesticides in R&Db (μg/L/week)Interpolation of the PEARLc metatable with the actual Kom partitioning coefficients of the pesticides and the local soil organic fraction. The DT50(d) in soil has been calculated from the degradation rate constant, kS (1/week). The interpolation result is subsequently multiplied by the total soil loading of the given pesticide obtained in step 9.
11Precipitation surplus (L/m2/week)Subtraction of PETd (mm/week) from precipitation (mm/week).
12Volume of R&Db water received by the ditch (L/L/week)Correction of precipitation surplus for leaching to deeper groundwater storage, taking the ratio between the wet and dry surface areas into account; according to Meinardi and Schotten (forthcoming), there is a log-linear relationship between soil permeability and the fraction of precipitation surplus transferred to surface water; a maximum of 80% transfer to surface water corresponds to a low permeability of 0.01 (m/d), and a minimum of 5% is reached at a high permeability of 20 (m/d).
13Iteration of final concentration of 261 pesticides in the ditch (μg/L/week)All the above calculations are performed for a single, grid cell at a time, for all 261 pesticide ingredients, and for all 52 weeks in the year. Per ingredient, the final water-concentration iteration starts with calculating the concentration that is left over from last week's concentration after 1 week of degradation in the water of the ditches. Subsequently, the drifted and the dry-deposed concentrations are added to the initial concentration to form a subtotal concentration after “dry addition.” Final concentration for the present week is calculated by adding the water containing pesticide input from rain and R&D water, with a correction for volume change. To stabilize the concentration of pesticide ingredients remaining from prior week, this iteration loop is continued for 5 · 52 weeks (5 y). The 1st week of the 2nd year, the final concentration from the prior week is set to the final concentration of the last week of the 1st year, etc.
14Temporary storage of the exposure-assessment resultsThe weekly, individual, pesticide concentration in the field ditches of the grid cell is attributed to the origin of the exposure in percentages of pesticide ingredients originating from the past week exposure (old), drift exposure (drift), dry deposition (dry), wet deposition (wet), and R&D, respectively.
Figure Figure 1..

Cumulative probability distribution of species sensitivity fitted to the observed chronic no-observed effect concentrations (NOEC) for different species. The arrows indicate the inference of a potentially affected fraction (PAF value) of species and the hazard concentration for 5% of the species (HC5).

Combined toxic risk of all pesticide ingredients—The combined toxic risk (multi-substance potentially affected fraction, msPAF) of all 261 pesticide ingredients was evaluated by sequentially applying the mixture toxicity mixed-model (Traas et al. 2002; de Zwart and Posthuma 2005). For pesticide ingredients with the same toxic mode of action (TMoA), concentration additivity was assumed. The weekly, calculated concentrations for each pesticide ingredient were transformed to hazard units per taxonomic group (HUIngredient, Tax.Grp), by dividing by 10math image, followed by summation (ΣHUTMoA, Tax.Grp). The weekly, combined theoretical risk per TMoA and per major taxonomic group (msPAFTMoA, Tax.Grp) was then calculated by applying the function NORMDIST(10log[ΣHUTMoA, Tax.Grp], 0, Average[SDTax. Grp], 1). For groups of ingredients with different TMoA, the response of each major taxonomic group is calculated applying a response additive model. It was assumed that aquatic species do not share a significant correlation in their sensitivity for different toxicants: msPAFTax.Grp = 1 – Pcy(1 – msPAFTMoA, Tax.Grp). The final theoretical risk (msPAF) was calculated as the average msPAFTax.Grp among all taxonomical groups considered in this analysis, assuming equal weight of major taxonomical groups. Because the data set generated by this calculation was too large to store electronically, the average msPAF per 4-week period was retained.

Verification of toxic risk with ecological field observations

Pesticide toxicity is not the only environmental condition governing species composition because other physical, chemical, and biological characteristics also affect the expected community. The observed species composition in the field, in terms of the number and abundance of species, may to a limited extent, be directly related to the predicted toxic risk associated with pesticide exposure. However, in view of the absence of extreme exposure levels and the expected relevance of other environmental stressors, this approach was considered unlikely to yield sufficient explanatory power.

Because the available data set on biological and chemical observations in field ditches covers only 212 sites, it is statistically impossible to include many of the variables that possibly govern species occurrence. The number of predictors (related to degrees of freedom) should be at least a factor of about 10 less than the number of observations. Especially with habitat characteristics that are generally expressed in categories, the increase in degrees of freedom will quickly exceed this requirement.

Therefore, a few chemical water characteristics (Cl, TP, KN, DO, and pH) were selected based on an earlier analysis of the importance of factors determining aquatic community composition (Ertsen and Wortelboer 2002). The influence of individual, environmental predictors can only be discerned if the variables are not highly correlated. The data set of chemical observations in field ditches, joined with the corresponding yearly average estimates of total toxic risk (msPAF) was analyzed to reveal correlation structure. The association between the observed abundance of macrofauna species and macrophytes and the 6 abiotic predictors was established by generalized linear modeling (GLM) (McCullagh and Nelder 1989), yielding a GLM for each species.

Assuming a Poisson distribution, species-specific regressions take the form of Equation 2.

equation image((2))

The 6 predictors were added stepwise to the model with linear and quadratic terms. The quadratic terms were introduced to address nonlinear response relationships such as the ecological optima. The stepwise procedure used the Bayesian information criterion (BIC) (Schwarz 1978) to restrict the addition of terms to those that have a significant contribution to the overall model (p < 0.05), making the full model highly significant. Calculations were conducted using S-Plus 2000 software (MathSoft, Cambridge, MA, USA). The models were used to isolate the driving force of predicted pesticide toxicity on species assemblage. The relationship between toxic risk and both field-observed species richness and total individual counts was also evaluated.

RESULTS AND DISCUSSION

Toxic risk of pesticide ingredients

Only 18% of the 261 pesticides produced a nonzero risk in 1 or more grid cells. The national average of the toxic risk (Avg PAF) for individual pesticides and the number of grid cells with nonzero risk are shown in Table 7. Because each pesticide is linked to specific crops, as well as to the proportion of the pesticide applied to a particular crop, Table 7 makes it possible to score crop categories according to their respective impact on total pesticide toxic risk. Per crop category, the sum over pesticide ingredients is taken using the Avg PAF multiplied by the number of grid cells and multiplied by the proportion of crop use. With 58% weighted risk attribution, potatoes contribute most prominently to the total toxic risk predicted in field ditches.

On a nationwide scale, it can be concluded that only 7 pesticides account for approximately 96% of total pesticide risk to the aquatic community. This was calculated by multiplying the Avg PAF of a pesticide by the number of grid cells in which a pesticide exhibited a nonzero risk and expressing the product relative to the total risk calculated for all pesticides (Table 7). The top 7 pesticide ingredients were fungicide maneb (36%), fungicide fentin-acetate (24%), pyrethroid insecticide lambda-cyhalothrin (11%), pyrethroid insecticide deltamethrin (10%), insecticide chlorpyrifos (8%), herbicide isoproturon (5%), and herbicide monolinuron (2%).

Table Table 7.. Pesticide ingredients, associated crops, and the proportion of pesticide use, arranged according to the calculated, average toxic risk (Avg PAF) for each pesticide and the number of grid cells for which a nonzero risk result was calculated
Pesticide ingredientAvg PAF (%)No. of grid cellsCrop% Proportion of use in crops
Fentin-acetate4436,560Potato97
Maneb3470,558Potato57
   Onions18
   Flowers14
Pirimiphos-methyl233,906Flowers98
Tolylfluanide20209Fruit trees90
Isoproturon1917,657Cereals97
Captan18821Fruit trees84
   Flowers14
Monolinuron177,553Potato94
Esfenvalerate17814Flowers47
   Potato43
Lambda-cyhalothrin1546,153Potato58
   Flowers21
   Cereals9
   Vegetables8
Deltamethrin1546,959Potato55
   Flowers13
   Onions8
Diflubenzuron126,314Fruit trees97
Thiram11 Fruit trees80
   Strawberries11
Fentin-hydroxide8668Potato95
Phosalone64,648Fruit trees81
   Vegetables19
Chlorpyrifos692,874Flowers37
   Potato31
   Grass20
Fenbutatinoxide52,335Fruit trees43
   Garden plants33
   Strawberries19
Metsulfuron-methyl51,017Cereals89
   Grass11
Permethrin5290Vegetables38
   Grass36
   Flowers12
   Garden plants5
Propachlor5387Onions94
Table Table 8.. The percent (%) average, pesticide-specific contribution to overall exposure and risk to the aquatic, ecological community posed by spray drift, wet or dry deposition, and surface runoff or drainage to field ditches in The Netherlands
IngredientResidue from previous weekSpray driftDry depositionWet depositionRunoff and drainage
Atrazin72101080
Azinfos-methyl8911000
Captan1841130
Carbaryl8911000
Carbendazim8119000
Chloorfenvinfos7317451
Chloorthalonil4149180
Chlorpyrifos72126100
Cyhexatin6634000
Deltamethrin7129000
Diazinon59230190
Diflubenzuron8416000
Dimethoate48361221
Dinoterb7921000
Diquat dibromide3952009
DNOC6004360
Esfenvalerate4258000
Fenbutatinoxide5149000
Fentin-acetate6238000
Fentin-hydroxide6634000
Heptenofos3961000
Isoproturon8812000
Koperoxychloride6234005
Lambda-cyhalothrin5743000
Lindane46512370
Linuron5347000
Mancozeb5050000
Maneb49390013
MCPA44180335
Methiocarb3063070
Metoxuron4258000
Metribuzin8911000
Metsulfuron-methyl7525000
Mevinfos34511410
Monolinuron6335002
Permethrin6139000
Phosalone7624000
Pirimiphos-methyl6436000
Propachlor65115200
Simazin5541004
Terbutryn928000
Terbutylazin74100151
Thiram4159000
Tolylfluanide3664000
Triazofos6733000
Zineb4258000
Total Average5833151

Although pyrethroid insecticides (Erstfeld 1999) are known to readily degrade, the predicted risk to the aquatic community associated with the use of these pesticides is quite high (approximately 21%). It should be noted, however, that exposure is not only dependent on persistence, but also on the frequency and amount of application. For pyrethroid insecticides, the application rate is frequent; despite rapid degradation, the high, intrinsic toxicity of this class of pesticides may still produce adverse effects on the aquatic community.

For 46 pesticide ingredients, Table 8 gives the average percentage of the origin of pesticide loading to field ditches. For most ingredients, the residue from the previous week's exposure is most prominent (average 58%). Spray-drift exposure is the 2nd most important source of exposure (average 33%), followed by the amount of pesticide in wet deposition (average 5%). Dry deposition and run off and drainage are negligible sources of exposure to the aquatic community.

The frequency distribution of toxic risk in the aquatic ecosystem associated with the agricultural use of pesticides is given in Figure 2. The aquatic community in up to 75% of the grid cells is predicted to suffer minor impact, with up to 5% of the species affected. A maximum of 51% impact on the aquatic community was predicted for exposure to the mixture of ingredients.

Figure Figure 2..

Frequency distribution of pesticide risk for all grid cells of 25 ha in The Netherlands during 4-week intervals in 1998.

The 4-week, average, total toxic risk of pesticide use on the aquatic assemblage of species in field ditches is depicted in the maps presented in Figure 3. From left to right and from top to bottom, the maps represent the thirteen 4-week intervals in 1998. White grid cells indicate a lack of data or areas where no pesticides are used (e.g., large water bodies). The lightest grey color shown in Figure 3 indicates an average toxic risk affecting less than 5% of the aquatic species potentially present in field ditches. Darker colors represent increasingly higher levels of predicted adverse effects.

As shown in Figure 3, during the first 3 months of the year hardly any pesticides are used, and the toxic risk of the pesticide mixture is below 5%. Pesticide use begins in April, which is evident in Figure 3 on the maps that appear darker in color. Overlaying the risk maps with the known distribution of crops, it can be shown that the culture of flower bulbs, potatoes, and fruit trees are mainly responsible for the onset of toxic risk. The months of the year that indicate the highest risk of pesticide use for nontarget aquatic species are June to August. The risks predicted for the past three 4-week periods of the year are associated with the application of soil-fumigation disinfectants, mainly in flower bulb growing areas.

Figure Figure 3..

Predicted ecotoxicological risk (msPAFNOEC) of pesticide use in field ditches. The 13 maps from left to right and from top to bottom represent the development of pesticide risk for 4-week periods throughout the year 1998. Darker colors indicate higher risk, up to the maximum level of 51% msPAF.

Generalized linear modeling results

The abiotic set of 5 field observations and msPAF at 212 field ditch sites in The Netherlands, used to explain the observed abundance of species, has a certain correlation (Table 9). In Table 9, the lower part of the table shows correlation coefficients (r), whereas the upper, shaded portion of the table describes the significance of the correlation. In the upper part, bold print indicates the few significant relationships between the variables. Pesticide toxic risk (msPAF) has a significant correlation only with chloride concentration (p < 0.001). The correlation, however, was due to 4 brackish outliers in the chloride data set. Because the objective of this study was only to relate the modeled pesticide risk to the species composition in field ditches, the observed significance of the correlations in the other variables was not considered important nor was the omission of habitat characteristics that generally have an important influence on species composition.

One or more of the 6 abiotic predictor variables show sufficient explanatory capacity for the GLM-predicted, numerical abundance of 306 out of 344 macrofauna species and 92 out of 113 macrophytes. Figure 4 shows the frequency distributions of the explained deviance of the GLMs for macrofauna (average 35%) and macrophytes (average 20%). By excluding msPAF, the average explained deviances reduce considerably to 29% and 18%, respectively.

When the values of the abiotic predictors for each of the 212 field ditch sites are substituted into the calibrated regression formulas for each species, the part of the linear predictor related to msPAF (li · msPAF + mi · msPAF2) indicates the “driving force” of toxic risk in terms of the natural, log-transformed numerical abundance of species. The msPAF portion of the linear predictor is called the “contribution of msPAF.” Negative values for the contribution indicate a force that lowers species abundance; positive values result from forces that increase species abundance.

Table Table 9.. Correlation cross tablea and the significance of each correlation coefficient for the abiotic set of observations in field ditches
Chemical parameterbCLTPKNmsPAFpHDO
  1. a In the correlation cross table, r results are presented in the lower-left portion of the table, and p results are presented in the upper-right portion of the table; bold print indicates p results that were statistically significant at p < 0.05.

  2. b Cl = chloride; TP = total phosphorous; KN = Kjeldahl nitrogen; msPAF = pesticide risk; pH = acidity; DO = dissolved oxygen.

Cl 0.5640.2840.0000.0001.000
TP0.13 0.0050.9920.0000.090
KN0.160.25 0.8740.3630.007
msPAF0.40–0.08–0.11 1.0000.069
pH0.420.49–0.150.04 0.000
DO–0.060.19–0.24–0.190.57
Figure Figure 4..

Frequency distributions of the explained statistical deviance in the generalized linear regression for the observed abundance of macrofauna and macrophyte species as a function of 6 environmental factors.

In Figure 5, the msPAF contribution, irrespective of positive or negative contributions, is averaged over the respective groups of species and plotted against the local toxic risk (msPAF) for the aquatic community. Figure 5 illustrates that macrophytes are relatively insensitive to the modeled toxic risk of pesticide mixtures. This is most probably because only 2 herbicides, comprising 7% of the total risk, are among the top 7 pesticides used on agricultural lands. The macrofauna assemblage in field ditches appears to be more strongly sensitive to the risk of mixed pesticides; up to a toxic risk of about 10%, the toxicity does not reduce species abundance. When pesticide mixture risk is increasing, the average predicted species abundance in field ditches is gradually forced lower. In terms of the predicted abundance of species (related to the probability of occurrence), the negative trend of the msPAF-contribution with increasing toxic risk means that the abundance of species that may be predicted by other predictors is gradually multiplied by an increasingly small number (minimum is: e–50 ≈ 10–22) at higher toxic risk. This implies that it becomes much more likely that that average numerical abundances of species, and thus the probability of occurrence, is reduced at higher toxic risks of pesticide use.

Figure Figure 5..

The relationship between mixture toxic risk and the predicted contribution of the multisubstance potentially affected fraction (msPAF) in terms of the predominant environmental factors that may influence the abundance of macrofauna and macrophyte species.

Comparison of model results to field observations

Only a weak relationship was observed between predicted mixture risk values (msPAF) and macrofauna species composition in field ditches, in either the total number of species or the overall abundance of individuals. For the macrophytes, such a relationship was completely lacking. To be able to observe the weak regression trends in the macrofauna data, abundant species (> 1,400) and scarce species (<10 individuals) at each site had to be removed from the analysis. The increased scatter of the data introduced by the most abundant and scarce species in the data set obscured any trends in the data. By reducing the data set to 299 macrofauna species, the slopes of the regression lines are clearly not significant (R2macrofauna, # species is 0.027, and R2macrofauna, total abundance is 0.01), and the percent difference between the predicted number of macrofauna species at the calculated risks of 0% and 38% is 43%. For the number of macrofauna individuals, the percent difference is 38%. Both maximum reductions in number of species and individuals correspond remarkably well to the maximum predicted risk of pesticide use (38%) on the aquatic community.

Table Table 10.. Top 10 sensitive and opportunist macrofauna and macrophyte species observed in field ditches in The Netherlands
MacrofaunaMacrophytes
SensitivesOpportunistsSensitivesOpportunists
Arrenurus crassicaudatusAnisus leucostomusButomus umbellatusCallitriche sp.
CollembolaChironomus plumosusGalium palustreCeratophyllum demersum
Erpobdella octoculataCricotopus sylvestrisLycopus europaeusGlechoma hederacea
Helochares sp.Haliplus lineatocollisMentha aquaticaJuncus effusus
Limnesia maculataNeomysis integerNuphar luteaLemnaceae
Mideopsis orbicularisPhysa acutaNymphaea albaPhragmites australis
Piona conglobataValvata cristataPeucedanum palustrePotamogeton pectinatus
Piona imminuta Ranunculus sceleratusPotamogeton pusillus
Polypedilum nubeculosum Rumex hydrolapathumSpirodela polyrhiza
Radix ovata Solanum dulcamaraWolffia arrhiza
Figure Figure 6..

Distribution of the correlation coefficient (r) between pesticide toxic risk (the multisubstance potentially affected fraction, msPAF) and the observed abundance of individual macrofauna species (n = 299) and macrophytes (n = 106).

Figure 6 shows the ranking distribution of the correlation coefficients between the pesticide msPAF values and the observed abundance of individual macrofauna and macrophyte species. At the 212 monitored field ditch sites, 29% of the macrofauna species (n = 299) have a positive correlation between their abundance and toxic risk. These species may be marked “opportunists” because they most probably display indirect effects by filling the gap left by the 71% of “sensitive” species that are reduced in abundance with increasing toxic risk. For macrophytes (n = 106), opportunist and sensitive species represent 20 and 80% of the community, respectively. Figure 6 clearly illustrates why diversity indices, species richness, and total abundance are not very sensitive indicators for ecological effects over a wide range of toxic exposure. Diversity effects are obscured by a shift in species composition: The abundance of some species is reduced; at the same time, other species increase, leaving the overall biodiversity indices unchanged, even though biodiversity itself changes considerably. Without attributing a tolerance score to the individual species or relating species composition to a reference community, it is generally not possible to demonstrate toxic effects on species diversity, unless an extremely high level of toxicity is detrimental to the majority of species.

Table 10 gives the top 10 sensitive and opportunist species in macrofauna and macrophyte assemblages, respectively. To my knowledge, insufficient data exist to relate this grouping to known sensitivities of the individual species. Studies by Klepper et al. (1999) and Mulder et al. (2003) indicate the differences in endpoint sensitivity between responses at the macroscale of observation (e.g., species numbers and diversity indices) and at the microscale of observation (e.g., the wax and wane of individual species). In Mulder et al. (2003), for example, the responses of nematode communities to environmental stress, in terms of total density, was the most sensitive, followed by numbers of taxa and the Shannon-index, whereas NOECs for separate species covered a broad range from sensitive to tolerant species.

CONCLUSIONS

Although highly complex both in concept and in computer programming, a GIS-based analysis of the potential for exposure of aquatic communities in field ditches in The Netherlands to all pesticides used in outdoor agricultural practice can be accomplished. Regarding the uncertainties in the exposure model, it can be stated that the applied rates of major, individual transfer and attenuation processes are all supported by experimental evidence (Leistra et al. 2000; de Nie 2002). For validation, it would have been nice to be able to compare the predicted concentrations to concentrations measured in the monitoring network. Unfortunately, this is impossible because of the lack of data. The monitoring network focuses little attention on field ditches, and only 2 of the top 7 pesticides (chlorpyrifos and isoproturon, with only 8 and 5% of modeled impact, respectively) have some coverage in the monitoring data of larger water bodies.

The conversion of pesticide-exposure concentrations in field ditches to a measure of toxic risk allows for a toxicity-scaled comparison and ranking of pesticides attributable to particular types of crop. The maps produced on the aquatic risk of combined-pesticide exposure seem plausible from underlying knowledge on the where, what, and when of crops and pesticide use.

The field-verification study was not able to draw firm conclusions regarding the predicted impact of pesticide use on overall biodiversity. The field-verification work performed in this study only gives an indication that the effects on aquatic ecosystems, predicted from the crop-based use of pesticides, may be realistic, even though the available data on species abundance proved to be low (only 212 field ditch sites were available in 1998). However, the GLM-regression and the correlation between pesticide toxic risk and individual abundance of species showed a highly significant relationship between toxic pesticide risk and the abundance of some individual aquatic species. A toxicity-related shift from sensitive to tolerant or opportunistic species can be observed for a few species. The majority of aquatic species are indifferent with respect to the predicted level of pesticide exposure believed to occur in field ditches in The Netherlands. The ecological implications on the aquatic community are an area that deserves further research.

Acknowledgements

Acknowledgement—This study has been performed by order and for the account of the Dutch Ministry of Spatial Planning, Housing, and the Environment, within the framework of project M/500002, “Modeling ecological effects.” D. de Zwart thanks John Deneer, Alterra, for providing the ISBEST 4.0 data; Jan Duyzer, TNO, for providing air and deposition data; Marijke Vonk, Theo Vermeire, Ton van der Linden, Kees Meinardi, and Dik van de Meent, RIVM, for stimulating discussions; and last but not least, the IEAM Editorial Office and Christian Mulder, RIVM, for their review and assistance in the preparation of this manuscript.

Ancillary