2

3

4

5

67

11121314151617

Journal of Hydrology (2006) xxx, xxx–xxx

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

ava i lab le a t www.sc iencedi rec t . com

journal homepage: www.elsevier .com/ locate / jhydrol

OO

FTowards an improved understanding of thenitrate dynamics in lowland, permeableriver-systems: Applications of INCA-N

RAndrew J. Wade *, D. Butterfield, P.G. Whitehead

P

Aquatic Environments Research Centre, School of Human and Environmental Sciences, The University of Reading,Whiteknights, Reading RG6 6AB, UK DAccepted 12 April 2006 E

NC

OR

REC

T

Summary The Integrated Catchment Model of Nitrogen (INCA-N) was applied to the Lambournand Pang river-systems to integrate current process-knowledge and available-data to test twohypotheses and thereby determine the key factors and processes controlling the movement ofnitrate at the catchment-scale in lowland, permeable river-systems: (i) that the in-streamnitrate concentrations were controlled by two end-members only: groundwater and soil-water,and (ii) that the groundwater was the key store of nitrate in these river-systems. Neitherhypothesis was proved true or false. Due to equifinality in the model structure and parametersat least two alternative models provided viable explanations for the observed in-stream nitrateconcentrations. One model demonstrated that the seasonal-pattern in the stream-water nitrateconcentrations was controlled mainly by the mixing of ground- and soil-water inputs. An alter-native model demonstrated that in-stream processes were important. It is hoped further mea-surements of nitrate concentrations made in the catchment soil- and ground-water andin-stream may constrain the model and help determine the correct structure, though otherrecent studies suggest that these data may serve only to highlight the heterogeneity of the sys-tem. Thus when making model-based assessments and forecasts it is recommend that all pos-sible models are used, and the range of forecasts compared. In this study both models suggestthat cereal production contributed approximately 50% the simulated in-stream nitrate load inthe two catchments, and the point-source contribution to the in-stream load was minimal.

�c 2006 Elsevier B.V. All rights reserved.

KEYWORDSWater quality;Pollution;Nitrogen;Nitrate;Model;Uncertainty

0d

U

18

1920

022-1694/$ - see front matter �c 2006 Elsevier B.V. All rights reservedoi:10.1016/j.jhydrol.2006.04.023

* Corresponding author. Fax: +44 118 975 5865.E-mail address: a.j.wade@reading.ac.uk (A.J. Wade).

Introduction

The Water Framework Directive (2000/60/EC) requires thatall European Union water-bodies, that are not heavily mod-

.

C

2122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182

838485868788899091929394

9596979899100101102103104105106107108109

110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143

2 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

UN

CO

RR

E

ified, must achieve good ecological status by 2015. Thus,mathematical models that simulate the storage and trans-port of pollutant inputs through the soils and groundwaterof a river-system are required to understand and quantifythe likely response-time between a pollutant entering thesystem and its subsequent emergence in the surface-waters,and thereby determine what management options are viableto improve river ecology within the time-scale of the Direc-tive. However, water quality modelling is difficult becauseof an inability to scale point-flux measurements typical ofroutine monitoring to a value representative of an areawhich is generally required in models (Beven, 1993). Thisinability prevents the identification of the model which bestrepresents a system since the optimum model structure andparameter set cannot be found (Beven, 2002).

The Cretaceous-chalk rivers of southern England are amajor national resource providing water for drinking andagriculture. Moreover, they provide key trout fisheries andare species-rich in terms of plants, macro-invertebratesand fish (Department of the Environment, 1995). Giventhe different demands on these resources and their impor-tance in the provision of water for the lowland United King-dom, a scientific programme, The Lowland CatchmentResearch (LOCAR) initiative was instigated to investigatethe hydrological, hydro-chemical and ecological functioningof lowland, permeable river-systems in the UK commensu-rate with the spatial-scale of catchment management(Wheater and Peach, 2004). In particular, an ambitiousand highly detailed monitoring programme is in progressmeasuring water quantity and quality in the soils, ground-water and riparian zone of the catchments in addition toin-stream. These data are being brought together with his-toric long-term data collected through routine monitoringby the Environment Agency and Thames Water plc. As such,the LOCAR initiative provides an example of a science pro-gramme that integrates the different aspects of science re-quired to understand how pollutants and climate affectriver ecology. The data available in LOCAR provide anopportunity to develop and test new models of hydro-chem-ical functioning, providing information that can be usedpotentially to specify and test the structure of models toa greater extent than water chemistry data alone.

Nitrate is an important water quality determinand sinceit is a key plant nutrient; though over-enrichment of river-systems with nitrate can lead to problems of eutrophica-tion, acidification and reduced biodiversity (Heathwaiteet al., 1993). As such, there is a need to understand thetransport, storage and fate of nitrate in lowland river-sys-tems which are subject to elevated stream-water concen-trations due to inputs from effluent, farming andatmospheric deposition. Initial analysis of the surface-waterchemistry collected in the Pang and Lambourn riversshowed an increase in stream-water nitrate concentrationswith flow in these two systems. This was explained by an in-creased input of soil-water under high-flow conditions whichwas enriched in nitrate relative to the groundwater; point-sources were relatively unimportant (Neal et al., 2004a).In the same work, the authors suggest that in-stream pro-cesses may control the observed nitrate dynamics to someextent, but the relative importance of the ground- andsoil-water contributions and in-stream processing was notdetermined. The two river-systems were also observed to

TED

PR

OO

F

be sinks for nitrate (Neal et al., 2004b), with a correlationbetween groundwater nitrate concentrations and groundwa-ter age observed (Gooddy et al., this issue). The latter worksuggests an increase in groundwater nitrate concentrationbetween 1950 and 1980, which corresponds to an increasein fertiliser use on arable crops within the UK from approx-imately 90 to 140 kg N ha�1 a�1 (Vinten and Smith, 1993).The belief is that fertiliser use in Chalk catchments has cre-ated a store of nitrate in the groundwater (Wellings andBell, 1980). Given these issues it is proposed that the keyscience questions relating to nitrate in Chalk river-systemsare:

1. Is arable agriculture the main cause of nitrate concentra-tions above those which might be expected in theabsence of a pollutant effect?

2. If the river-systems are acting as a sink for nitrogen, isthe main store in the groundwater?

3. When will the bulk of the nitrogen stored be released tothe surface-waters?

4. For how long will the nitrogen be released?5. What are the ecological consequences of elevated nitro-

gen concentrations and loads in-stream?6. What management options are available to mitigate or

reverse any adverse ecological impacts?7. Can a reliable model be produced that can integrate all

the measurements made in LOCAR to explain theobserved in-stream chemistry?

The Integrated Catchment Model of Nitrogen (INCA-N) isa model of the factors and processes that control nitrate ex-port from the land, and the subsequent in-stream processing(Whitehead et al., 1998; Wade et al., 2002; Wade and Neal,2004); though this may be one of an infinite set of possiblemodels (Beven, in press). Despite uncertainty regardingmodel-structure and parameters, the INCA-N model has al-ready proved successful in simulating the flow and nitrateconcentrations observed in the River Kennet, a major riv-er-system draining the Cretaceous chalk in southern England(Wade et al., 2002; Limbrick et al., 2000). Moreover, themodel has been subject to a sensitivity-analysis based onits ability to simulate observed flow and stream-water ni-trate and ammonium concentrations from the River Kennetand the model structure was found to be reasonable for thistask though problems of identifying the internal structure ofthe soil-component were recognised (McIntyre et al., inpress).

The aim of the work presented is to apply INCA-N to theLOCAR Lambourn and Pang river-systems to test twohypotheses: (i) that the in-stream nitrate concentrationsare controlled by two end-members only: groundwater andsoil-water, and (ii) that the groundwater is the key storefor nitrate in these river-systems. If the first hypothesis istrue, the implication is that the in-stream nitrate dynamicsare controlled alone by the relative inputs of soil- andgroundwaters: in-stream processing and point or point-dif-fuse sources are unimportant (Neal et al., 2004a). If the sec-ond hypothesis is true, then nitrate transported by lateralthrough-flow from the soil store or by over-land flow is ofsecondary importance to the input of nitrate from ground-water in controlling the in-stream nitrogen concentrationsand loads. In addition, the applications of the model will

C

144145146147148149150151152153154155156157158159160161162

163

164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202

203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254

255

256257258259260261

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 3

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

UN

CO

RR

E

provide an assessment of the relative contributions fromdiffuse and point sources to the in-stream nitrate load,and the mass of nitrate removed from the stream by a com-bination of plant uptake and denitrification. Thus, it ishoped that this work will provide a modelling assessmentof the factors and processes controlling the in-streamdynamics of nitrate concentrations at the river-systemscale, and directly address the first two of the proposed sci-ence questions. The model was applied using manual cali-bration rather than automated uncertainty-analysistechniques to explore the benefits of using typically avail-able hydrological and land management data to explainthe in-stream nitrate concentrations. The results were alsoused to provide comment on the remaining questions posedin this Introduction. During the course of this study it wasfound that at least two alternative model configurationscould be identified to explain the observed stream-water ni-trate concentrations; in this paper, the term ‘model’ refersto an alternative structure and/or parameter-set of INCA-N.

Study area and data resource

The Lambourn and Pang rivers are described in detail else-where (Neal et al., 2004b). Briefly, the two rivers are adja-cent, though the Lambourn (261 km2) drains to the RiverKennet and the Pang (174 km2) to the River Thames(Fig. 1). Both river-systems are predominantly underlainby Cretaceous Chalk, though Palaeogene deposits (ReadingBeds, London Clay and alluvium and drift) cover the lowerreaches of the Pang from Frilsham downstream to Pang-bourne. Both catchments are rural, with the main land-usesbeing cereal production and pasture. There are no majortowns in either catchment, though there are villages inthe Lambourn at Lambourn, East Shefford, Boxford and Bag-nor. The catchment also includes the suburban fringe ofNewbury near the confluence with the Kennet. In the Pang,there are villages at Compton, Frilsham, Bradfield, Tid-marsh and Pangbourne. Small sewage treatment works(STWs) serve many of these villages.

The long-term annual-rainfall in the Lambourn is 739 mmwhich produces 231 mm (1.71 m3 s�1) of runoff; in the Pang(at Pangbourne) the rainfall and runoff are 702 and 114 mm(0.62 m3 s�1) respectively (Marsh and Lees, 2003). The WestBerkshire Groundwater Scheme reduces flow in both river-systems except under low-flow conditions when the bore-holes are used to augment the river-flow. Given the low-land, permeable nature of both systems, the in-streamflow dynamics are dominated by the groundwater contribu-tion (Bradford, 2002).

The Environment Agency maintain and collect data fromone flow-gauge on the Lambourn at Shaw and three on thePang at Frilsham, Bucklebury and Pangbourne; continuousdata are available from these gauges (Fig. 1a). The waterquality has been monitored at three sites on the Lambourn(East Shefford, Boxford and Shaw) and at four sites on thePang (Frilsham, Bucklebury, Rotten Row and Tidmarsh).The measurements, at all sites except Tidmarsh, were takenweekly from April 2002 to December 2003 for a broad suiteof determinands including nitrate and ammonium. Weeklymeasurements for the Pang at Tidmarsh extend from July1997 to December 2003. A detailed description of the chem-

TED

PR

OO

F

ical monitoring sites and the methodologies used is given inNeal et al. (2004b).

Data describing the land-use and livestock numbers ofthe two catchments were derived from the Department ofthe Environment, Food and Rural Affairs (DEFRA) Agricul-tural Census for 2000, based on farm-scale returns. Due tothe Non-disclosure Act whereby the identities of individualfarmers were protected, the data were treated before re-lease to third parties and therefore data for some areas ofLambourn and Pang were missing. Furthermore, since thereturns were based on farms, it was difficult to translatethe data to sub-catchments as, although the farm mightbe inside the catchment, the farm land might not. A de-tailed description of how the data were used in this studyis provided in the ‘‘Model set-up and calibration’’ section.

Inputs of nitrogen from fertiliser and livestock were de-rived from the inputs assumed in recent applications ofthe Export Co-efficient Model (ECM, Johnes, 1996; Johnesand Butterfield, 2003), and the timing of the fertiliser appli-cations was assumed to be the same as those observed inthe survey of fertiliser practice from the River Ant (Johneset al., 2003). The latter was a pragmatic response to datascarcity, and whilst the exact timing of applications is unli-kely to be the same in all systems, all three catchments aredominated by cereal production and located in lowland Eng-land. Thus, it is expected that the timing of fertiliser appli-cations will be similar: with the main applications in autumnand spring. Estimates of total nitrogen deposition (as wetand dry, nitrate and ammonium loads, kg N ha�1 a�1) werederived from the MATADOR-N model (Whitehead et al.,1998).

Daily output from the Meteorological Office Rainfall andEvaporation Calculation System (MORECS, Hough et al.,1997) model for the single stations at Lambourn and Comp-ton were used to describe the hydrological effective rain-fall (HER) and soil moisture deficit (SMD) of the Lambournand Pang, respectively, from 1 January 1999 to 17 May2004. The data supplied by the Meteorological Office alsoincluded daily actual precipitation and air temperaturedata (Fig. 2).

Final effluent nitrate and ammonium concentrationswere provided by the Environment Agency for sites on theLambourn at East Shefford, Boxford and the Trout Farm up-stream of Newbury, and sites on the Pang at Compton andHampstead Norreys operated by Thames Water plc. Littledata was available to describe the flow of final effluent ex-cept at Compton, and for the other STWs the effluent flowswere calculated from population equivalent data suppliedby the Environment Agency and Thames Water plc and theassumption of 180 l per population equivalent. Other minorworks are privately owned and data could not be obtainedto describe the effluent flow and quality.

Model set-up and calibration

The model was set-up and calibrated for the period 1 Janu-ary 2002 to 31 December 2003 in the Lambourn and for the 1January 1999 to 31 December 2003 in the Pang. These dateswere chosen because they cover the concurrent samplingperiod of the weekly water chemistry data and readily avail-able meteorological data in each river-system.

UN

CO

RR

EC

TED

PR

OO

F

262

263264

265266267268

Figure 1 Schematic of the Lambourn and Pang Catchments. The inset shows the location of Cretaceous Chalk within England. Map(a) shows the location of the Environment Agency flow gauges (s) at 1. Frilsham, 2. Bucklebury, 3. Pangbourne, 4. Shaw and theLOCAR chemical monitoring sites (m) at A. East Shefford, B. Boxford, C. Shaw, D. Frilsham, E. Bucklebury, F. Rotten Row, G.Tidmarsh. Map (b) shows the locations of the larger villages and towns, and the sewage treatment works. Map (c) shows the INCAreach boundaries.

4 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

Model set-up

For the Lambourn and Pang, 16 and 17 reaches were definedfor the respective applications of INCA-N; the sub-catch-

ments are shown in Fig. 1c and described in Tables 1 and2. Reach boundaries were chosen to coincide with the loca-tions of the hydrochemical sampling sites and flow-gauges.The remaining reach boundaries were chosen to prevent

NC

OR

REC

TED

PR

OO

F

269270271272273274275276277278

279280281282283284285286287288

Figure 2 Soil Moisture Deficit, Hydrologically Effective Rainfall, air temperature and actual precipitation at (a) Lambourn in theLambourn and (b) Compton in the Pang.

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 5

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

Uover-long reaches (typically an arbitrary value of 10,000 mis set as an upper limit).

The river network was taken from the Ordinance Survey1:50,000 map held digitally at the Centre of Ecology andHydrology. The sub-catchment boundaries were suppliedby the Centre of Ecology and Hydrology based on the grid-references for the downstream reach boundaries. The sub-catchment boundaries were derived from Ordinance Surveydigital elevation data (Moore et al., 1994; Morris and Flavin,1990).

Six land-use classes were used in the two applications:cereal, other arable, grassland, woodland, set-aside andurban. These land-use classes were derived by amalgamat-ing the land-use classes defined in the Agricultural Census.The mapping between the two is given in Wade et al.(2004). The DEFRA data includes totals of different cropareas and livestock numbers for the whole catchment aswell as for each of the INCA sub-catchments. To accountfor the missing data in a particular sub-catchment, the re-ported areas or livestock numbers were summed and

OR

REC

TED

PR

OO

F

289290291292293294295296297298299300301302303

304305306307308309310311312313314315316317318

Table 2 Reach structure used for the INCA-N modelling of the Pang

Sub-catchment Name Cumulativearea (ha)

Grid reference Base flowindex

Water qualitymonitoring

Flowgauge

1 Above Compton STW 5800 SU527791 0.952 Between Woodend and Flood Cottage 6100 SU527780 0.953 Hampstead Norreys 7100 SU528766 0.954 Everington Barn 8400 SU536752 0.955 Everington House 8800 SU535744 0.956 M4, Frilsham 9000 SU539738 0.957 Frilsham 9200 SU537729 0.95 X X

8 Marlston 10,000 SU536720 0.879 Bucklebury 11,000 SU555711 0.87 X X

10 Stanford Dingley 11,500 SU576715 0.8611 Rotten Row 12,500 SU583716 0.86 X

12 Folley Bridge 12,600 SU590720 0.8613 Road Bridge, Bradfield 13,100 SU603727 0.8614 50 m below Bradfield STW 13,200 SU606729 0.8615 A340 Tidmarsh 16,700 SU633737 0.8616 Tidmarsh 16,800 SU635746 0.86 X

17 Pangbourne 17,400 SU635765 0.86 X

Table 1 Reach structure used for the INCA-N modelling of the Lambourn

Sub-catchment Name Cumulativearea (ha)

Grid reference Base flowindex

Water qualitymonitoring

Flowgauge

1 Upper Lambourn 3400 SU322798 0.942 Lambourn 6800 SU333783 0.943 Eastbury 7500 SU349770 0.944 East Garston 9400 SU367765 0.945 Great Shefford 10,100 SU381754 0.946 East Shefford 14,200 SU390747 0.94 X

7 Weston 14,700 SU397740 0.948 Welford 15,600 SU412731 0.949 Westbrook Farm 15,800 SU426722 0.9410 Boxford 17,600 SU427716 0.94 X

11 Hunts Green 18,100 SU434702 0.9412 Woodspeen 18,300 SU442695 0.9413 Bagnor 22,900 SU453691 0.9414 Below Lambourn Trout Farm 23,000 SU457690 0.9415 Shaw 23,300 SU470682 0.94 X X

16 Newbury 26,100 SU487674 0.94

6 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

UN

Csubtracted from the whole catchment number. This gave afigure for the ‘undisclosed’ crop area or head of livestock,which was distributed evenly over the remaining catch-ment area. The predominant land-use in both catchmentsis cereal cultivation, which occupies around 60% of the to-tal land-area in the upper reaches of both river-systems;progressing downstream this percentage falls to around40% as grassland and woodland occupy a greater percent-age. Fallow or set-aside occupies around 10% of eachsub-catchment, and ‘other arable’ cultivation only approx-imately 4%.

The fertiliser application rates for the six land use classeswere taken from Johnes et al. (1998). The application rateswere differentiated by geoclimatic region. The nitrogen loadto permanent and temporary grassland and rough grazing

was calculated separately because they have different fertil-iser input rates in the geoclimatic scheme. The input loadfrom permanent and temporary grassland and rough grazingwas summed to provide a total input to ‘grassland’. Theapplication rates were multiplied by the land-use area ineach sub-catchment to provide annual application loads(kg N a�1) for each land-use type. An estimate of themonthly load was made by distributing the fertiliser loadbased on the monthly application-timings observed in farmsurvey of the Ant catchment (Johnes et al., 2003). Thesemonthly loads were converted to a daily application rate(kg N ha�1 day�1) by dividing the monthly load by the num-ber of days per month and the land class area within eachsub-catchment. The partition between nitrate and ammo-nium in the inorganic-N fertiliser was unknown and therefore

319320321322323324325326327328329330331332333334335336337338339340341342

343344345346347348349350351352353354355356357358359360361362363364365366

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 7

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

the ratio of nitrate to ammonium was 50:50 in the modelsimulations.

The calculated annual nitrogen inputs from fertiliser forthe land-classes, ‘other arable’ and cereal were approxi-mately constant for each sub-catchment in both river-sys-tems at 174 and 175 kg N ha�1 a�1, respectively, whereasthe input rate to ‘grassland’ varied between 90 and170 kg N ha�1 a�1 depending on the ratio of permanentand temporary grassland, and rough grazing. The input ratesfor ‘other arable’ and cereal were approximately constantsince both river-systems were within a single geo-climateregion: mixed arable and dairying on a permeable geology.

The annual nitrogen inputs per hectare from grazing live-stock are shown in Fig. 3. The Ant farm survey identifiedthat cattle and sheep predominately grazed grassland,while pigs were found on sugar beet and potatoes. Thus,the nitrogen inputs to the two river-systems from cattleand sheep were added to the ‘grassland’ land class andthe nitrogen inputs from pigs and poultry to ‘other arable’.Manure is comprised mainly of organic-N and ammonium-Nat the time of land application. However, the exact formof the manure N-content varies depending on livestock spe-cies, feed and moisture content. The availability of nitro-gen from the manure is also dependent on the form in

UN

CO

RR

EC

Figure 3 N inputs (kg N ha�1 a�1) from livestock to the sub-cat

PR

OO

F

which the manure is applied: solid or liquid. Once spreadthe organic-N is mineralised and the ammonium is nitrified.INCA-N simulates the store of organic-N in the soil as anunlimited N pool and the rate of mineralization can not,currently, be increased to simulate an input of manure,rather the rate of mineralization is fixed at a backgroundrate to characterise mineralization throughout the majorityof the year. Given this and the uncertainties in the form ofmanures applied to farmland in the Lambourn and Pang,and the rate of conversion of the applied N to crop-avail-able forms, then the nitrogen from manure was added tothe ammonium and nitrate pools in the model to representmineralization and nitrification of the manure-N. A 50:50split of the manure-N to the ammonium and nitrate poolswas assumed as a starting point, though these proportionswill be refined as the understanding of N availability frommanures improves. Given an absence of information regard-ing when and how many head of livestock where housedduring winter, it was assumed that all livestock grazed out-side for the entire year or the waste from housed-animalswas spread on the land. The higher inputs from livestockto ‘other arable’ in the upper reaches of the Pang andreaches below East Garston in Lambourn were caused byhigh estimated numbers of pigs in the sub-catchment.

TED

chments of (a) the Lambourn and (b) the Pang river-systems.

367368369370371372373374375376377378379380381382383384385386387388389390391392

393

394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425

426427428429430431432433434435

436437438

439

440441442443444445446447448449450451

8 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

The highest inputs to grassland of approximately350 kg N ha�1 a�1 occurred in the lower reaches of theLambourn, which have small sub-catchment areas but rela-tively high stocking densities. This result may be an anom-aly due to the limitations of the data. Despite thisproblem, the data-set remains the only publicly availablesource to describe land-use and livestock statistics. The in-puts from fertiliser and livestock were much greater thanthose from total nitrogen deposition, which is approxi-mately 15 kg N ha�1 a�1 across the catchment.

Within MORECS estimates of the duration of the growingseason for different land use types are used to calculateevapo-transpiration. These estimates also provide datesfor crop planting and harvest, and tree bud burst and leaffall across the UK, which were used to constrain the simu-lated growing-season in INCA-N. These data are availablefor each MORECS grid square; the estimates for square159, which includes the Lambourn and Pang systems, wereused (Hough et al., 1997). Where estimates were availablefor a number of different crops types forming part of a sin-gle INCA-N land class, the range of dates that incorporatedall growing seasons was used. Grass roots were assumed tobe active all year, and therefore the growing season was as-sumed to be year round, with nitrogen uptake rate a func-tion of SMD and soil temperature in the INCA-N model(Wade et al., 2002).

C452453454455456457458459460461462463464465466467468469470471472473

474

475476477478479480481482483484

UN

CO

RR

E

Calibration

The INCA-N model was calibrated manually in three phases.Firstly, the model was calibrated to the observed flows. Sec-ondly, the model parameters were adjusted until the simu-lated stream-water nitrate concentrations matched thedynamics observed, and the land-phase loads were withinthe range of those observed in UK conditions (Whiteheadet al., 1998; Wade et al., 2002). Finally, an iterative processof adjusting all parameters was done in an attempt to findthe best fit to the observed flow and nitrate concentrations(McIntyre et al., in press). The model performance was as-sessed using the Nash–Sutcliffe coefficient (Nash and Sutc-liffe, 1970), the explained variance and the correlationcoefficient for a calibration period only. The purpose of thisstudy is to use the model to explore potential explanationsfor the observed stream-water nitrate concentrations; atthis stage the model is not used for forecasting.

The daily MORECS time-series describing the hydrologi-cally effective rainfall gave a poor fit to the observedhydrology in both river-systems during preliminary calibra-tion: there was no HER when peaks where observed in thehydrograph. To overcome this limitation the Identificationof Unit Hydrographs and Component Flows from Rainfall,Evaporation and Stream-flow Data (IHACRES) model wasused to generate a new HER time-series for each river-sys-tem (Jakeman et al., 1990). IHACRES was applied using atwo-box in-parallel configuration to capture the differentresponse of the soil-water and groundwater. While it isknown that the contribution from different hydrologicalpathways and stores within the soil-water and groundwateris highly heterogeneous in both space and time, the applica-tion of IHACRES represents a practical approach to the der-ivation of the HER.

TED

PR

OO

F

To further improve the fit between the simulated and ob-served flows within INCA-N, the initial flows in the soil-water and groundwater boxes for each land-use type wereadjusted until the base-flow response closely matched thatobserved. The base flow index (BFI), which represents thepartition between the soil-water and groundwater and wasobtained from the hydrometric register for the flow-gauges(Tables 1 and 2; Gustard et al., 1992).

The residence times in each of the land use classes werethen adjusted since:

• the residence times of the soil-box controlled the time topeak of the simulated hydrograph;

• the residence time in the groundwater box controlled thedecay of the falling limb.

As such, the residence times relate more to lag times be-tween a precipitation event and the peak in the hydrographand the decay time, rather than a true measure of a soil- or aground-water turnover-time. The initial values of the vol-umes in the soil-water and groundwater boxes were setbased on preconceived notions of equivalent water depthsper km2. The drainage volume in the soil-box relates to theamount of water in the macropores or drains which is readilyable to move and, by a piston effect, most strongly influ-ences the rising hydrograph limb. The soil retention volumerepresents the water volume stored in the soil that respondsmore slowly, and may make up the majority of the waterstorage in the soil (similar to the field capacity concept).This water may be thought of as stored in the micropores.

The relationship between the velocity, v and the dis-charge, Q in each reach is v = aQb (Whitehead et al.,1998). The values of the ‘a’ and ‘b’ parameters affect thetime of the response of the flow to the rainfall input, andthe initial storage in the reach. Thus, the parameters ‘a’and ‘b’ were calibrated to ensure the in-stream reach vol-umes and the associated depths were reasonable, and to im-prove the hydrological fit.

The initial nitrate and ammonium concentrations in theland phase of the model were also set by calibration. Tosimulate the spatial variations in the stream-water nitrateconcentrations observed in the Pang, INCA-N had to be mod-ified so that the groundwater initial flow and nitrate andammonium concentrations varied by sub-catchment geologyrather than land-use. The terms relating to mineralization,immobilisation, nitrification, denitrification and plant-up-take within the soil were adjusted until the simulated an-nual output load was within the range specified in theliterature (Wade et al., 2002).

The ‘in-stream’ and ‘groundwater’ models

The parameters controlling the in-stream nitrification anddenitrification were then adjusted to produce the best fitto the stream-water nitrate concentrations (the ‘in-stream’model). It was found that by adjusting these parameters agood-fit to the observed stream-water nitrate concentra-tions could be achieved even when the groundwater concen-trations remained constant. Discussion over yet unpublisheddata suggests that groundwater nitrate concentrations inthe Lambourn and Pang generally varied with season possi-bly due to fluctuations in the water-table intercepting

485486487488489490491492

493

494

495496497498499500501502503504505506507508509510511512513

514515516517518519520521

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 9

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

nitrate stored at different depths in the unsaturated-zone.A previous study by Wellings and Bell (1980) also suggest an-nual-cycles in the measured nitrate concentrations in theunsaturated zone. When this sinusoidal pattern in thegroundwater concentrations was simulated (the ‘groundwa-ter’ model), the seasonal pattern in the observed stream-water nitrate concentrations was reproduced without theneed to invoke any in-stream processes.

522523524525526527528529530531532533534535536537538539540541542543544545546

Results

The observed and simulated daily-mean flows

During the period 2002–2003 the observed daily-mean flowsin the Lambourn at Shaw varied from 1 to 5.6 m3 s�1. Thehydrograph showed a rapid response to individual stormssuper-imposed on a slow-response groundwater dynamic(Fig. 4). Annual flow peaks occurred in March 2002 and Jan-uary 2003, and the lowest flows occurred in October 2002after the summer dry period.

In the Pang, the observed flow dynamics were similar atall three gauging stations which indicated that the flowswere correlated (Fig. 4). As observed in the Lambourn,the flow-pattern was dominated by the groundwater re-sponse, but with a more rapid response to storm-eventsthan observed in the Lambourn; this is probably due tothe influence of the Palaeogene which is less permeablethan the chalk. The peak flows were generally observed inwinter (December/January/February) though annual peakswere observed in May 2000 at Frilsham. The lowest flowsin the Pang typically occurred during September and Octo-ber. Thus in both systems, the highest observed stream-

UN

CO

RR

EC

Figure 4 The observed (dashed-line) and simulated ‘in-stream’Lambourn at (a) Shaw and the Pang at (b) Frilsham, (c) Bucklebsimulations are identical.

TED

PR

OO

F

flows were between December and May and the lowest flowsoccurred in September and October.

In general INCA-N was able to simulate the flow dynamicsobserved at all gauging stations on the Lambourn and Pang:the key dynamics of the base-flow increase and recessionwere simulated and the model captured the onset andrecession of individual storm events (Fig. 4; Table 3). Boththe ‘in-stream’ and ‘groundwater’ models produced thesame results since the hydrological component of the twomodels was identical.

The simulated flow failed to reproduce the flow peaksobserved during winter 2002 in the Lambourn at Shaw, sug-gesting an under-estimate of the HER during this event sincethere was insufficient volume of flow simulated (Fig. 4a).The simulated rate of increase of the rising limb and therate of decay of the falling limb for the highest flow-peakwas good, and the model was able to simulate the timingof individual storm responses.

The simulated flow was a good fit to that observed at Tid-marsh on the Pang though failed to capture the extremefloods observed during November and December 2000(Fig. 4d). The simulation of the base-flow recession wasgood at all three gauged sites on the Pang, but the flowsat Frilsham were over-estimated (Fig. 4b). The upperreaches of the Lambourn and Pang are ephemeral. Whilstthe observed flow approaches zero at the most upstreamgauged-site on the Pang, the simulated flow over-estimatedthe low-flows, though the simulated flows did approachzero. This over-estimation occurred for two reasons. Firstly,the estimate of HER was derived by generating a relation-ship between the flows at Shaw and Pangbourne with the ac-tual precipitation measured in the Lambourn and Pangcatchments, respectively; consequently the simulated HER

(solid-line) and ‘groundwater’ (solid-line) hydrographs in theury and (d) Pangbourne. The ‘in-stream’ and ‘groundwater’

C

OO

F

547548549550551552

553

554

555

556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586

587588589590591592593594595596597598599600601602603604605606607608609610611612613614615

616617

618619620621622623624625626627

Table 3 Model performance during calibration to flow and stream-water nitrate concentration data measured in the Lambournand Pang systems using the ‘in-stream’ and ‘groundwater’ models, the results for the latter are presented in brackets

Determinand River System Site Nash–Sutcliffe Coefficient Explained Variance Correlation Coefficient

Flow Lambourn Shaw 0.78 (0.78) 0.78 (0.78) 0.89 (0.89)Pang Frilsham 0.20 (0.20) <0.00 (<0.0) 0.44 (0.44)

Bucklebury 0.75 (0.75) 0.62 (0.62) 0.87 (0.87)Pangbourne 0.56 (0.56) 0.37 (0.37) 0.75 (0.75)

Stream-water nitrateconcentration

Lambourn East Shefford 0.40 (0.20) 0.42 (0.42) 0.63 (0.45)

Boxford <0.00 (<0.00) <0.00 (0.42) <0.00 (<0.00)Shaw 0.60 (0.65) <0.00 (<0.00) 0.78 (0.80)

Pang Frilsham 0.56 (0.49) 0.66 (0.56) 0.75 (0.70)Bucklebury 0.23 (0.48) 0.26 (0.52) 0.48 (0.69)Rotten Row <0.00 (0.25) <0.00 (0.62) <0.00 (0.50)Tidmarsh <0.00 (<0.00) <0.00 (0.00) <0.00 (<0.00)

The Nash–Sutcliffe coefficient is defined in Nash and Sutcliffe (1970).

10 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

UN

CO

RR

Ewas likely to be an over-estimate for the upper reaches ofeach system. Also INCA-N included conditions on the modelequations which prevented the flow in the stream reachingzero; this was to prevent numerical-problems solving equa-tions based on concentration when there was no simulatedflow.

The observed and simulated stream-water nitrateconcentrations

Observed stream-water nitrate concentrations

The Lambourn at Shaw and the Pang at Pangbourne werethe only two gauging-stations with continuous flow recordsover the entire study-period considered for each system.As such, each measurement (of nitrate or stream tempera-ture) made in the Lambourn was plotted against the daily-mean flow at Shaw (Figs. 5a–c and 6), and each in the Pangagainst the flow at Pangbourne (Figs. 5d–g and 7). Whilstthis approach is a simplification since the flows varied inabsolute terms at each sample-site, the purpose was toexamine general patterns in the relationship between themeasurements and flow. Given the flows in the Pang werecorrelated, it is assumed that this was true of the Lambournalso and therefore using the flows from the most down-stream gauge was reasonable (Fig. 4b–d). The period from1 January 2002 to 31 December 2003 was considered at allsites only to provide a consistent time-period forcomparison.

In the Lambourn, the stream-water nitrate concentra-tions varied between 4.9 and 8.2 mg N l�1, though themajority of the observed concentrations were between 6.5and 8.0 mg N l�1 (Fig. 5a–c). The Pang exhibited a greaterrange in extremes (Fig. 5d–g). The highest stream-water ni-trate concentrations of approximately 11 mg N l�1 werefound at Tidmarsh under the highest flow-conditions, whilstthe lowest concentrations of approximately 3–4 mg N l�1

were also found under high-flow conditions at Bucklebury.The observed relationship between stream-water nitrate

concentrations and flow produced the same general patternin both the Lambourn and Pang rivers, and across all sitesexcept Tidmarsh (Fig. 6; Neal et al., 2004a). In the Lambo-urn as flow increased from the minimum to approximately

TED

PR2 m3 s�1, there was a dilution of the in-stream concentra-

tions; between 2 and 3.5 m3 s�1 the concentrations then in-creased with flow; between 3.5 and 5 m3 s�1 theconcentrations remained constant until there was a furtherincrease in concentration above a flow-threshold of approx-imately 5 m3 s�1.

In both river-systems there was a weak-negative correla-tion between stream-water temperature and flow, and thestream-water nitrate concentration was negatively corre-lated with temperature also (Figs. 6 and 7). Stream-watertemperature is a surrogate for season: the temperaturesare highest in July and August, and lowest in January. Thehighest observed stream-water nitrate concentrations oc-curred in January and February which coincided with thehighest flow conditions and the lowest temperatures. Thelowest observed nitrate concentrations occurred betweenJuly and October, a period which started with the warmestmonths and ended with the lowest flows. Thus the seasonaldynamics in the stream-water nitrate concentrations wererelated to both flow and temperature (season), but the sea-sonal-signal in the time-series of the observed stream-waternitrate concentrations did not compare exactly to the sea-sonal pattern in the observed flows (Figs. 6 and 7). Thestream-water nitrate concentrations were also observed todecrease in the stream during some storm events whenthe soil-water contribution increased. This observed patternsuggests that the soil-water contribution had lower concen-trations than the groundwater during some storm-events asthe concentration in the stream was diluted.

Simulated stream-water nitrate concentrations: ‘in-stream’ model

The simulated stream-water nitrate concentrations usingthe ‘in-stream’ model match the dynamics observed at thethree sites on the Lambourn, though the increase in concen-tration in autumn 2003 was under-estimated (Fig. 5a–c).The calibrated model set-up was able to simulate the dilu-tions in the stream-water concentrations observed in re-sponse to storm-events and reproduced the generalincrease in concentration during the winter and early springmonths and the decrease during the spring through to au-tumn. The simulated effect of final-effluent inputs from

NC

OR

REC

TED

PR

OO

F

628629630631632633634635636637638

639640641642643644645646647648649

Figure 5 The observed (points) and simulated stream-water nitrate concentrations (mg N l�1) using the ‘in-stream’ (solid-line)and ‘groundwater’ (dashed-line) models in the Lambourn at (a) East Shefford, (b) Boxford, (c) Shaw and in the Pang at (d) Frilsham,(e) Bucklebury, (f) Rotten Row and (g) Tidmarsh.

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 11

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

Uthe STWs on the stream-water nitrate concentrations wasalmost negligible as nearly all works, apart from Comptonin the Pang, discharge only small (<0.01 m3 s�1) volumesof effluent. Removing all the nitrate from the final effluentat Compton, caused the simulated stream-water nitrateconcentrations to decrease by a maximum of approximately1.5 mg N l�1 immediately downstream of the works duringthe summer months when there was a lack of dilution forthe effluent in the stream. At Pangbourne the effect ofremoving the nitrate from the Compton effluent was notdiscernable.

If the in-stream processes of denitrification or plant-uptake were omitted from the model run, then theseasonal-pattern observed in the stream-water nitrate con-centrations was not reproduced with a constant groundwa-ter end-member concentration. Instead, the simulatedpattern was one of a high base-flow concentration derivedfrom groundwater, diluted by inputs of soil-water duringstorm events. Thus, the simulated mixing of a groundwaterand soil-water end-member, the latter which varied season-ally, and the inclusion of inputs from point-sources, wasinsufficient to explain the patterns observed in the

UN

CO

RR

EC

TED

PR

OO

F

Figure 6 The observed relationships between stream-water NO3 concentrations and (a) flow and (b) stream-water temperature,and between (c) stream-water temperature and flow in the River Lambourn. All the flows are from the measurements made at Shaw.

12 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

UN

CO

RR

EC

TED

PR

OO

F

Figure 7 The observed relationships between stream-water NO3 concentrations and (a) flow and (b) stream-water temperature,and between (c) stream-water temperature and flow in the River Pang. All the flows are from the measurements made atPangbourne.

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 13

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

F

650651652653

654655

656657658659660661662663664665666667668669

670671672673674675676677678679680681

682

683

684

685686687

14 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

stream-water nitrate concentrations: in-stream processesappeared important in controlling the seasonal pattern inthe stream-water nitrate concentrations, though point-source inputs controlled the summer minima.

Simulated stream-water nitrate concentrations:‘groundwater’ model

To the eye the ‘groundwater’ model simulates similarstream-water nitrate concentrations to the ‘in-stream’ mod-el for both the Lambourn and the Pang (Fig. 5). Though thestructure of each model is quite different, it is impossibleto state which model is best given the similar results of thethree test of model-fit (Table 3). The sinusoidal variationsin the groundwater concentrations generate a sinusoidal pat-tern in the in-stream nitrate concentrations, which are mod-ified during storm events by simulated lateral inputs from thesoil store. These inputs either dilute or concentrate thein-stream concentrations depending on the season. No in-stream processes were included in the ‘groundwater’ model.

Removing all the nitrate from the final effluent at Comp-ton in the ‘groundwater’ model, caused the simulated

UN

CO

RR

EC

Figure 8 Nitrate export (kg N a�1) from the land to the river. Cumuusing the (a) ‘in-stream’ and (b) ‘groundwater’ models, and the Pan

stream-water nitrate concentrations to decrease by a max-imum of approximately 2.0 mg N l�1 immediately down-stream of the works during the summer months whenthere was a lack of dilution for the effluent in the stream.At Pangbourne the effect of removing the nitrate from theCompton effluent was still discernable.

For both models, the simulated stream-water nitrate con-centrations match poorly those observed at Tidmarsh on thePang which is the monitoring site with the most measure-ments. This highlights that either model is able to simulatethe general seasonal variations and inter-annual variability,but is unable to accurately simulate daily variations.

RO

OThe simulated soil and groundwater nitrateconcentrations

Simulated soil nitrate concentrations

The simulated soil-water nitrate concentrations displayedthe same patterns in both the Lambourn and Pang systemsfor both models since the soil-box processes were set-up

TED

P

lative inputs of nitrogen along the main channel of the Lambourng using the (c) ‘in-stream’ and (d) ‘groundwater’ models.

688689690691692693694695696697698699700701702703704705706707

708

709710711712713714715716717718719720721722723724725726727

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 15

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

identically. The soil-water concentrations under the cerealand ‘other arable’ land-uses increased during the wintermonths and peaked in March (with a maximum simulatedconcentration of 10 and 33 mg N l�1, respectively). The min-imum occurred in August (at approximately 3 mg N l�1 forall land-use types). This pattern coincided with the applica-tion of fertilisers in autumn and spring, and the uptake ofnitrate from the soil during the growing season by the crops.The grassland and woodland land-uses had peak concentra-tions during October and minimum concentrations in July.The simulated soil-moisture conditions were wet in Octoberand the simulated soil temperature was still warm. Conse-quently the simulated mineralization and nitrification wererelatively high compared to other times of the year. In Julythe soils were dry which inhibited the simulated formationof nitrate in the soils. Only the ‘cereal’ and ‘other-arable’land use classes had simulated soil-water concentrationsgreater than those simulated in the groundwater, and thenonly for three months following the spring fertiliseraddition.

UN

CO

RR

EC

Fig 8. (cont

OO

F

Simulated soil nitrate concentrations

The simulated groundwater nitrate concentration for allland-use types in the Lambourn was constant at approxi-mately 9 mg N l�1, which corresponded to the simulatedand observed stream-water concentrations at high flows inwinter. This simulated groundwater end-member concen-tration was higher than the range of observed groundwaterconcentrations measured in boreholes located on eithersides of the River Lambourn at Boxford (National Grid Refer-ence 442500 172300); where the observed range was 4.9–6.9 mg N l�1 (Gooddy et al., this issue); though unpublishedgroundwater nitrate concentration data, measured by theEnvironment Agency, in boreholes across the Pang andLambourn catchments show that the range of the mean con-centrations is 0.2–23 mg N l�1 (measured as total oxidisednitrogen).

In the simulation of the Pang river-system, the sub-catchments downstream of Bucklebury had an initialgroundwater concentration of 8.5 mg N l�1 whereas thoseupstream had an initial concentration of 11 mg N l�1. The

TED

PR

inued)

728729730731732733734735736737738739740741742743744745746

747

748

749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786

787788789790791792793794795796797798799800801802

803

16 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

lag time between the effective rainfall and in-stream re-sponse was the same (45 days) irrespective of spatial loca-tion though the base flow index decreased down thesystem reflecting the increased importance of near-surfaceflow pathways in the Palaeogene deposits (Table 2). Thesimulated groundwater concentrations in the Pang also re-mained constant over the simulation period.

The simulated groundwater nitrate concentration variedas a sinusoid from 7.9 to 6.1 mg N l�1 in February andAugust, respectively, for each land use in each reach alongthe Lambourn. For the Pang, the concentration variedfrom 9.5 (February) to 7.5 mg N l�1 (August) for eachland-use in each from the upper reaches to Rotten Row,and between 7.2 (February) to 5.5 mg N l�1 (August) down-stream. These results indicate spatial as well as temporalvariation in the groundwater input, and that the simulatedgroundwater nitrate concentrations were lower than the‘in-stream’ model which simulated higher in-stream re-moval of nitrate.

C804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845

UN

CO

RR

EThe relative contribution from point and diffusesources

The modelled output suggested that a smaller mass of ni-trate was exported from the Lambourn than the Pang. Inthis case, export was defined as the nitrate mass exportedfrom the land-phase through the combined soil and ground-water pathways, and was averaged over the period simu-lated in each system. Specifically, the ‘in-stream’ modelsimulations suggest that approximately 500,000 kg N a�1

(Fig. 8a) are exported from the Lambourn compared to600,000 kg N a�1 from the Pang (Fig. 8c). The exported massfrom the ‘groundwater’ model simulations was 400,000from the Lambourn (Fig. 8b) compared to 490,000 kg N a�1

from the Pang (Fig. 8d). Thus, simulated mass exportedfrom diffuse-sources was higher by approximately 20% inthe ‘in-stream’ model compared to the ‘groundwater’ mod-el. This was due to the removal of nitrogen mass from theriver by the in-stream processes to match the observed ni-trate concentrations. The greater export from the Pangwas due to two factors: (a) the higher simulated groundwa-ter concentrations and (b) the wetter period of 2001 whichwas included in the simulation. In both systems just over50% of the nitrate observed in the river was derived fromland in cereal production (55% Lambourn; 51% Pang), thesecond major source was grassland which provided approx-imately 25% of the nitrate exported by the rivers. The pointsource contribution in terms of load was negligible ataround 2% (Fig. 8). These proportions were approximatelyconstant along the length of the main channel in each sys-tem, though the load increased where a tributary joinedthe main river-channel, such as where the Bourne joinedthe Lambourn (reach 12, Fig. 8a and b) and where the Win-terbourne joined the Pang (reach 14, Fig. 8c and d), or whenthere was a STW input, for example from Compton STW inreach 2 of the Pang. For the ‘groundwater’ model the sim-ulated export, in terms of nitrate export per hectare peryear (averaged over the simulation period) from cerealand grassland in the Pang was in the range 30–40 kg N ha�1,and for ‘other arable’ land in the range 35–55 kg N ha�1,with the highest values found upstream of Bradfield, which

OF

has the highest numbers of livestock and the highest simu-lated groundwater concentration. Export from cereal,grassland and ‘other arable’ ranged between 17 and25 kg N ha�1 for all land-use types for both the ‘in-stream’and ‘groundwater’ models, including ‘set-aside’ whichwhilst it received nitrogen from atmospheric deposition onlyhad high simulated groundwater nitrate-concentrations dueto the history of nitrogen applications.

The simulated in-stream combined denitrification andplant-uptake over the 2-year simulation period in theLambourn was estimated as approximately 225,000 kg for2 years. As such, the transfer of nitrogen to the atmosphereand in-stream plants was approximately 18% of the totalinorganic (nitrate plus ammonium) load exported in the riv-er: in-stream dissolved inorganic nitrogen, plus denitrifica-tion and plant-uptake.

TED

PR

ODiscussion

This modelling assessment of the key factors and processescontrolling the in-stream nitrate dynamics in the Lambournand Pang was inconclusive. At least two models could beproposed to explain the observed seasonal and annual pat-terns observed in-stream: the ‘in-stream’ and ‘groundwa-ter’ models. This result confirms an earlier sensitivity-analysis of INCA-N when applied to flow and in-stream ni-trate and ammonium concentrations measured in the RiverKennet; the study indicated that an optimum model struc-ture could not be identified with certainty (McIntyreet al., in press).

The ‘in-stream’ model suggest that the patterns in theobserved nitrate concentrations can be explained as threekey controls in the Lambourn: (i) an input of groundwaterwith a constant nitrate concentration over the simulatedperiod which increased in volume as the flows increased,(ii) inputs of soil-water with generally lower concentrationsthan the groundwater during storm-events, and (iii) an in-stream biological processing of the groundwater and soil-water nitrate contributions. The nitrate concentrations inthe River Pang at Bucklebury and up-stream (on the Chalk)can be explained by the same three controlling factors thatoperate in the Lambourn. Downstream of Bucklebury the in-stream nitrate dynamics were complicated due to the influ-ence of the Palaeogene deposits. At Tidmarsh, the model-ling assessment using the ‘in-stream’ model suggests thatthe same factors still control the in-stream nitrate dynam-ics, but the influence of near-surface pathways appears rel-atively more important than on the Chalk: during spring2000 the stream-water nitrate concentration increased withflow, whereas if the in-stream biology where more impor-tant, the observed nitrate concentrations would be ex-pected to decrease.

Alternatively the ‘groundwater’ model suggests that thein-stream nitrate concentrations can be explained as: (i) agroundwater concentration that varies with fluctuatingwater-table level and (ii) inputs of soil-water with generallylower concentrations than the groundwater during storm-events. The ‘groundwater’ model therefore supports thefirst hypothesis, whereas the ‘in-stream’ model suggestthe first hypothesis is false. Again to explain the observednitrate concentrations in the Pang a spatially, as well as

C

846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907

908909910911912913914915916917918919920921922923924925926927

928929930931932933934935936937938939940

941942943944945946947948949950951952953954

955956957958959960961962963964965966967968

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 17

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

UN

CO

RR

E

temporally, varying groundwater end-member wasrequired.

Without more detailed data it is difficult to identify thecorrect model. Preliminary analysis of the LOCAR groundwa-ter data suggests that the ‘groundwater’ model is the cor-rect description. If the groundwater concentrationobserved at Boxford is indicative of the general groundwa-ter concentration then the implications are that the simu-lated groundwater concentration is too high (Gooddyet al., this issue). The range of nitrate concentrations inriparian groundwater at Westbrook Farm (BGS, unpublisheddata), just up-stream of Boxford, match those used in thesimulation of the Lambourn, but may have been influencedby surface water-groundwater interactions (Gooddy et al.,this issue). Measurements from other boreholes at West-brook highlight the spatial heterogeneity in the concentra-tions and it is unclear if all groundwater exhibits aseasonal pattern in the nitrate concentrations caused bythe rise and fall of the water-table. Another study of nitratein Chalk suggests a seasonal pattern in the upward anddownward movement of the solute due to changes in theevapotranspiration demand (Wellings and Bell, 1980). Datadescribing in-stream nitrate processes has not yet been pub-lished. Only once both data-sets have been analysed can thetwo models proposed here be rejected or accepted. How-ever, it is unclear if these new data will help to identifythe optimum model or only serve to the highlight the spatialand temporal heterogeneity of the catchment (Kirchneret al., in press).

Despite the current limitations, consistent predictionsbetween the models are enlightening regarding the key fac-tors and processes operating. Also the consistent predictionsindicate that the manual calibration of models, though lim-ited in the number of different model structures and param-eter sets that can be explored compared to automateduncertainty-analysis techniques (e.g. Freer et al., 1996;McIntyre et al., in press), can be used to explore how a riv-er-system may function. There were four consistent predic-tions. The first was the simulation of the observation thatthe stream-water nitrate concentrations decreased duringevents during spring, summer and autumn. This contradictsthe simple suggestion that enhanced agricultural run-offduring storm-events increases the stream-water nitrate con-centration; rather such inputs can either concentrate or di-lute the groundwater inputs in the stream depending uponseason, and perhaps varying between storm events.

The second consistency was that both models were ableto simulate the seasonal and inter-annual variations in thestream-water chemistry suggesting that at larger spatial(>50 km2) and temporal scales (seasonal) medium complex-ity models in combination with readily-available data-setscan be used to provide an estimate of the stream-waterchemistry to a first approximation (Wade et al., 1999; Wadeet al., 2005).

The third consistency was that both model simulationsalso suggest that the contribution from point-sources tothe in-stream nitrate load was almost negligible in compar-ison to that exported from arable land. Reducing point-source inputs is unlikely to have a significant effect on thein-stream load but may help lower the in-stream nitrateconcentrations during the summer low-flow periods closeto the effluent input.

TED

PR

OO

F

Finally, the fourth consistency was that both models sug-gested that there was no trend in groundwater nitrate con-centrations simulated over the study-period. This indicatesthat the second proposed hypothesis was true: the mainstore of nitrate is in the groundwater. Given the concentra-tion of this store did not change (over a simulated 5 yearperiod) this indicates the store was being recharged, or onlyslowly depleted. The volume of water in the groundwaterstore was set to 6 · 107 m3 km�2 in the model calibration,which equates to a depth of 200 m when assuming a porosityof 0.3. The effective depth of the Chalk aquifer is approxi-mately the upper 50–60 m of the saturated zone (Priceet al., 1993) and therefore the simulated depth is too largeand further work is necessary to explore the uncertainty inthe size of this store before the importance of this store canbe determined accurately. Both models suggested that themain contributor to the store of nitrate in the groundwaterwas land used for cereal production.

Based on these findings the following recommendationsare made:

• Given an inability to determine the optimum model it issuggested that multiple models, which could be multiplemodel-structures and/or multiple parameter-sets shouldall be used to forecast the potential changes in thestream chemistry to give an estimate of the affects ofequifinality on model predictions (Beven, 2002). Thisstudy has shown that two model structures which givegood simulations of the in-stream nitrate concentrationsdiffer by approximately 20% in terms of the nitrogen loadexported in the river-water; which is a first estimate ofthe uncertainty in the exported load due to modelstructure.

• The cycle of model-development and data-analysis ofnew laboratory and field-based measurements to testand re-structure models appears to have been broken.The rapid increase in computing power has allowedmodel development beyond the capability to test modelstructures and identify parameters with existing data-sets as demonstrated in this study. There is a need toexpend research-effort on measurement. Though theuse of automated uncertainty-analysis techniques isstrongly recommended, this study has demonstratedthe value of manual calibration since the model was usedas a tool with which to learn about the interaction ofpotential factors and processes controlling nitrate move-ment and fate in river-systems.

• Current emphasis within the UK, Europe and the USseems to be upon large integrated-projects that measuremany catchment variables in an attempt to explain thehydrology and hydro-chemistry observed in-stream, asin the case of LOCAR. It is unclear if such data will helpto define model structures and parameter sets giventhe infinite-set of possible combinations, or overcomethe inability to scale point measurements to area-basedflux estimates. The monitoring of stream-water chemis-try at high-frequencies (hourly) over long time-periods(decades) for selected determinands may help to charac-terise the range of residence times for different pollu-tants and determine any fractal character that is notsimulated by current ‘bucket’ models such as INCA-N(Kirchner et al., 2001). The latter approach is appealing

C

969970971972973974975976

977978979980981982983

984985986987988989990991992993

9949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027

102810291030

10311032103310341035103610371038103910401041104210431044104510461047

1048

1049105010511052105310541055105610571058105910601061106210631064106510661067

1068

1069107010711072107310741075107610771078

1079

10801081108210831084

18 A.J. Wade et al.

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

UN

CO

RR

E

because it does not rely on constructing a conceptualmodel of the system but allows the residence-time distri-bution to be determined from measurement. The fractalapproach should, however, be used in conjunction withprocess-based studies to help discriminate between‘‘true age’’ and ‘‘apparent age’’, the latter which iscaused by the mixing of waters of different age.

• Irrespective of the modelling approach, to achieve anaccurate nitrogen mass-balance, more accurate data onthe spatial distribution of crops and livestock, and areasof crops and numbers of livestock, are required to esti-mate diffuse inputs; and more accurate effluent flowsand concentrations are required to quantify point-sourceinputs.

• This study has not considered septic tank inputs for whichlittle information could be gathered within the project.Further work is required to estimate the importance ofthis source, which has the potential to affect the in-stream nitrate dynamics. In particular this source maycause an increase in nitrate concentration under high-flow conditions which may explain the increase in in-stream concentrations above the 5 m3 s�1 threshold inthe Lambourn (Neal et al., 2004a), and thereby createa third alternative model.

• This work has not considered the potential role of wet-lands and buffer strips within the catchment. Given theimportance of the groundwater contribution to thestream it is important to establish the amount of ground-water that flows through the wetlands and if nitrate isremoved. The timing of nitrate flushing which is likelyto occur during winter-events is also important and is apossible source for higher nitrate concentrations duringwinter-months. Once flushed the stores in such riparianzones may re-fill thus the contribution to the stream willbe diminished. The area of wetlands in the Lambourn andPang river-systems is small, approximately 3% of the totalarea. Their effect therefore depends upon the fraction ofgroundwater intercepted and the flushing-store cycle. Asimple mass-balance is required to establish theirimportance.

• The modelled estimate of the total N deposition derivedfrom the MATADOR-N model is low compared to measure-ments reported in a recent study (Neal et al., 2004b):though the modelled estimate and measured total nitratedeposition are similar (approximately 7 kg N ha�1 a�1),the modelled total ammonium deposition is lower thanmeasured (7 compared with 24 kg N ha�1 a�1). It isunclear if the measurements include catchment sourcesor not. INCA-N includes the simulation of ammoniumuptake by vegetation and the conversion to nitrate. Fur-ther work is required to modify the model structure tosimulate the movement of ammonium with sediment,and to establish if the ammonium measurements werederived only from atmospheric sources. Running modelsimulations with altered deposition inputs would providean estimate of the impacts on the water quality of differ-ent nitrogen inputs, though such simulations will proba-bly lead to more possible model structures and furtheruncertainty.

Without confidence in the model structures and parame-ter-sets it is not possible to answer questions 3–7 posed in

TED

PR

OO

F

the Introduction at present; to find answers to these ques-tions it is recommended that INCA-N be used within an auto-mated procedure to identify as many acceptable models aspossible, and then use all the models for forecastsimulations.

The ‘groundwater’ model was created by substituting asine-wave into the groundwater component of INCA-N.Structural changes are required to the current version ofINCA-N to simulate groundwater turnover times and themovement of water vertically down through the unsatu-rated zone. As such a new conceptual model of the unsatu-rated zone and groundwater has been created to addressthis gap (Jackson et al., this issue). It is envisaged that thisnew model will be linked to the soil-nitrogen model of INCA-N to simulate the whole river-system. This new model mustbe applied with high-frequency data-sets to determine theappropriateness of the model structure.

Conclusions

This model-based assessment of the key factors and pro-cesses controlling in-stream nitrate concentrations in theLambourn and Pang river-systems was inconclusive. Fur-ther data are required to provide information on thegroundwater nitrate concentrations and in-stream nitrateprocesses to help identify the model structure, though itis unclear how helpful these data will be given the likelyspatial and temporal heterogeneity. Until there is suffi-cient research evidence to determine whether modelscan be adequately defined using internal catchment mea-surements and/or high frequency in-stream measurementsit is recommended that all models which can explain theobserved data are used for forecasts so that the uncer-tainty which arises due to equifinality can be factored intomanagement decisions. Whilst this is most efficientlyachieved using automated uncertainty-analysis techniques,manual calibration and exploration of alternative modelstructures and parameter-sets can improve understandingof river-system functioning.

Acknowledgements

This work was funded by the Natural Environment ResearchCouncil and the Environment Agency as part of the LowlandCatchment Research program (project NER/T/S/2001/00942), and by the European Union under the Euro-limpacsresearch project (contract number GOCE-CT-2003-505540).The authors are grateful for comments on drafts of the textby Paul Shand and Colin Neal, and for the data supplied bythe Department of the Environment, Food and Regional Af-fairs, the Environment Agency, the Meteorological Officeand Thames Water plc.

References

Beven, K.J., 1993. Prophecy, reality and uncertainty in distributedhydrological modelling. Adv. Water Resour. 16, 41–51.

Beven, K., 2002. Towards an alternative blueprint for a physicallybased digitally simulated hydrologic response modelling system.Hydrol. Process. 16, 189–206.

C

1085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148

1149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212

Towards an improved understanding of the nitrate dynamics in lowland, permeable river-systems 19

HYDROL 15339 No. of Pages 19, Model 6+

13 May 2006 Disk Used ARTICLE IN PRESS

UN

CO

RR

E

Beven, K.J., in press. A Manifesto for the equifinality thesis. J.Hydrol.

Bradford, R.B., 2002. Controls on the discharge of Chalk streams ofthe Berkshire Downs, UK. Sci. Total Environ. 282/283, 65–80.

Department of the Environment, 1995. Biodiversity: The UKSteering Group Report. HMSO, London.

Freer, J., Beven, K.J., Ambroise, B., 1996. Bayesian estimation ofuncertainty in runoff production and the value of data: anapplication of the GLUE approach. Water Resour. Res. 32, 2161–2173.

Gooddy, D., Darling, W.G., Abesser, C., Lapworth, D.J., this issue.Using chlorofluorocarbons (CFCs) sulphur hexafluoride (SF6) tocharacterise groundwater movement and residence time in alowland chalk catchment. J. Hydrol.

Gustard, A., Bullock, A., Dixon, J.M., 1992. Low flow estimation inthe United Kingdom. Institute of Hydrology Report No. 108,Institute of Hydrology, Wallingford.

Heathwaite, A.L., Burt, T., Trudgill, S.T., 1993. Overview – thenitrate issue. In: Burt, T.P., Heathwaite, A.L., Trudgill, S.T.(Eds.), Nitrate: Processes, Patterns and Management. Wiley,New York, pp. 3–21.

Hough, M., Palmer, S., Weir, A., Lee, M., Barrie, I., 1997. TheMeteorological Office Rainfall and Evaporation Calculation Sys-tem: MORECS Version 2.0 (1995). Meteorological Office,Bracknell.

Jackson, B.M., Wheater, H.S., McIntyre, N., Butler, A.P., Mathias,S., this issue. A simple model of variable residence time flow andnutrient transport in the Chalk. J. Hydrol.

Jakeman, A.J., Littlewood, I.G., Whitehead, P.G., 1990. Computa-tion of the instantaneous unit hydrograph and identifiablecomponent flows with application to two small upland catch-ments. J. Hydrol. 117, 275–300.

Johnes, P.J., 1996. Evaluation and management of the impact ofland use change on the nitrogen and phosphorus load deliveredto surface waters: the export coefficient modelling approach. J.Hydrol. 183, 323–349.

Johnes, P.J., Butterfield, D., 2003. Export coefficient model runsfor the Hampshire Avon and the Herefordshire Wye catchments,based on 1 km2 grid scale data from the 1995 Annual AgriculturalCensus returns. Report to DEFRA, Aquatic EnvironmentsResearch Centre, University of Reading.

Johnes, P.J., Bennion, H., Curtis, C., Moss, B., Whitehead, P.,Patrick, S., 1998. Trial Classification of Lake Water Quality inEngland and Wales. Report from Environment Agency Project E2-i721, Environment Agency, Bristol, p. 264.

Johnes, P.J., Whitehead, P.G, Davies, S, Wade, A., Butterfield, D.,2003. Effectiveness of Eutrophication Control by PhosphorusReduction. Environment Agency R&D Technical Report, Bristol,149pp.

Kirchner, J.W., Feng, X., Neal, C., 2001. Catchment-scale advec-tion and dispersion as a mechanism for fractal scaling in streamtracer concentrations. J. Hydrol. 254, 82–101.

Kirchner, J.W., Feng, X., Neal, C., Robson, A., in press. The finestructure of water-quality dynamics: the (high-frequency) waveof the future. J. Hydrol.

Limbrick, K.J., Whitehead, P.G., Butterfield, D., Reynard, N., 2000.Assessing the potential impacts of various climate changescenarios on the hydrological regime of the River Kennet atTheale, Berkshire, south-central England, UK: an application andevaluation of the new semi-distributed model, INCA. Sci. TotalEnviron. 251/252, 539–556.

Marsh, T.J., Lees, M.L. (Eds.), 2003. Hydrological Data UnitedKingdom, Hydrometric Register and Statistics 1996–2000. Cen-tre for Ecology and Hydrology, Wallingford, 208pp.

1213

TED

PR

OO

F

McIntyre, N., Jackson, B., Wade, A.J., Butterfield, D., Wheater,H.S., in press. Sensitivity analysis of a semi-distributed catch-ment nitrogen model. J. Hydrol.

Moore, R.V., Morris, D.G., Flavin, R.W., 1994. Sub-set of UK digital1:50,000 scale river centre-line network. NERC, Institute ofHydrology, Wallingford.

Morris, D.G., Flavin, R.W., 1990. A digital terrain model forhydrology. In: Proceedings of the 4th International Symposiumon Spatial Data Handling, Zurich, pp. 250–262.

Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting throughconceptual models. Part 1 – a discussion of principles. J. Hydrol.10 (3), 282–290.

Neal, C., Jarvie, H.P., Wade, A.J., Neal, M., Wyatt, R., Wickham,H., Hill, L., Hewitt, N., 2004a. The water quality of the LOCARPang and Lambourn catchments. Hydrol. Earth Syst. Sci. 8 (4),614–635.

Neal, C., Skeffington, R., Neal, M., Wyatt, R., Turner, H., Hill,L., Hewitt, N., 2004b. Rainfall and runoff water quality ofthe Pang and Lambourn, tributaries of the River Thames,south-eastern England. Hydrol. Earth Syst. Sci. 8 (4), 601–613.

Price, M., Downing, R.A., Edmunds, W.M., 1993. The Chalk asan aquifer. In: Downing, R.A., Price, M., Jones, G.P.(Eds.), The Hydrogeology of the Chalk of North-WestEurope. Oxford Science Publications, Clarendon Press,Oxford, pp. 35–58.

Vinten, A.J.A., Smith, K.A., 1993. Nitrogen cycling in agriculturalsoils. In: Burt, T.P., Heathwaite, A.L., Trudgill, S.T. (Eds.),Nitrate: Processes, Patterns and Management. Wiley, Chiches-ter, pp. 39–73.

Wade, A.J., Neal, C., 2004. Assessing nitrogen dynamics inEuropean ecosystems: integrated measurement and modelling.Hydrol. Earth Syst. Sci. 8 (4), 597–860.

Wade, A.J., Neal, C., Soulsby, C., Smart, R.P., Langan, S.J.,Cresser, M.S., 1999. Modelling streamwater quality undervarying hydrological conditions at different spatial scales. J.Hydrol. 217, 266–283.

Wade, A.J., Durand, P., Beaujouan, V., Wessel, W.W., Raat, K.J.,Whitehead, P.G., Butterfield, D., Rankinen, K., Lepisto, A.,2002. Towards a generic nitrogen model of European ecosys-tems: INCA, new model structure and equations. Hydrol. EarthSyst. Sci. 6 (3), 559–582.

Wade, A.J., Butterfield, D., Raat, K.J., Whitehead, P.G. 2004.Effectiveness of eutrophication control by phosphorus reduc-tion: development of the INCA-P model. R&D Final Report P2-137, Science Report SC980009/SR, Environment Agency, Bristol,151pp.

Wade, A.J., Neal, C., Whitehead, P.G., Flynn, N.J., 2005. Modellingnitrogen fluxes from the land to the coastal zone in Europeansystems: a perspective from the INCA project. J. Hydrol. 304,413–429.

Water Framework Directive (2000/60/EC), OJ, L327, 1–72.Wellings, S.R., Bell, J.P., 1980. Movement of water and nutrients in

the unsaturated zone of upper chalk near Winchester, Hants.,England. J. Hydrol. 48, 119–136.

Wheater, H.S., Peach, D., 2004. Developing interdisciplinaryscience for integrated catchment management: The UK LOwlandCAtchment Research (LOCAR) programme. Int. J. Water Resour.D 20, 369–385.

Whitehead, P.G., Wilson, E.J., Butterfield, D., 1998. A semi-distributed Integrated Nitrogen model for multiple sourceassessment in Catchments (INCA): Part 1 – model structureand process equations. Sci. Total Environ. 210/211, 547–588.