WATER RESOURCES RESEARCH, VOL. 36, NO. 10, PAGES 3077-3080, OCTOBER 2000 A formula for predictingadvection-dispersion in the vadosezone at unevendrainage intervals Vincent J. Bidwell Lincoln Environmental,Lincoln University,Canterbury,New Zealand Abstract. A computationalmethodis presentedfor predictingone-dimensional advective-dispersive solutetransportwith linear equilibriumsorption,for complexsolute input historiesassociated with unevenintervalsof transportingwater flux. The suggested applicationis monitoringof leachedsolutesfrom land use and near-surface transformations, which are assumedto be transportedthroughthe vadosezone to the underlyinggroundwaterwithout further degradation.Real-time forecastsof solute concentrationat the groundwatersurfaceprovidea smoothfeedbacksignalto operational managementof land use for protectionof groundwater,as a surrogatefor groundwater monitoringwhich hasunacceptabletransportlag. The conceptualbasisis the seriesof mixingcellsas an analoguefor advection-dispersion, in the form of a linear systemfor which cumulativepore-waterdrainageis the continuousindex. This mixing-cellformula is unconditionallystablefor any interval of drainageand assuresconservationof solutemass. A demonstrationexampleis presented. 1. Introduction When landuseis consideredasa nonpointsourceof groundwater contamination,the effectsmay be estimated,for the purposesof operationalmanagement,by assumingthat solute whichhasbeentransportedbeyondtransformationin the near- tion generatedfrom beneaththe soillayer.The purposeof this computationis to be ableto forecast,in real time, the ultimate effect on groundwaterqualityof nonpointsourcecontamination generatedby land usessuchasland treatmentof wasteor applicationof fertilizersandpesticides for agriculturalproduction. Where leachatequality and quantityare monitored,or estimatedfrom a mathematicalmodel, these data together with the vadosezone transport computationcan provide a feedbacksignalfor operationalmanagementof land use to achievegroundwaterprotection. surfacezone eventuallyreachesgroundwaterwithout further degradation.Examplesof thiskind of solutebehaviorare some pesticidesand their degradationproducts,which have been leachedbelowthe organiclayersof the soil,and nitrate anions leachedbeyondthe influenceof the plant root zone.The subsequenttransportof thesecontaminants throughthe unsaturated vadosezone to the groundwatersurfacemay be consid- 2. Transfer Function Approach ered asan advective-dispersive process, whichattenuatespeaks One-dimensionalsolutetransportthroughthe unsaturated in solute concentration but otherwise conserves solute mass. subsurfacezone can be representedin terms of a transfer This processdescriptioncan also include linear equilibrium function[Jury,1982]whichspecifiesthe distributionof passage sorption(onto solidsurfacesor into immobilewater) by ap- times to a particulardepth. This conceptis the same as the propriatescalingof the "watercontent"parameterwith a "re- transferfunctionof a linearsystem, whichcanbedefinedasthe tardation" coefficient. Leachingof contaminantsbelow the near-surfacezone is climaticallydriven, and inputsof contaminantsto the vadose zone may be consideredas intermittentamountsof drainage from the soilprofile,eachassociated with a massof contaminant in solution or suspension.The transport time to the groundwatersurfacemaybe severalyears,dependingon depth to groundwaterand geologicalproperties.More significantly, the transportprocessis drivenby the verticalwater flux in the vadosezone, and transporttime is therefore the cumulative effectof a seriesof drainageamountsgeneratedby climateand soil-plantproperties.Thusvadosezone transportis a stochastic processin termsof time, but in thisnote it will be presented as a deterministicprocessin termsof cumulativedrainage. The aim of this note is to providea methodfor estimating the soluteflux enteringthe groundwatersurfacein responseto eachdrainageamountand associated contaminantconcentra- outputresponseto a unit inputimpulse.The time indexcanbe replacedby cumulativedrainageI for field-scalerepresentation of solutetransportundertransientwater flow [Juryet al., 1990]. Numerical solutions[Wierenga,1977] show that this transformation allows unsteady advective-dispersive solute transport to be adequatelyrepresentedby the steady state solution.This characteristicdependson the dispersioncoefficient D being linearly related to pore-watervelocity V by a value of dispersivity• which is invariant with water content. The water content 0 for use with the steadystate solutionis suggested [Wierenga, 1977]to be near the valuefor field capacity.Other numericalresults[De Smedtand Wierenga,1978] supportthe useof a depth-averaged valueof 0 for nonuniform water content. The time-basedtransferfunctionf(t, L), at depth L, for advection-dispersion in soil under steadyflow conditions,is [Juryand Roth, 1990,example2.2] Copyright2000 by the AmericanGeophysicalUnion. Paper number2000WR900192. f(t,L)=tx/4,rDt exp 4Dt 1' 0043-1397/00/2000WR900192509.00 3077 (1) 3078 BIDWELL: TECHNICAL NOTE Asa probability density function (pdf)of traveltimet, (1) has meanL/V andvariance 2DL/V3. The pdff(t, L) canbe transformed to a cumulative drainagebasisf(I, L) by means of the equivalence f(t, L)d! = f(I, L)dI, substituting D = trationCin to thefirstcell.Thisprocedure canbe developed fromthegeneralsolution of theordinarydifferential equation of r mixingcellsin series,eachcell with residencetime T. The generalformof vanderMolen[1973,equation(35)]is then XV, and then replacingthe travel distanceVt with I/0 so that =Cin +[C(0) --Cin ]exp 1+m• øm!Tm' L exp [(L-I/O) 2]' (2) Cr(t) f(I,L)=I x/4•rXI/O 4XI/O (3) profiles,definedby Thepdf(2) hasmeanL 0 andvariance 2LX02andrepresentsFor nonuniforminitial concentration of superposition leadsto sequential applithe transferfunctionof solutetransported by advection with Cr(0),theprinciple of the hydrodynamic dispersion. Thereforedispersion is assumed to cationof (3) to the seriesof cellsfor computation profileCr(t) at time t, whichcanbe illustrated be determined by the distance I/0 traveledby the cumulative concentration water flux and is independentof the transientflow behavior. The transferfunction(2) completelycharacterizes advective-dispersive solutetransport,but it isnot a convenient math- ematicalform for prediction,especially for complexsolute input historiesand unevenintervalsof drainage.The next sectiondescribes an analogous linear systemwhichis more convenientfor computational purposes. 3. Mixing-Cells Analogue Solutetransportthrougha seriesof cells,within each of for the first three cells as Cl(t)=Cin+[Cl(O)--Cin] exp (--•) c2(t) --C,n+ [c•(O) --Cin] exp (--•) (1+•) +[c2(0) - el(0)] exp (-•) (4) c3(t) = Cin q-[Ci(0 ) --Cin ] exp whichperfectmixingoccurs, isa concept witha longhistory of dispersion modeling in hydrology [e.g.,Bear,1969;Bajracharya andBarry,1994].Thismodelisoftenpresented asan approximation to the finite difference solution of the advection- 1+ • + 2.• +[c2(0) - c•(O)] exp (--•) (1+•) +[c3(0) - c2(0)] exp (•) . dispersion equation, anddevelopment hasfocused on problemsof numerical dispersion andstabilityrelatedto thesizeof the spaceand time steps[Bajracharya andBarry,1992].In contrastto thesediscrete-time models,seriesof mixingcells The generalform of (4) canbe rearrangedas canalsobedescribed in continuous timebysystems of ordinary differential equationswhich include solute transformations lumpedatthescaleofeachcell[Sardin etal.,1991].Themixing cellvolumes arepredetermined sothattheirdynamic response simulates advective-dispersive transport, andthesetofcoupled differentialequationsfor transportand transformation are solvedby standard methods for complex linearsystems [e.g., Bidwell,1999]. (3), (4), and (5) incorporate the imMuch of the mathematicsof mixing-cellsystemshas Thesegeneralsolutions that the inputfluxconcentration c•, is conemergedfrom chemicalengineering andthe theoryof chro- plicitassumption matography.Van der Molen [1956] applied the chromato- stant over the time inte•al { 0 t }. Bidwell[1999]developed a form of the m•ng cellmodel, graphic theoryof Glueckauf[1949], whichincludes linearequilinear,equilibrium and libriumsorption,to leachingof salinesoils.He describes the with flowbasisI, whichincorporates solutetransformation. For linearequilibrium transportprocess in termsof soil-waterflow (I), appliesthe nonequilibrium, eachcell hasa residentvolumeper unit crosssecprincipleof superposition of analytical solutions for complex processes, tion [L] equal to 2XOR(instead of residence timeT), where salinity profiles, andestimates theparameter "theoretical plate Cr(t) =exp • 5 m•rmCr-m(O) ()•r-l( tm )• +{1-exp (•)[••o (mt•;m)]}Cm ß (5) thickness" from measurements of chloride ion concentration. the retardation coefficient R is the ratio of total mass to dissolved mass on a soil volume basis.This is the same as the van Thisapproach hasessentially thesametheoretical basisas(2), involves onlyequilibrium and the theoreticalplatethickness is equivalent to 2•. In a derMolen[1973]modelif theprocess transfer of solute within the water phase, for which R = 1. For subsequent monograph [vanderMolen,1973]thisdimension is ontosolidsurfaces, for whichR > 1, theproduct R0 called"effective mixinglength,"andthe soilprofileis consid- sorption asa singleparameter for calibration pureredas a seriesof mixingreservoirs for whichthe dynamic canbe considered Solutemass response is shownto simulatecloselythe analyticalsolutions posesif 0 andR are not knownindependently. balance of a cell is described by a differential equation with fromGlueckauf[1949]. Thiscomparison isachieved byimplicdrainage I asthecont•uous index(•steadoft•e t): itlyequating thetimebasist andresidence timeT of a mixing cumulative reservoir to theflowbasis I andthewatercontent 2X0,respecdc tively,of a theoreticalplate.This equivalence is the sameas 2XOR • + c = C,n, (6) that between(1) and (2). A computational methodisalsoprovided for calculating the where c•, is the influent soluteconcentration.A seriesof n soluteconcentration Cr(t) in cellr at timet, giventhe initial cells,eachdescribed by (6), is a lineardynamicsystem with a concentration c(0) in eachof the cellsandinputfluxconcen- transferfunctionwhichis the gammadistribution BIDWELL: f(I, n)= (n- 1)!(2AOR) nexp-2XOR' TECHNICAL (7) Thepdf(7) hasmean2nXORandvariance 4nX2(0R)2.If the NOTE E 70 -a- Monitoredat 0.7 rn c 60 ._o -e- Forecast at 15 m • 50 observationdepthL is representedby n cells,then L = 2nX, (• the meanis L OR, andvarianceis 2LA(0R)2. The transfer o 30 functionsof the advection-dispersion process(2) and the mixing-cellmodel (7) havethe samemean and variancefor n = L/2A and R = 1. The matchbetweenthe two functions(2) and (7), over the full rangeof drainageI, improveswith increasingnumber of cellsand is very closefor n greaterthan 40 z & 20 z 0 ¾•_ _,i . i- ._i -l•r.-,=,•li i i i = i.. 0 100 200 300 400 500 600 , , 700 800 900 Cumulativedrainage(mm) about 10. Figure 1. 4. 3079 Forecasts of nitrate-N concentration at the groundwatersurface(15 m depth), associated with monitored leachateconcentrations from 0.7-m-deeplysimeters,as demonstratedfor water content 0 = 0.13 and dispersivityX - Computational Form For practical monitoringof leachateinput to the vadose 0.88 m. zonefrom land treatmentof wastewater,for example,drainage collectedfrom monolithsoil lysimetersmay be measuredand sampledfor bulk solute concentrationat time intervals of o(k) m about i month. The volume collectedat the kth monitoring bin(k)=m! (12) eventis expressed asa drainageintervaldI(k), whichdepends on soil-waterresponseto climateand irrigationevents,and the into (10) and calculatingthe coefficientsbm(k) from the secorresponding bulk soluteconcentrationci,,(k) is considered quence to be constantduring dI(k). By substitutingthe following bo(k) = 1 equivalences into (5) b(k) = .(k)b0(k) ½r(k) = Cr(t) (13) ½r(k -- 1) • ½r(O) ½in(k)•½in (8) bin(k) - dI(k) = t m bm-l(k) ß ß 2XOR = T ß and writing d/(k) a(k)= 2XOR' (9) The valueof the bm(k) coefficients mayhaveverysmallvalues as m increases,dependingon the value of a(k), and truncation of the process(13) may be considered.However, the author prefersto completethe full sequenceand thus avoid havingto test for lossof solutemass. (5) then becomes 5. ½r_m(k1)) ] Demonstration Example An exampleof the use of the computationalmethod (9)cr(k ) = exp[-a(k)] m! (13) is demonstratedwith a seriesof flux-averagednitrate-N concentrationdata ci,,(k) and averagedrainagefluxesdI(k), collectedfrom four monolithsoil lysimetersusedfor monitor(lO) ing a land treatmentsitefor meat-processing wastewater[Bidwell, 2000]. These data were collected at approximately monthlytime intervalsduringa 2-year trial for which the avThe formula (10) is unconditionally stablefor any magnitude of drainageinterval and assuresconservationof solutemass. eragedrainagewas400 mm/yr.The data (Figure 1) includea significant"contaminationevent" when the waste treatment The number of mixingcellsn requiredfor a model is deterarea was temporarilyoverloaded.The transportparameters0 mined from the dispersivityX and the depthL at which pre- and X have not been determined for the vadose zone beneath diction is desired,accordingto the site, and for the purposeof the presentdemonstration, L valueshavebeenusedfrom measurements byEngesgaard et al. n =2X' (11) [1996] for the unsaturatedzone of a sandyaquifer receiving estimatedinfiltration of about 450 mm/yr. Their resultsfor transportmodel to the tritium The number of mixingcellsn, from (11), may be large, for fitting an advective-dispersive example,n - 100 for depth to groundwaterL - 20 m and profile in the --•20-m-deepvadosezone providevaluesof 0 dispersivity X - 100mm.When the summationindexm in (10) 0.13 and X = 0.88 m. For the presentdemonstrationthe depth has these large values,evaluationof the ratio a(k)m/m! by to groundwaterwas set to 15 m. It was assumedthat for nitrate-N, R - 1, and there was no meansof softwarelibraryfunctionsmaycausenumericalproblems.These terms are better calculatedby substituting N transformationduring advective-dispersive transport from + 1- exp [-a(k)] m• ø m! ½in(k)' = 3080 BIDWELL: TECHNICAL NOTE beneaththe soil profile, at the 0.7-m monitoringdepth of the countsfor residentsoluteconcentrationthroughoutthe vadose lysimeter,downthe remainingL = 14.3 m to the groundwater zone depth. However, the parametersOR and X are lumped surface. The initial nitrate-N concentration in the vadose zone valuesfor the depthof observation,andthe real distributionof was set to the flux-averagedconcentrationfor the first 12 concentration with depthmaybe different. The computationalprocedureis intended for operational months, of4.4g/m3.Foreachinputof drainage fluxdI(k) with averageconcentration Cin(k), the concentration profilecr(k) monitoring and management rather than as a conceptual was computed.Then a forecastedprofile c'r(k) was obtained model for scientificinvestigationand can supporta rangeof to processing monitoringdata.The examby applyingan additionalflux incrementdI' (k), at a concen- feasibleapproaches tration c•,, equal to the averagesolute concentrationin the ple in this note demonstratesthe use of forecastedcontamimixingcells,to transportci,,(k) to the groundwater surface.In nant effect at groundwaterlevel for supportingoperational managementdecisions,whereasthe actual contaminantcontermsof (9) and (10) thiscomputationis centrationin the groundwatermay be the result of land-use dI' (k) = L O - dI(k) eventsseveralyearsprevious. c = 2 c(a- Acknowledgments. This note is basedon researchinto groundwater protectionfundedby the New Zealand Foundationfor Research, Scienceand Technology,under contractLVL805. An anonymousreviewer alerted me to the previouscontributionof W. van der Molen. r=l dI'(k) a'(k)= 2X0 (14) References ølt (k)m cr_m(k ) m! c;(k)= exp[-a'(k)] Bajracharya,K., and D. A. Barry, Mixing cell modelsfor nonequilibrium singlespeciesadsorptionand transport,WaterResour.Res.,29, _- 1405-1413, 1992. olt(k) rn + 1 - exp[- a'(k)] m! cIn(k). = Bajracharya,K., andD. A. Barry,Note on commonmixingcellmodels, J. Hydral., 153, 189-214, 1994. Bear, J., Hydrodynamicdispersion,in Flow ThroughPorousMedia, edited by R. J. M. De Wiest, chap.4, pp. 121-124,Academic,San Diego, Calif., 1969. Bidwell,V. J., State-space mixingcell modelof unsteadysolutetransport in unsaturated soil,Environ.Model.Software,14(2-3), 161-169, By meansof (14), the estimatedeffect at groundwaterlevel, c,,(k), can be computedon a real-timebasisfor eachsolute 1999. flux input monitoredat the baseof the surfacesoil.The com- Bidwell,V. J., Decisionsupportfor protectinggroundwaterfrom land treatment of wastewater,Environmetrics, in press,2000. putation takes into accountadvective-dispersive transportof the new input and the "background"concentrationprofile De Smedt,F., and P. J. Wierenga,Solutetransportthroughsoilwith nonuniform water content, Soil Sci. Sac.Am. J., 42, 7-10, 1978. generatedby the previousinput history.Figure 1 showsthe Engesgaard,P., K. H. Jensen,J. Molson,E. O. Frind, and H. Olsen, input data set and the corresponding forecastsat groundwater Large-scaledispersionin a sandyaquifer:Simulationof subsurface levelfor the parametersspecifiedfor the demonstration.These transportof tritium, WaterResour.Res.,32, 3253-3266, 1996. ! forecastedvalues can be used, in real time, as feedback infor- mationfor operationalmanagementof the land treatmentsite. Groundwaterlevel forecastsprovide a more realistic signal than lysimeterdata in terms of actual effect on groundwater, and the smoothingeffectof the transportprocessimprovesthe signal-to-noise ratio. Monitoringthe actualgroundwatercontaminationis not usefulfor operationalmanagementbecause of the long transportlag,whichfor the presentdemonstration is about 4.6 years. 6. Discussion invariant data, J. Chem. Sac., 152, 3280-3285, 1949. Jury, W. A., Simulationof solutetransportusinga transferfunction model, Water Resour.Res., 18, 363-368, 1982. Jury, W. A., and K. Roth, TransferFunctionsand SoluteMovement ThroughSoil, 226 pp., BirkhauserVerlag, Basel,Switzerland,1990. Jury,W. A., J. S. Dyson,and G. L. Butters,Transferfunctionmodelof field-scalesolutetransportundertransientwater flow,SoilSci.Sac. Am. J., 54, 327-332, 1990. Sardin,M., D. Schweich,F. J. Leij, and M. T. Van Genuchten,Modeling the nonequilibriumtransportof linearlyinteractingsolutesin porousmedia:A review,WaterResour.Res.,27, 2287-2307,1991. Van der Molen, W. H., Desalinization of saline soils as a column The methodologydescribedin this note is for estimationof one-dimensional advective-dispersive transportin porousmedia, in responseto complexdata seriesof solute inputs at unevenintervalsof transportingwater flux. Dispersionis assumedto be linearly dependenton pore-watervelocitybut otherwise Glueckauf,E., Theoryof chromatography, part VI, Precisionmeasurements of adsorptionand exchangeisothermsfrom column-elution with water content and is therefore com- process,Soil Sci.,81, 19-27, 1956. Van der Molen, W. H., Salt balance and leachingrequirements,in DrainagePrinciplesand Practice,vol. II, Publ. 16, pp. 60-100, Int. Inst. for Land Reclam.and Impr., Wageningen,Netherlands,1973. Wierenga,P. J., Solutedistributionprofilescomputedwith steady-state and transient water models, Soil Sci. Sac. Am. J., 41, 1050-1055, 1977. pletelydescribedby the dispersivity X. Linear equilibriumsorpV. J. Bidwell, Lincoln Environmental, P.O. Box 84, Lincoln Univertion processes can be includedby meansof a retardationco- sity,Canterbury8150,New Zealand.(bidwellv@lincøln'ac'nz) efficientR appliedto the transportingwater fraction 0 or by calibration of the combinedparameter OR. In conceptual (ReceivedMay 4, 1999;revisedJune 16, 2000; terms, the mixing-cellanalogueof advection-dispersion ac- acceptedJune 19, 2000.)