modelling pesticide run-off to surface waters. part ii: model application

Pestic. Sci. 1998, 54, 121È130

Modelling Pesticide Run-Oþ to Surface Waters.Part II: Model ApplicationRichard J. WilliamsInstitute of Hydrology, Wallingford, OXON OX10 8BB, UK

(Received 2 March 1998 ; revised version received 4 June 1998 ; accepted 3 July 1998)

Abstract : A conceptual model of pesticide run-o† to surface water from agricul-tural land has been tested against data collected from a catchment study. Ingeneral the model was able to simulate total run-o† and pesticide loss and peakpesticide concentrations to an acceptable level (much better than an order ofmagnitude) for a number of pesticides in run-o† events over Ðve seasons. Hour-by-hour variations in pesticide run-o† were reasonably well estimated, althoughthe timing of the estimated peak in pesticide concentrations was always inadvance of that observed. A simple sensitivity test of the model showed the sorp-tion coefficients and half-lives of the chemicals simulated were important in con-trolling model outputs, although the impact of the latter was reduced if eventsoccurred soon after application. Other important parameters were the extent ofthe enhanced conductivity area above the drains and the parameters controllingthe rate of Ñow of water between the model boxes. 1998 Society of Chemical(Industry

Pestic. Sci., 54, 121È130 (1998)

Key words : mathematical model ; pesticide ; surface water ; bypass Ñow; agricul-ture

1 INTRODUCTION

A conceptual model aimed at predicting short-termpesticide concentrations in surface waters arising fromrainfall events has been described in a companionpaper.1 It is important that any model is tested againstobserved data from sites representative of the condi-tions under which the model might be expected to beused. In this case, the model is designed to estimatepesticide run-o† from agricultural land. A suitable dataset for testing was available for a small catchment inHerefordshire, UK which was within the conÐnes ofADAS Rosemaund, an agricultural research facility. Inthis paper, model outputs are compared to those mea-sured during a Ðve-year monitoring programme.2h4 Theresearch programme that generated these data con-tained within it additional information that helped todeÐne not only the conceptualization of the model, but

Contract/grant sponsor : National River Authority(Environment Agency).Contract/grant sponsor : National Environment ResearchCouncil.

also several of the model parameters. It is therefore anexcellent data set on which to conduct an initial test ofthe model performance for a number of pesticides overa range of hydrological conditions. Ideally, applicationsof the model to other catchment data sets would berequired to perform an independent test.

In addition to testing the model against observeddata it is important to know about the sensitivity ofmodel outputs to variations in input parameters. Asimple approach to sensitivity analysis is presented inorder to assess this, based on two run-o† events undercontrasting hydrological conditions. Such informationis not only useful in the context of the current applica-tion, but helps in determining the emphasis on param-eter estimation in future applications.

2 MODEL APPLICATION

2.1 Experimental data

The experimental data that were used to test the modelwere obtained over Ðve cropping seasons (November

1211998 Society of Chemical Industry. Pestic. Sci. 0031È613X/98/$17.50. Printed in Great Britain(

122 Richard J. W illiams

TABLE 1Values of the Box Thickness (Depth), Moisture Volume Fraction Equivalent toMinimum Water Content (SMIN), Field Capacity (SFC) and Saturation (SMAX),

used in the Model, together with the Bulk Density and Organic Carbon Content

BulkDensity Organic

Box No. Depth (mm) SMIN SFC SMAX (g cm~3) carbon (%)

1 and 5 0È50 0É19 0É27 0É42 1É21 1É72 and 6 50È500 0É24 0É32 0É40 1É42 0É9

3 500È1000 0É30 0É35 0É38 1É38 0É54 1000È2000 0É24 0É25 0É26 1É38 0É57 500È1000 0É30 0É25 0É40 1É38 0É5

1987ÈMarch 1993) from experimental catchments estab-lished at ADAS Rosemaund.2h4 Four nested catch-ments were used : two were sited on the stream (151 haand 35É5 ha, referred to as sites 0 and 1) and the twoothers at the outlets from Ðeld drainage systems (5É3 haand 2 ha, referred to as sites 3 and 5). The soils atRosemaund are predominantly silty clay loam intexture, predominantly from the Bromyard series, butwith Compton and Middleton series also represented inthe valley bottoms. These soils are prone to seasonalwater-logging and consequently nearly all the Ðelds atRosemaund are drained, (typically 1 m depth, 20 mspacing). During the summer the soils can crack andthese cracks may persist at depth through part or all ofthe drainage period. There are also macropores extend-ing to depth, and voids around soil peds due to the veryblocky nature of the soil structure in the lower parts ofthe proÐle.

The monitoring strategy adopted was designed tomeasure pesticide concentrations in the stream resultingfrom rainfall events falling on recently treated Ðeldswithin the catchment. Automatic sampler systems wereused to take a series of water samples at short intervals(usually one hour, but intervals of half an hour and fourhours were also used) over the duration of rainfallevents. In order that these concentrations could belinked to the hydrological response of the catchment,Ñow was measured at each of the monitoring sites.Rainfall and parameters to estimate potential evapo-ration were measured hourly by an automatic weatherstation.

2.2 Selection of model parameters

A complete list and description of the model parametersis given in Part 1 of this paper.1 The general approachwas to assign values to the parameters from measuredor literature data. Where this was not possible becauseof the conceptual nature of the model, values were esti-mated by calibration against observed data.

The values used for the minimum water content,SMIN, the saturated water content, SMAX and the

Ðeld capacity, SFC of each soil box are given in Table 1.These were based on observations of water content andsoil water potential made in one of the Ðelds in theRosemaund catchment.5 The organic carbon contentand bulk density of the soil in each of the model boxeswas estimated from soil proÐle analyses carried out bythe Soil Survey and Land Research Centre6 (Table 1).The soil proÐles also gave indications of the percentageof macropores, old root channels and Ðne and severecracking at a range of depths. These were used to esti-mate the macropore fractions in each of the modelboxes (Table 2). The parameters controlling the move-ment of water between the boxes representing the di†er-ent hydrological compartments (i.e. those parametersbroadly equivalent to the vertical and horizontalunsaturated and saturated hydraulic conductivities)were obtained by calibration. Measured saturatedhydraulic conductivities were available for four siteswithin one of the Ðelds at ADAS Rosemaund (Table 3).6Unfortunately, the sites were selected to identify di†er-ences between the shallow and normal soil phaseswithin the Bromyard series rather than located relativeto the sub-surface drainage arms. Therefore, these datawere most useful for estimating vertical changes in themodel parameters equivalent to hydraulic conductivity.

TABLE 2Values of the Parameters in the Model that are broadly equiv-alent to Vertical and Horizontal Conductivity and Macropore

Extent (CF)

SaturatedConductivity conductivity

(mm h~1) (mm h~1)

Box No. V ertical Horizontal V ertical Horizontal CF

1 0É01 0É01 0É1 0É1 0É022 0É001 0É001 0É005 0É005 0É053 0É001 0É001 0É005 0É005 0É024 0É0 0É0002 0É0 0É0004 0É015 1É0 0É20 2É0 0É2 0É026 1É0 0É20 2É0 0É2 0É057 0É1 0É20 0É4 0É2 0É02

Modelling pesticide run-o†: model applications 123

TABLE 3Saturated Vertical Hydraulic Conductivities (mm h~1) mea-

sured at Four Sites at ADAS Rosemaund.

Depth(mm) Site 1A Site 1B Site 2A Site 2B

0È300 35É8 11É7 4É2 2É1300È450 28É8 12É9 3É3 2É5450È600 37É1 11É3 0É8 1É7600È800 3É3 2É9 1É7 1É2

Sites 1 and 2 were on the shallow and normal phases of theBromyard soil series respectively. Values shown are theaverage of two to four measurements at each site.

The observed data showed a large variation in valuesbetween the Bromyard soil phases and a much lesservariation within a given soil phase. In both soil phasessaturated vertical conductivities were similar over thetop 450 mm. In the Bromyard normal phase there wasthen a marked drop in conductivities below this depth.In the Bromyard shallow phase this drop did not occuruntil 600 mm. This pattern, together with the results ofthe soil water movement study,5 were used to help setthe values for the initial estimates of the model verticalsaturated hydraulic conductivities. Some modiÐcationsto the values were required because, within the model,the driving force for water movement from a box is theexcess of the water content over the minimum watercontent, and the size of the box will therefore inÑuencethe value of the conductivity parameter. There is, there-fore, some scope for using measurements of saturatedconductivity to help in setting these parameter valueswithin the model. The initial calibration involvedsetting values which resulted in drain Ñow of the correctmagnitude beginning at the correct time and was basedon observed data from one of the Ðeld drain outlets (site5) for the 1992/1993 season. The model was then testedagainst Ñow data from the outlet from the whole catch-ment (site 0) for the same season. The values for theparameters used are given in Table 2. The base Ñowrate was set at 4É0 litre s~1, estimated from the observedÑow leaving the catchment in early autumn.

The model was driven by hourly rainfall and dailyestimates of potential evaporation according to themethod of Penman.7 The model was used to simulateconcentrations of the pesticides isoproturon, lindane,simazine, mecoprop, triÑuralin and dichlorprop leavingboth Ðeld drains and at both stream monitoring sites.Parameters for the six pesticides are given in Table 4.The applications were made to crops as part of theirnormal use on the farm, using a tractor-mountedsprayer, subsequently updated to a self-propelled unitfrom spring 1990. For each pesticide for each season inwhich it was studied, the application rates were takenfrom records kept by sta† at ADAS Rosemaund (seeTable 6). Values of the organic carbon partition coeffi-

TABLE 4Physicochemical Properties of the Pesticides used in the

Model Simulations

koc

a Half-lifeaChemical (litre kg~1) (days)

Isoproturon 130b 20cLindane 1100 400Simazine 130 60Mecoprop 7d 21Dichlorprop 1000 10TriÑuralin 8000 60

a Reference 12.b D N Brooke, pers comm.c Fitted from Rosemaund data.d Reference 13.

cients, and degradation rates were obtained from thekoc

literature or from data collected during the Ðeld experi-ments.

3 RESULTS AND DISCUSSION

3.1 Model performance

When applying a pesticide run-o† model, it is necessaryto ensure that the movement of water is adequatelymodelled and that this part of the assessment is madeÐrst.8 The potential evaporation is assumed to equal theactual evaporation whenever there is sufficient wateravailable to meet the demand.

It is clear from Fig. 1 that the model was able torepresent the patterns of Ñow from the catchment verywell. However, the highest peaks were underestimatedby the model while the fall back to base Ñow conditionswas predicted to occur too quickly. This resulted in asubstantial underestimate in the volume of waterleaving the catchmentÈ11É8 mm predicted over theperiod 1 October 1992 to 8 December 1992 comparedto an observed total of 22É5 mm. This was at leastpartly due to an overestimate of the observed Ñowscaused by backing up of the water below the weir,which under high Ñows drowned out the weir. StreamÑow was estimated using a theoretical relationshipbetween stream level above the weir assuming freemovement of water away from the downstream side ofthe weir. When the weir was drowned out, the theoreti-cal relationship no longer held. The result was thatlevels upstream of the weir were made higher for a givenÑow rate and were maintained at a higher level forlonger after rainfall had ceased. However, the shape ofthe simulated hydrograph was very similar to theobserved data for all of the storm events. In the secondhalf of the simulation there was good agreementbetween the modelled and measured decline in riverÑows to base Ñow conditions. Observed data on the


Fig. 1. Modelled and observed Ñow at the outlet from the ADAS Rosemaund catchment from September 1992 to March 1993.

occurrence of overland Ñow were compared to theoccasions on which the model predicted this to occurfor the 1990/91 season (Table 5).5 The model generatedoverland Ñow less often than it was observed within thesingle instrumented 1-m2 plot for which data wereavailable. The model required large amounts of rainfallin one event or consecutive events to generate overlandÑow. The Ðeld in which the plot was located had a slopeof 6%, almost twice the value used in the model for thecatchment as a whole. Therefore, it is not surprisingthat this Ðeld might generate overland Ñow more oftenthan predicted for the whole catchment.

The model parameters obtained by calibration todrain Ñow were kept constant for all simulations. Thesampling programme at ADAS Rosemaund was event-based, with many samples taken over short time-periodsduring each of the seasons studied. There were manyobserved events against which to compare the modeloutput. From these, two contrasting events within asingle season have been chosen to illustrate the modelperformance in detail for the herbicide simazine. Theevent of 25 December 1990 was the Ðrst major run-o†event of the season, which occurred particularly late inthe year, 32 days after the pesticide application of4É3 kg of simazine on 23 November 1990. This eventtherefore combined a long time-lag from applicationwith the difficulties which might arise from modellingthe Ðrst substantial run-o† event of the season. The

TABLE 5Comparison between Observed and Modelled Predictions ofOccurrences of Overland Flow for Selected Events during the

1990/91 Season

OverlandÑow generation

RainfallData (mm) Observed Modelled

19 November 1990 6É5 Yes No23 November 1990 9É5 Yes No24 November 1990 4É0 Yes Yes9 December 1990 3É0 Yes No20 December 1990 5É5 Yes No25 December 1990 17É5 Yes Yes5 January 1991 6É0 Yes No8 January 1991 17É5 Yes Yes9 January 1991 6É0 Yes Yes15 March 1991 1É0 No No16 March 1991 9É0 Yes No17 March 1991 5É0 No No18 March 1991 5É0 No No20 March 1991 3É5 No No22 March 1991 0É5 No No2 April 1991 7É0 No No4 April 1991 9É5 Yes No6 April 1991 4É5 No No


event of 16 March 1991 occurred 113 days after theapplication described above, but only two or three daysafter a second simazine application of 2É2 kg to di†erentÐelds. This second application occurred at a time ofyear when the catchment was relatively wet at depth,but some drying had occurred near the surface. Figure 2shows the results of the model simulations for theDecember event. At this short temporal scale it is clearthat the modelled Ñow peaked an hour earlier and atalmost twice the magnitude of the observed data.Overall, the modelled run-o† total of 3É6 mm was some

Ðve times that of the observed (0É7 mm). The simazinesimulations were similar to the Ñow simulations, withthe peak value occurring at the correct time but at twicethe magnitude. The estimated mass of pesticide leavingthe catchment was overestimated at 7É5 g compared to1É8 g calculated from the observed data. This is asimilar ratio to that for the run-o† estimates. The fallfrom the peak value was similar in both the simulatedand the observed concentrations. It is interesting tonote that the model predicted elevated increased con-centrations in the period before samples were taken

Fig. 2. Modelled and observed Ñow and simazine concentrations at the outÑow from the Rosemaund catchment during a rainfallevent on 25 December 1990.

Fig. 3. Modelled and observed Ñow and simazine concentrations at the outÑow from the Rosemaund catchment during a rainfallevent on 16È17 March 1991.


from the stream. Similar information is given for theMarch event in Fig. 3. Here, with the catchment wetter,the simulated hydrograph was in better agreement withthe observed data, with the modelled run-o† over theevent 1É9 mm compared to 2É3 mm calculated from theobserved data. On this occasion the peak simazine con-centration was underestimated by a factor of about 4and the timing of the peak was some three hours inadvance of the observed peak. The model predicted apesticide loss of 3É3 g of simazine during the event com-pared to 11É8 g calculated from the observed data. Theunderestimation of the peak simazine concentrationmay, in part, be explained by the proximity to thestream monitoring point of the sites receiving a secondapplication two or three days before the event. In themodel, run-o† in a given hour from treated Ðelds will bediluted instantaneously by mixing within the streamwith water from untreated Ðelds in the rest of the catch-ment. In reality, water from Ðelds closer to the monitor-ing site will arrive Ðrst and any pesticide associated withthis water will not be subjected to as much dilution asassumed in the model. However, the Rosemaund catch-ment responds quickly to rainfall and this explanationshould only be valid for, perhaps, the Ðrst two hours ofan event.

The results of all the model simulations carried outfor events monitored at ADAS Rosemaund are sum-marized by pesticide in chronological order for eachmonitoring site in Table 6. Comparisons are madebetween observed and modelled data in terms of thetotal run-o†, total pesticide loss, peak pesticide concen-tration and the error in the prediction of the timing ofthe pesticide peak. The model produced good estimatesof total run-o† and pesticide losses, generally betterthan in the contrasting events discussed above. In allbut one case the estimates of peak pesticide concentra-tion were within one order of magnitude of the mea-sured values and often much better estimates weremade. A bias of two to one in favour of overestimatingobserved values was evident from the data, althoughthis seemed not to be related to either the pesticide orthe sampling point. The time of the peak concentrationwas generally not predicted well and the modelled peakalways anticipated the observed peak by several hours.In several cases this was due to an observed time seriesthat did not show a simple response of the kind seen inFigs 2 and 3. Rather, the observed pesticide concentra-tion showed either several peaks or little relationship torun-o† at all. Compounds with a range of physi-cochemical properties were used in the model testing(Table 4). It is of particular interest that the model pre-dicted concentrations of triÑuralin, which is highlysorbed litre kg~1), at least as well as for any(k

oc\ 8000

of the other compounds that were simulated. The otherhighly sorbed chemical studied was lindane (k

oc\

1100) and the model simulated the observed concentra-tions well in two of the four events studied. Simulations

of isoproturon II concentrations for the(koc

\ 130)same event give an interesting contrast. The model gen-erally overestimated the isoproturon values. For theevents at site 1 on 13 December 1989 and site 3 on 8November 1989 this was probably due to missingsamples from the initial part of the run-o† event due toautomatic-sampler failure.9 Events at ADASRosemaund have previously been shown to have ini-tially high pesticide concentrations that fall rapidlywithin hours of the start of the hydrograph.2h4 Theoverestimate of isoproturon is consistent with the modelsimulations for lindane given the di†erence in the sorp-tion properties of the chemicals (as would be expectedsince these are explicitly modelled). It follows thatlindane appears to have run o† in larger amounts thanmay have been expected compared with the isoproturonvalues. However, this is probably explicable in terms ofunderestimating the isoproturon run-o† due to theautomatic-sampler failure described above.

Other modelling approaches to estimating pesticiderun-o† which have been applied to the Rosemaundcatchment data are the SoilFug10 and SWAT11 models.These approaches were aimed at predicting the meanand peak pesticide concentrations, respectively, associ-ated with a run-o† event, but make no attempt tomodel the run-o† event in detail. The estimates made ofthe mean and peak pesticide concentrations by thesemodels were generally better than one order of magni-tude and are therefore of the same order as the esti-mates made using the model results presented in thispaper. The advantage of the model presented here isthat since it simulates at an hourly time step, it canprovide information on the duration of pesticide run-o†events. The observed event data were not sufficient todemonstrate convincingly that the model adequatelyestimated duration of events. However, Figs 2 and 3show a reasonable similarity between model and pre-dicted event durations, which at least indicates somefunctionality of the model in this respect.

3.2 Sensitivity testing

Sensitivity testing was carried out to allow some esti-mate of the importance, expressed in terms of the e†ecton model output, of the model parameters used, bothrelatively and in absolute terms. Relative importanceallows intelligent selection of input parameters, concen-trating most e†ort on those parameters that have themost e†ect. Absolute importance is similar, but thepurpose is to identify parameters where uncertainty intheir value is magniÐed through the modelling process.The approach to sensitivity testing adopted here issimple, but still allows this information to be obtained.

Each of the model parameters has been subjected toincreases and decreases in its value in line with thedegree of uncertainty that might be expected for that

Modelling

pesticiderun-o†

:modelapplications

127

TABLE 6Summary of the Results of the Simulation of Pesticide Transport at ADAS Rosemaund for a Number of Rainfall Events

Amount Rainfall T otal run-o† (mm) T otal pesticide loss (g) Predicted T imeSite applied Date of Days from amount Obs. max. max. error

Pesticide No. (kg) event application (mm) Obs. Predicted Obs. Predicted (lg litre~1) (lg litre~1) (h)

Isoproturon 1 6/3É75 13/12/89 45/27 54 4É3 4É9 3É8 5É8 5É4 15É1 123 2É1 8/11/89 8 28É5 1É0 5É2 0É1 0É4 8É4 20É2 63 10/11/89 10 10É5 0É4 2É7 0É1 0É2 13É7 11É4 83 13/12/89 45 54 4É7 8É7 0É7 0É5 8É8 16É5 90 13É8/23É4/5É2 25/12/90 75/44/32 17É5 0É7 3É3 0É6 8É9 1É8 6É7 130 5/01/91 86/55/43 9É5 1É3 1É4 0É7 2É0 5É2 1É8 200 8/01/91 89/58/46 20 Èa 4É9 6É6 6É7 2É1 200 21/02/91 132/101/89 11É5 2É8 2É3 BDLb 0É5 \0É02 0É2 È1 10É8/11É1 25/12/90 72/33 17É5 0É3 0É9 1É0 2É4 17É2 17É0 11 8/01/91 86/47 20 1É3 2É9 0É5 1É7 2É6 4É3 31 21/02/91 132/93 11É5 1É3 1É4 0É4 0É1 2É1 0É3 25 4É2 21/02/91 91 11É5 2É8 2É9 7É5 ] 10~2 7É5 ] 10~2 2É6 1É5 65 4/03/91 113 13É5 2É8 2É5 8É0 ] 10~2 5É0 ] 10~2 2É5 0É9 24

Lindane 1 3É4 13/12/89 45 54 4É3 4É9 0É1 1É1 0É3 2É9 103 1É2 8/11/89 6 28É5 1É0 5É2 6É5 ] 10~2 7É5 ] 10~2 4É5 3É6 63 10/11/89 8 10É5 0É4 2É7 2É8 ] 10~2 4É2 ] 10~2 4É1 3É6 53 13/12/89 43 54 4É7 8É7 3É5 ] 10~2 0É12 0É5 3É0 7

Simazine 0 6É9/7/6É4 24/02/89 79/9/8 13É5 1É4 1É3 58É3 71É9 68É0 90É1 30 2/03/89 79/9/8 10É5 0É8 0É2 9É8 2É2 15É7 27É2 81 5É75 24/02/89 79 13É5 3É9 0É5 0É3 1É6 1É8 20É4 120 4É5 25/12/90 32 10É5 0É7 3É6 1É8 7É5 4É1 7É5 10 5/01/91 43 9É5 1É3 1É4 1É7 2É0 1É5 2É1 70 8/01/91 46 20É5 È 4É9 È 7É7 0É7 2É8 10 21/2/91 90 11É5 2É9 2É3 1É1 1É9 0É4 0É8 40 4É5/1É2/1 16/03/91 113/3/2 9É5 2É3 1É9 11É8 3É3 15É3 4É3 3

Mecoprop 1 10É4 15/5/90 55 12 0É2 0É4 1É5 ] 10~2 0É8 ] 10~2 1É4 3É0 16Dichlorprop 1 41É6 15/5/90 55 12 0É2 0É4 1É8 ] 10~2 1É2 ] 10~2 1É0 0É6 14TriÑuralin 5 2É2 11/11/90 5 11É5 1É1 2É6 5É3 ] 10~3 8É0 ] 10~3 14É1 1É9 2

5 15/11/90 9 9 1É4 1É1 1É5 ] 10~2 0É2 ] 10~2 2É2 0É4 1

a È: No Ñow data available.b BDL: below detection limit.


parameter. Corresponding changes in model estimatesof total pesticide loss and maximum pesticide concen-trations have been quantiÐed as a percentage changerelative to the prediction arising from the originalmodel estimate. This procedure was carried out for thetwo simazine run-o† events described above. The resultsfrom the sensitivity testing for those parameters in themodel that relate to the Ñow of water1 are given inTable 7. Model predictions were most sensitive to theparameter setting the proportion of the catchment thatcould be considered to be acting as a high conductivityarea (HCA). Increasing this fraction increased the lossof pesticide predicted in the stream for both the Decem-ber and March events by more than the increase in theparameter value. The increase was due to increasedamounts of water reaching the stream during the events.The e†ect on concentration was less marked and wasonly signiÐcant for the March event which occurredsooner after the pesticide application. As less degrada-tion was likely to have taken place, more pesticidewould be available to travel by the newly increasedrapid route. Decreasing the fraction considered to be ofhigh conductivity caused a corresponding decrease inpesticide concentrations. Other than in this case, noneof the changes in parameter magnitude gave rise to agreater change (in percentage terms) in either the pre-dicted pesticide losses or maximum concentrations ineither run-o† event. Errors in the accuracy with whichindividual parameters are known will not therefore bemagniÐed though the modelling process ; in fact, in mostcases, such errors will, to some extent, be ameliorated.However, only single parameter errors have been con-sidered here, whereas, in reality, there is likely to be an

error associated with each parameter. These errors mayeither be additive to produce a greater model error thanindicated above (likely) or, by chance, may be ofopposite sign leading to a reduced error (unlikely). Ineither case, it is important to consider the interactionbetween parameters when applying this or any model.

Increasing the horizontal conductivity parameterswas the only other signiÐcant change with(SAT kh1h7)

respect to the change in the parameter value. The e†ectwas greater for the December event than for the Marchevent, indicating that the model predicts an increasedcontribution of lateral Ñow to the stream Ñow early inthe drainage season (the Ðrst drain Ñow was not initi-ated until soon before this event). Perhaps most inter-esting was the insensitivity of the model outputs tochanges in the parameters describing the extent ofmacropores through the soil proÐle This implies(CF

i).

that in this model conceptualization it is not necessaryto consider macropores explicitly because they areassumed implicitly by the introduction of the high con-ductivity zone above the drains. In this zone, the largevalue for the vertical saturated conductivity parameter

e†ectively allows rapid Ñow from the(SAT kv5h7)surface layers to the layers feeding the drains undersaturated or near saturated conditions.

The model predictions were also sensitive to thoseparameters related to the physicochemical properties ofthe pesticide (Table 8). This is because the half-life andsorption coefficient e†ectively determine how much ofthe pesticide is available for transport. This sensitivity isimportant because measurements of these parametersreported in the literature clearly show variations similarto those used in this sensitivity analysis.12 Increasing

TABLE 7Sensitivity of Model Estimates of Simazine Loss and Maximum Concentrations to Variations in hydrological ParameterValues

Change in total Change in thepesticide loss (%) maximum concentration (%)

Parameters Change 25 Dec. 1990 16 Mar. 1991 25 Dec. 1990 16 Mar. 1991

SMAX4 ]10% [4É6 [0É6 ]1É9 ]2É9SAT kh4 ]10 [4É1 [0É5 [0É4 [0É2SAT kh4 ]0É1 ]0É3 0É0 [15É3 [0É2SAT kv1h3 ]10 [63É6 [60É0 [57É3 [19É8SAT kv1h3 ]0É1 ]36É5 ]53É9 [0É8 ]37É1SAT kh1h3 ]10 ]272É3 ]124É8 ]25É9 ]76É2SAT kh1h3 ]0É1 [46É4 [49É7 [14É8 [16É2SAT kv5h7 ]10 [62É3 [37É5 [77É5 [62É4SAT kv5h7 ]0É1 ]87É4 ]1É1 ]52É0 ]1É0SAT kh5h7 ]10 ]145É5 ]30É3 ]28É0 ]60É2CF1h4 Double number of macropores ]0É4 ]7É5 ]2É8 ]0É5CF5h7 Double number of macropores ]2É1 ]2É5 ]7É1 [0É5CF1h4 No macropores in slow Ñow zone ]5É2 ]7É1 [1É0 ]3É0CF1h7 No macropores in any box ]4É0 ]7É8 [1É0 ]6É0HCA ]100É0 ]102É8 ]111É1 [8É1 ]31É4HCA [50É0 [51É8 [51É1 [6É7 [17É5


TABLE 8Sensitivity of Model Estimates of Simazine Loss and Maximum Concentrations to

Variations in Pesticide Half-Life and Soil Sorption

Change in Change inModel Change Simazine event pesticide loss maximum

parameter (%) date (%) concentration (%)

Koc

[50 25 December 1990 ]55É3 ]21É216 March 1991 ]55É2 ]64É0

]100 25 December 1990 [41É5 [47É316 March 1991 [41É8 [41É7

Half-life [50 25 December 1990 [35É8 [45É65 January 1991 [44É0 [44É38 January 1991 [47É5 [47É121 February 1991 [71É6 [72É516 March 1991 [24É3 [9É5

]100 25 December 1990 ]24É9 ]5É55 January 1991 ]37É5 ]36É28 January 1991 ]37É4 ]37É121 February 1991 ]89É5 ]80É016 March 1991 ]34É4 ]11É7

the sorption coefficient decreases the amount of pesti-cide in solution and therefore reduces the concentrationin the stream. Decreasing the sorption coefficient hasthe opposite e†ect. Variability in the value of the degra-dation half-life had an e†ect on model predictions thatbecomes greater the longer the interval between theapplication date and the event of interest. This e†ect isclearly seen in Table 8 where data on all Ðve run-o†events following simazine application are shown. TheMarch event shows the least change in model outputdue to the close proximity of this event to a secondsimazine application.

Clearly the model is sensitive to a small range ofparameters, although, in general, the sensitivity dis-played by the model is less than the variation imposedon the parameter values. Those parameters that need tobe known well are the extent of the high conductivityarea and the properties of the pesticide. The formerrequires some knowledge of the extent of drainagewithin the catchment, which may be available from theoriginal drainage plans or from local knowledge. Pesti-cide properties are certainly best measured using soilsamples collected for the catchment being studied. As aÐrst estimate, the parameters could be extracted fromappropriate reference works.13

4 CONCLUSIONS

A pesticide run-o† model has been tested againstobserved data collected as part of a pesticide run-o†study at ADAS Rosemaund. The model provided goodestimates of observed data (within an order ofmagnitude) for both the total pesticide loss and peakpesticide concentrations occurring in individual rainfall

events. The model, therefore, provides the basis of a toolfor estimating pesticide loss from agricultural catch-ments to surface waters. However, the model has beenapplied to only one catchment and the transferability toother catchments would need to be proven before itcould be considered for wider use.

ACKNOWLEDGEMENTS

The funding for this work was provided by the NationalRivers Authority (now part of the Environment Agency)and the Natural Environment Research Council whosesupport is gratefully acknowledged. The author wouldalso like to thank the Centre for Environment, Fisheriesand Aquaculture Science laboratory at Burnham-on-Crouch for allowing the use of their data in Table 4 andthe sta† at ADAS Rosemaund for providing the pesti-cide application data.

REFERENCES

1. Williams, R. J., Modelling pesticide run-o† to surfacewaters, Part I : Model theory and development, Pestic.Sci., 54 (1998) 113È120.

2. Williams, R. J., Bird., S. C. & Clare, R. W., Simazine con-centrations in a stream draining an agricultural catch-ment. J. IW EM, 6 (1991) 80È84.

3. Matthiessen, P., Allchin, C., Williams, R. J., Bird, S. C.,Brooke, D. & Glendinning, P. J., The translocation ofsome herbicides between soil and water in a small catch-ment. J. IW EM, 6 (1992) 496È504.

4. Williams, R. J., Brooke, D. N., Matthiessen, P., Mills, M.,Turnbull, A. & Harrison, R. M., Pesticide transport tosurface waters within an agricultural catchment. J.IW EM, 9 (1995) 72È81.


5. Bell. J. P., Abbott, C. L. & Batchelor, C. H., The Soilhydrology of “LonglandsÏ, ADAS Rosemaund, Here-fordshire : Second Interim ReportÈCrop Year 1990/91. InPesticide Runo† Study at Rosemaund, Report of Y ears 2 to5, ed. C. M. Hack. ADAS, Oxford, UK, 1994.

6. Carter, A. D. & Beard, G. R., Interim report on the soilwater sampling and soil characterization programmewithin a small catchment at Rosemaund EHF (1990È1991). Soil Survey and Land Research Centre, Silsoe,Beds., UK, unpublished.

7. Penman, H. L., Natural evaporation from open water,bare soil and grass, Proc. R. Soc. L ondon, Ser. A., 193(1948) 120È46.

8. Armstrong, A. C., Portwood, A. M., Leeds-Harrison, P.B., Harris, G. L. & Catt, J. A., The validation of pesticideleaching models, Pestic. Sci., 48 (1996) 47È55.

9. Williams, R. J., Brooke, D. N., Glendinning, P. J., Mat-thiessen, P., Mills, M. J. & Turnbull, A., Measurement and

modelling of pesticide residues at Rosemaund Farm. InProc. Brighton Crop Prot. Conf.ÈW eeds, 1991, BritishCrop Protection Council, Farnham, Surrey, 2 (1991)507È14.

10. DiGaurdo, A., Williams, R. J., Matthiessen, P., Brooke, D.N. & Calamari, D., Simulation of pesticide runo† atRosemaund Farm (UK) using the SoilFug model. Environ.Sci. Pollut. Res., 1 (1994) 151È60.

11. Brown, C. D. & Hollis, J. M., SWATÈA Semi-empiricalmodel to predict concentrations of pesticides enteringsurface water from agricultural land. Pestic Sci., 47 (1996)41È50.

12. Wauchope, R. D., Buttler, T. M., Hornsby, A. G., Augus-tin Beckers, P. W. M. & Burt, J. P., The SCS/ARS/CESpesticide properties database for environmental decisionmaking. Rev. Environ. Contam. T oxicol., 123 (1992) 1È164.

13. Tomlin, C. (ed.), T he Pesticide Manual, 10th edn. BritishCrop Protection Council, Farnham, UK, 1994.

modelling pesticide run-off to surface waters. part ii: model application

Documents