Assessment of the predictive quality of simple indicator approaches for nitrate leaching from agricultural fields

lable at ScienceDirect

Journal of Environmental Management 91 (2010) 1305e1315

Contents lists avai

Journal of Environmental Management

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

Assessment of the predictive quality of simple indicator approaches for nitrateleaching from agricultural fields

Uwe Buczko a,*, R.O. Kuchenbuch b, B. Lennartz a

a Institute for Land Use, University of Rostock, Justus-von-Liebig-Weg 6, 18059 Rostock, GermanybAgricultural Analysis and Research Institute (LUFA) Rostock, Graf-Lippe-Straße 1, 18059 Rostock, Germany

a r t i c l e i n f o

Article history:Received 25 March 2009Received in revised form28 October 2009Accepted 7 February 2010Available online 3 March 2010

Keywords:Nitrate leachingNitrogen lossNon-point source pollutionAgri-environmental managementEnvironmental indicatorValidation

* Corresponding author. Tel.: þ49 381 4983103; faxE-mail addresses: [email protected] (U.

lufa.de (R.O. Kuchenbuch).

0301-4797/$ e see front matter � 2010 Elsevier Ltd.doi:10.1016/j.jenvman.2010.02.007

a b s t r a c t

Diffuse N losses from agriculture are a major cause of excessive nitrate concentrations in surface andgroundwaters. Leaching through the soil is the main pathway of nitrate loss. For environmentalmanagement, an anticipatory assessment and monitoring of nitrate leaching risk by indicator (index)approaches is increasingly being used. Although complex Nitrogen Loss Indicator (NLI) approaches mayprovide more information, relatively simple NLIs may have advantages in many practical situations, forinstance, when data availability is restricted.

In this study, we tested four simple NLIs to assess their predictive properties: 1. N balance (Nbal); 2.Exchange frequency of soil solution (EF); 3. Potential nitrate concentration in leachate (PNCL); 4. Acomposite NLI (balance exchange frequency product, BEP). Field data of nitrate leaching from two sites innortheast Germany along with published data from several sites in Germany, Scotland and the USA wereutilized.

Nbal proved to be a relatively poor indicator of Nloss for the time frame of one year, whereas itsprediction accuracy improved for longterm-averaged data. Correlation between calculated EF andexperimental data was high for single-year data, whereas it was lower for longterm-averaged data. PNCLgave no significant correlations with measured data and high deviations. The results for BEP wereintermediate between those for Nbal and EF.

The results suggest that the use of EF is appropriate for assessing N leaching loss for single-year dataand specific sites with comparable N input and management practices, whereas for longterm-averageddata, Nbal is better suited. BEP is an appropriate NLI both for single year and longterm data whichaccounts for source and transport factors and thus is more flexible than source-based Nbal and transport-based EF. However, such simplified NLIs have limitations: 1. The N cycle is not covered completely; 2.Processes in the vadose zone and the aquifer are neglected, 3. Assessment of management factors isrestricted.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Diffuse nitrogen (N) losses from agricultural fields are a majorcause of excessive nitrate concentrations in ground and surfacewaters and have been of concern since decades (e.g. Bach, 1987;Wendland et al., 1993; Ten Berge, 2002; Behrendt et al., 2003;Delgado et al., 2008). Excessive nitrate concentrations in ground-water can have toxic effects when used as drinking water and causeeutrophication in surface waters (e.g. Townsend et al., 2003;Powlson et al., 2008). Gaseous N losses in form of N2O are an

: þ49 381 4982159.Buczko), rkuchenbuch@lms-

All rights reserved.

important factor in global warming and the destruction of thestratospheric ozone layer (IPCC, 2007), whereas ammonia volatili-zation contributes to soil acidification and eutrophication (Follettand Delgado, 2002). Moreover, fertilizer and manure N that is notused by growing crops but lost to the environment represents aneconomic loss.

For management and environmental planning purposes, it isnecessary to assess the risk and magnitude of diffuse N losses fromagricultural fields and how they are influenced by factors such asmanagement practices, climate and soil etc. (Meisinger andDelgado, 2002; Havlin, 2004). Utilization of experimentalmethods to determine N loading, such as analysis of leachate waterobtained by suction cups (Sieling and Kage, 2006), monolith-lysimeters (Bohne et al., 1997; Knappe et al., 2002), analysis ofpercolate from tile drains (Kladivko et al., 2004; Tiemeyer et al.,

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e13151306

2008), N concentrations in groundwater (De Ruijter et al., 2007),and also measurements of mineral nitrate content in the soil profile(Wehrmann and Scharpf, 1979), is restricted in practice. The mainreasons for this are that routine application of such labor-demanding methods is mostly not viable, measurements can bemade only afterwards, and the experimental data are often notsuited for generalization (because of the effects of different yearswith varying weather patterns, different management practices,fertilizer application rates, etc.). Moreover, it is often difficult toexplain results from direct measurements, since such data are inmost cases integrative and do not allow to separate the effects ofsoil, climate and management (e.g. Bockstaller et al., 2008).

On the other hand, physically based N transport models havebeen developed since decades (e.g. Cannavo et al., 2008). With suchmodels, it ise at least in principlee possible to quantify N losses forvarious environmental conditions and agricultural managementpractices. However, complex models require many input data,contain many parameters whose values are often not sufficientlyknown, and are often difficult to operate. This restricts their routineuse for assessment of Nloss from agricultural fields and also ona larger regional scale.

As an alternative, simplified qualitative or semi-quantitativeNitrogen Loss Indicators (NLI) have been developed and discussed(Follett et al., 1991; Schröder et al., 2004; Pervanchon et al., 2005;Delgado et al., 2008; Bockstaller et al., 2008). NLIs are a subset of(agri-)environmental indicators (Bockstaller et al., 2008; Makowskiet al., 2009). The great number of different NLI approaches differwith respect to their complexity, factors of Nloss, data requirementsand type of output (e.g. risk classes, quantified amounts of Nlossetc.). For instance Shaffer and Delgado (2002) presented variousleaching indices used in the USA; in Canada, the IROWC-N NLI hasbeen developed (De Jong et al., 2009). Several recent, more elabo-rated NLIs consider also Nloss by denitrification, ammonia volatil-ization, surface runoff or erosion: e.g. the “NIT-1” NLI has beendeveloped in the USA (Delgado et al., 2008), but also applied inother countries such as Spain (De Paz et al., 2009); or the French“IN” NLI (Pervanchon et al., 2005). Although even these “complex”NLIs are considerably simpler and quicker to use than full-fledged Nmodels, the utilization of very simple NLIs may be warranted insituations with restricted data availability. For instance, N balancesare commonly being used as an indicator for N losses at the land-scape and field scale (Goodlass et al., 2003; Drury et al., 2007).However, correlations of N balances with measured N losses oftenproved to be weak (e.g. Schröder et al., 2004; Sieling and Kage,2006; De Ruijter et al., 2007; Rankinen et al., 2007).

To support a decision as to which NLI method could be suited fora given site, management option, and data availability, outputvalidation of different NLIs using experimental data would beuseful. Such studies are relatively scarce: for instance, in theNetherlands several source-based NLIs were tested and compared(Ten Berge, 2002; De Ruijter et al., 2007), in France, output vali-dations of simple source-based NLIs using the “Receiver OperatingCharacteristic” methodology were conducted (Bockstaller et al.,2008; Makowski et al., 2009).

The objectives of this study were to test some simple NLIs usingfield data of Nloss and to compare the results obtained withdifferent NLI methods. We restrict the testing procedure here toa relatively small number of some basic NLIs. To simplify theanalysis, only leaching through the soil profile was considered. It isgenerally accepted that leaching through the soil profile is thedominant pathway of Nloss in many situations, and the impact byleaching losses of N on ground and surface waters is more directthan the environmental impact by gaseous N losses and surfaceerosion losses. We used experimental field data of Nloss both fromNortheast Germany (Miegel and Zachow, 2006; Tiemeyer et al.,

2008) and published data from different geographical regions,with different crop types and methods of Nloss measurement.

2. Material and methods

2.1. Field sites/experimental data

We utilized N leaching data from two field sites inMecklenburg-Vorpommern (NE Germany). We extended the database withpublished data from field experiments in Germany, Scotland, andNorth America (Tables 1 and 2). The field experiments encompassdifferent measurement methods of Nloss and different overallduration, a wide range of crop types, N input rates (Table 1), soiltextures, water holding capacity of the root zone at field capacity(WHC(rz)), seepage rates, rates of N leaching, residual soil N (RSN)and N export by harvested crops (Table 2). Additional informationabout the sites can be found in the references given in Table 1.

2.2. N loss indicators

We tested four relatively simple NLI approaches:

1 A field scale annual N balance (Nbal);2 The exchange frequency (EF) approach;3 The potential nitrate concentration in leachate (PNCL);4 A composite NLI, calculated as the product of EF and Nbal(“BEP” for “balance exchange frequency product”).

Criteria for choosing these NLIs were their simplicity, wide-spread use (in Germany at least) and good documentation (with theexception of BEP). Moreover, they cover the spectrum of purelysource-based NLIs (Nbal), transport-based NLIs (EF), and NLIswhich contain both source and transport factors (PNCL and BEP).

2.2.1. N balancesIn general, N balances can be calculated for a whole farm, for the

soil surface, or for the soil system (Oenema et al., 2003). In its mostbasic form, Nbal comprises merely N application rates (in form ofmineral fertilizer and manure) minus N export by harvestedproducts (e.g. Bach, 1987; Sieling and Kage, 2006), since thenecessary data for calculating a complete soil system N balancewhich accounts for all possibly relevant components of input(fertilizer/manure application, atmospheric deposition, minerali-zation, N from crop residues, N-fixation) and output (extraction byharvested crops, immobilization, ammonia volatilization, denitri-fication, soil leaching, erosion, surface runoff), are often not avail-able, especially under non-equilibrium conditions (i.e., shiftingcultivation, changing nutrient inputs, changing soil carbon pool)(Oenema et al., 2003). N balances have been used as an indicator forN losses in several studies (e.g. Delgado, 1998; Van Eerdt and Fong,1998; Jansons et al., 2003; Sieling and Kage, 2006; Rankinen et al.,2007; Schröder and Neeteson, 2008). However, it has been reportedthat the N balance is often a poor predictor of N amounts lostactually to the environment and only weakly correlated with Nlosses actually measured, at least on the field scale and for single-year data (Schröder et al., 2004; Sieling and Kage, 2006; De Ruijteret al., 2007; Rankinen et al., 2007). This is mainly because Nbal doesnot consider soil hydrology and water movement (Shaffer andDelgado, 2002), and dynamic changes in the N status of the soil(mineralization, immobilization) (Oenema et al., 2005). A furtherdrawback is that N balances per se yield no information about thepathways and factors of Nloss. Correlations between Nbal and Nlossare usually improved when longer-term (i.e., several years) data areconsidered (e.g. Haferkorn, 2000; Sieling and Kage, 2006).

Table 1Field sites and datasets used for testing the NLI approaches.

Site References Type ofdataa

Duration(years)

Crops N input rate(kg N ha�1 y�1),mean (min.max)

Atmospheric N deposition(kg N ha�1 y�1), mean(min.max)

Ames (Iowa, USA) Baker and Johnson (1981) Tile 5 Maize, soybean, oats 96 (9.261) 9.8 (7.11)Brandis (L1 e 11) (Sachsen,

Germany)Haferkorn (2000) Lysi 17 Winter wheat, barley,

potato, sugar beet131 (18.217) 36.2 (18.69)

Central Iowa (USA) Jaynes et al. (2001) Tile 4 Corn-soybean rotation 174 (70.220) 12.8 (11.15)Dummerstorf (Mecklenburg-

Vorpommern, Germany)Tiemeyer et al. (2008) Tile 4 Winter wheat-oilseed

rape-maize-peas241 (198.278) 14.5 (e)

Glencorse (Scotland) Vinten et al. (1994) Tile 6 Spring barley 89 (67.5.131) naNo 3 field, (Scotland) Vinten et al. (1994) Tile 4 Spring barley 130 (127.5.135) naGroß-Lüsewitz '97

(Mecklenburg-Vorpommern,Germany)

Bohne et al. (1997) Lysi 4 Potatoe, rye, wheat,barley, sugar beet

70 (0.140) na

Groß-Lüsewitz '06(Mecklenburg-Vorpommern,Germany)

Miegel and Zachow(2006)

Lysi 13 Potatoe, rye, rape,maize, wheat, barley

160 (e) na

Hohenschulen (Schleswig-Holstein,Germany)

Sieling et al. (1997),Sieling and Kage (2006)

Suction cups,drainagerates (est.)

9 Oilseed rape, winterwheat, winter barley

183 (93.340) na

Lamberton '97 (Minnesota, USA) Randall et al. (1997) Tile 6 Maize, soybean, alfalfa 137 (35.180) naLamberton '04 (Minnesota, USA) Strock et al. (2004) Tile 3 Maize 134(e) naPurdue (Indiana, USA) Kladivko et al. (2004) Tile 15 Maize 260 (205.307) naWaseca (Minnesota, USA) Randall et al. (2003) Tile 13 Maize 143 (135.150) naWillsboro 1 (New York, USA) Sogbedji et al. (2000) Tile 3 Maize 85 (22.134) naWillsboro 2 (New York, USA) Sogbedji et al. (2000) Tile 3 Maize 85 (22.134) naWittkoppenberg (Niedersachsen,

Germany)Pamperin (2002) Suction cups,

soil N, Nbal3 Winter rye-barley 102 (0.177) na

a Tile: tile drainage effluent; Lysi: Monolith-Lysimeter.

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e1315 1307

For the present comparative study, simple N balances, calcu-lated as the difference between input as fertilizer/manure minus Nexport by harvested crops were used. Atmospheric deposition wasincluded if data were given in the original studies (Table 1). Wecalculated Nbal values based on the data given in the originalreferences. If Nbal values were given in these references, ourcalculations matched those provided by the authors.

2.2.2. Exchange frequency of the soil solutionThe EF is often used in Germany as a NLI (Frede and Dabbert,

1999; Müller, 2004). It is calculated as the ratio of a transport

Table 2Data on the field sites and datasets used for testing the NLI approaches.

Site Soil texture WHC(rz) (mm)(aWHC/tWHC)

Seepage rate (mm y�1),mean (min.max)

Ames Silt loam 242/385 130 (84.216)Brandis L1 Silt loam 193/374 113 (7.314)Brandis L4 Loam 179/252 135 (24.352)Brandis L5 Sandy loam 93/143 175 (96.381)Brandis L7 Loam 171/247 98 (24.269)Brandis L8 Sandy loam 161/207 146 (50.351)Brandis L9 Silt loam 257/407 57 (0.184)Brandis L10 Silt loam 321/529 50 (0.184)Brandis L11 Sandy loam 128/197 114 (18.292)Central Iowa Silty clay loam 184/552 262 (127.320)Dummerstorf Sandy loam 150/235 180 (93.302)Glencorse Clay loam 177/450 547 (499.622)No 3 field Sandy loam 220/321 572 (515.649)Groß-Lüsewitz '97 Sandy loam 126/134 204 (140.290)Groß-Lüsewitz '06 Sandy loam 126/134 180 (40.290)Hohenschulen Sandy loam 162/323 292 (0.490)Lamberton '97 Clay loam 135/375 121 (0.510)Lamberton '04 Clay loam 135/375 219 (10.465)Purdue Silt loam 188/281 177 (67.325)Waseca Clay loam 134/407 161 (0.434)Willsboro CL Clay loam 134/407 148 (43.281)Willsboro LS Loamy sand 186/250 194 (121.296Wittkoppenberg Loamy sand 155/215 205 (143.317)

(SeepRate, or groundwater recharge, in mmy�1) and a storage term(WHC(rz)):

EF ¼ SeepRate=WHCðrzÞ � 100 (1)

Thus, EF has units of % per year. In this approach, it is assumedthat water moves through the soil profile as a homogeneous front(i.e., no preferential flow) and that no surface runoff occurs. Boththe available water holding capacity (aWHC) (Hölting et al., 1995)and the total water holding capacity (tWHC) (Frede and Dabbert,1999; Kersebaum et al., 2006) were used as storage terms in this

Measured N losses(kg N ha�1 y�1),mean (min.max)

Residual soil nitrate(kg N ha�1)

N export by harvested crops(kg N ha�1 y�1), mean(min.max)

37.3 (18.72) na 96.2 (0.172)27.7 (.2.97) na 124.8 (37.214)33.4 (2.108) na 117.3 (26.210)43.8 (14.105) na 87.9 (19.155)13 (1.58) na 121.9 (25.200)

25.8 (5.81) na 103.3 (21.157)5.6 (0.37) na 161.2 (31.250)45 (0.280) na 178.1 (24.423)

33.1 (0.142) na 127.2 (29.250)38 (13.61) 30 (17.46) 143.2 (53.208)

26.4 (16.52) na 182.5 (161.200)20 (6.33) na 70

39.3 (38.47) na na64.5 (51.94) 123 (33.250) na35.4 (5.70) na na33.2 (0.57) na 104.3 (75.3.137.7)35.4 (0.91) 85 (11.264) 149.4 (31.380)23.7 (1.46) 75 (40.108) 117.9 (96.2.130)31.3 (8.96) na 107.5 (44.2.157.3)24.6 (0.87) 123 (26.337) na9.5 (0.25) 44 (2e98) 160.1 (103.210)

17.6 (9.31) 30 (9.75) 162.6 (126.197)42.2 (4.85) 40 (6.87) 76.8 (20.132)

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e13151308

study (yielding “EF(a)” and “EF(t)”). A low EF corresponds to a longresidence time of water in the root zone and therefore to a low riskfor nitrate leaching. For the data used in the present study, directmeasurements of seepage rates were available, whereas WHC(rz)was assessed based on soil texture, bulk density, and soil organicmatter content. The effective rooting depth was estimated from soiltexture and bulk density (Müller, 2004).

2.2.3. Potential nitrate concentration in leachateThe PNCL in its original form (Bach, 1987) was used here:

PNCL ¼ Nbal=SeepRate� 100 (2)

where the factor “100” accounts for the transformation into units ofmg NO3eN L�1. PNCL is an indicator for the expected mean nitrateconcentrations in the leachate. In the original form (Bach, 1987), itis assumed that no net mineralization/immobilization occurs in thesoil (i.e., equilibrium conditions), and that the entire N surplus islost by leaching through the soil profile (i.e., no ammonia volatili-zation, no denitrification losses, no surface runoff). A more recentform of PNCL accounts for N mineralization and immobilization,and denitrification in the root zone (Frede and Dabbert, 1999).However, since these datawere not available formost datasets usedin this study, the more simple original form of PNCL was used.

2.2.4. Composite NLIsThe composite NLI used here is calculated as the product of EF

and Nbal and thus includes N sources, weather data, and soilproperties. Depending on whether EF(a) or EF(t) are used, twodifferent NLI formulations are obtained, “BEP(a)” and “BEP(t)”:

BEPðaÞ ¼ ðEFðaÞ � NbalÞ=100 (3)

BEPðtÞ ¼ ðEFðtÞ �NbalÞ=100 (4)

The multiplicative approach accounts for the fact that Nloss isdirectly proportional to both seepage rate (e.g. Juergens-Gschwind,1989; Sieling and Kage, 2006) and Nbal (e.g. De Ruijter et al., 2007).

0102030405060708090

100110120130140150

0 50 100 150 2

Nlo

ss (k

g N

ha-

1 y-

1)

Fig. 1. Illustration of the linear regression procedure for comparison of Nloss with EF(a) anmultiplied with the slope of the overall regression of Nloss vs EF(a) (0.236, solid line), or N

2.3. Comparison of NLIs with experimental data

The NLI approaches described in Section 2.2 were comparedwith measured NO3eN losses (Nloss) and NO3eN concentrations(Nconc) in the leachate. The comparisons were performed bothfor data of single years and data averaged over several years(2e17 years, depending on the site, Table 1). We utilized linearcorrelation analysis (correlation coefficient, R) as a measure forprecision and difference measures between calculated andmeasured values as a measure of accuracy of the NLIs (Yanget al., 2000). Correlation coefficients were calculated for alldatasets pooled and for separate sites. As a measure for thedifference between the respective NLI and measured data, rootmean square error (RMSE) and mean error (ME) were calculatedaccording to:

RMSE ¼"Xni¼1

ðNLI�measuredÞ2=n#0:5

(5)

ME ¼Xni¼1

ðNLI�measuredÞ=n (6)

With n denoting the number of observations. Whereas RMSEis a measure for the absolute magnitude of the differences, MEindicates the direction of differences (over- or under-prediction).It has been found that ME and RMSE are negatively correlatedwith each other (Yang et al., 2000). We calculated RMSE and MEdirectly for the comparisons Nbal or BEP vs Nloss and PNCL vsNconc. For the comparison of EF vs Nloss, RMSE and ME couldbe calculated only after a data transformation procedure, sinceEF and Nloss per se have different physical units and aretherefore not directly comparable. EF values were re-scaled bymultiplication with the overall slope of linear regression of Nlossvs EF (Fig. 1), i.e., 0.236 for EF(a) and 0.406 for EF(t). Effectively,EF values were transformed to a comparable scale as Nlossvalues.

y = 0.236x

y = 0.406x

00 250 300 350 400EF

EF(a)EF(t)

d EF(t). For calculation of ME and RMSE values (Tables 3, 4 and 6, ES1), EF values wereloss vs EF(t) (0.406, dashed line).

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e1315 1309

3. Results

3.1. N balance

For all datasets pooled, the correlation between Nbal and Nlossis rather low when single-year data are considered (althoughsignificant at the 0.01 level, Table 3, Fig. 2). Fig. 2 shows a largescatter, some datasets even show large negative Nbal values(<�150 kg N ha�1 y�1) with high Nloss. The correlation is distinctlyhigher when data averaged over several years are considered(Table 4). On the other hand, correlations between Nbal and Nconcin the leachate were not significant, neither for single year (Table 3)nor longterm-averaged data (Table 4). ME values are negative bothfor single year and longterm data, i.e., overall, average Nbal waslower than Nloss. The RMSE between Nbal and Nloss is70.4 kg N ha�1 y�1 for single years and 46.3 kg N ha�1 y�1 forlongterm-averaged data and thus exceeds average Nloss for indi-vidual sites in many cases (cf. Table 2).

For separate study sites, the correlation between Nbal and Nlossis highly variablewhendata for singleyears are considered (Table 5):Whereas for Dummerstorf, a negative correlation (with R ¼ �0.56)was found, small positive correlation coefficients are calculated formost other sites. Accordingly, RMSE values between Nbal and Nlossare mostly relatively high with values between 30 and 90 for mostsites for single-year data, but distinctly lower for longterm data. MEvalues are negative for 13 of the 20 datasets for annual data, indi-cating that Nloss is higher than Nbal. The correlation coefficients forannual data are relatively high for Central Iowa, Purdue, and Witt-koppenberg (see Section 4.1).

3.2. Exchange frequency of soil solution

The Overall correlation between EF and Nloss is higher than forNbal vs Nloss when data for single years are considered (Table 3),although Fig. 1 shows a large scatter. Both EF(a) and EF(t) yieldsimilar, highly significant, correlation coefficients. In contrast to theother NLI approaches, the correlation between EF and Nloss islower for longterm-averaged data than for annual data (Table 4).The overall correlation between EF and Nconc is negative, both foryearly (Table 3) and longterm data (Table 4), and correlation coef-ficients are also in this case lower for longterm than for yearly data.RMSE values for yearly data are about 25 for both EF(a) and EF(t),whereas ME values are slightly negative for annual data, and evenmore negative for longterm data (Table 4).

Table 3Number of datasets (n), correlation coefficients, RMSE and ME for comparisons ofNLI approaches (Nbal, EF(a), EF(t), PNCL, BEP(a), BEP(t)) with experimental data ofNloss (all datasets), yearly data.

n Nloss Nconc RMSEd MEd

Correlation coefficient

Nbal 269 0.25a 0.04 70.4 �6.7EF(a) 305 0.48a �0.20a 24.6 �5.0EF(t) 305 0.47a �0.16a 25.0 �4.3PNCLþb 175 �0.14 0.12 51.5 23.0PNCL(all)c 269 0.08 �0.006 325.0 �36.5BEP(a) 269 0.34a 0.050 70.8 2.6BEP(t) 269 0.32a 0.05 47.3 �10.1

a Correlation is significant at the 0.01 level.b Negative PNCL values not included in analysis; one outlier removed

(PNCL > 3000).c Including negative PNCL values and outliers.d RMSE and ME were calculated directly for the comparisons Nbal, BEP(a), BEP(t)

vs Nloss and PNCL vs Nconc. For the calculation of RMSE and ME for EF vs Nloss, EFvalues were linearly transformed (i.e., EF(a) multiplied by 0.236 and EF(t) by 0.406,see 2.3 in the main text for more details).

For separate sites, correlations between EF and Nloss are simi-larly high (Table 6 for EF(a) data, note that due to space constraintsthe corresponding table for EF(t) can be found in the electronicSupplement as Table ES1). EF(a) and EF(t) yield equal correlationcoefficients, because at a given site there was mostly only one soiltype, which has therefore a fixed difference between aWHC andtWHC (exception: site Willsboro with two different soils). RMSEvalues of EF vs Nloss for separate sites and yearly data are highlyvariable. For some sites (e.g. Brandis L10, Glencorse), RMSE valuesare larger than 50. For some sites, RMSE values are lower for EF(t) vsNloss than for EF(a) vs Nloss (9 of the 20 sites for yearly data),whereas it is vice versa for other sites (11 of the 20 sites). Corre-spondingly, ANOVA yielded no significant differences. For longtermdata, ME and RMSE values of separate sites are not distinctlydifferent from single-year data.

3.3. Potential nitrate concentration in leachate

Due to negative values for Nbal, many datasets have negativePNCL values. Concentrations per definition cannot be negative. Onthe other hand, when PNCL is viewed merely as an indicator,negative PNCL values may have some meaning. In Tables 3 and 4both cases are compared: without negative PNCL values (PNCLþ),and with both positive and negative PNCL values included (PNCL(all)). Although the results are different for both cases, correlationsbetween PNCL and Nconc or Nloss are not significant (exception:PNCL(all) vs Nloss for longterm data). ME and RMSE values arelower for longterm than for annual data. Conspicuously, the overallcorrelation between PNCL(all) and Nloss is relatively strong andstatistically significant for longterm data.

Since PNCL is interpreted as a concentration, presumably theexclusion of negative PNCL values is logically more appealing, andfor separate study sites, only this case is considered (Table 7). Thecorrelation between PNCL and Nconc is highly variable, with bothpositive and negative correlation coefficients and ME values(Table 7). RMSE values are in general lower when longterm-aver-aged data are considered (Table 8), whereas ME values exhibit bothpositive and negative values, irrespective of whether correlationcoefficients are positive or negative.

3.4. Composite NLIs

Both composite NLI approaches (BEP(a) and BEP(t)) showsignificant positive correlation with Nloss when all data are pooled(Tables 3 and 4). The correlation is higher for longterm data (Table 4)than for single years (Table 3). There is no significant correlationbetween BEP and Nconc. RMSE values between BEP and Nloss are inthe same range or lower than corresponding values for Nbal.

For separate sites, the correlations between BEP and Nloss arehighly variable but mostly positive (Table 8 for BEP(a) data, notethat due to space constraints the corresponding table for BEP(t) canbe found in the electronic Supplement as Table ES2). WhereasRMSE values between BEP and Nloss are mostly lower compared toRMSE values between Nbal and Nloss when single-year data areconsidered, the reverse is true for longterm data.

For pooled data, correlation between BEP(a) and Nloss is slightlyhigher than for BEP(t) vs Nloss, both for single years (Table 3) andlongterm data (Table 4). On the other hand, RMSE is distinctlyhigher for BEP(a) vs Nloss than for BEP(t) vs Nloss. ME values arecloser to zero for BEP(a) vs Nloss than for BEP(t) vs Nloss (Tables 3and 4), although they are very similar. For separate sites, RMSE is forall sites higher for BEP(a) vs Nloss than for BEP(t) vs Nloss, whereasME values are closer to zero for BEP(t) vs Nloss for more sites (9 outof 20) than for BEP(a) vs Nloss (7 out of 20).

R² = 0.064

R² = 0.190

0

20

40

60

80

100

120

140

-200 -100 0 100 200

Nbal (kg N / hay)

Nlo

ss (k

g N

/ ha

y)

yearly values

longterm values

Linear (yearly values)

Linear (longtermvalues)

Fig. 2. Nbal vs Nloss for all datasets. Data are for single years (“yearly values”) or averaged over 2e17 years (depending on the site) (“longterm values”).

Table 4Number of datasets (n), correlation coefficients, RMSE and ME for comparisons ofNLI approaches (Nbal, EF(a), EF(t), PNCL, BEP(a), BEP(t)) with experimental data ofNloss (all datasets), longterm average data.

n Nloss Nconc RMSEd MEd

Correlation coefficient

Nbal 69 0.44a 0.08 46.3 �8.6EF(a) 79 0.37a �0.36a 18.6 �7.4EF(t) 79 0.30a �0.38a 20.0 �6.2PNCLþb 53 �0.05 0.07 23.8 4.6PNCL(all)c 69 0.35a �0.03 41.3 �11.5BEP(a) 69 0.50a 0.06 47.5 �.5BEP(t) 69 0.47a 0.09 34.5 �14.2

a Correlation is significant at the 0.01 level.b Negative PNCL values not included in analysis; one outlier removed

(PNCL > 3000).c Including negative PNCL values and outliers.d RMSE and ME were calculated directly for the comparisons Nbal, BEP(a), BEP(t)

vs Nloss and PNCL vs Nconc. For the calculation of RMSE and ME for EF vs Nloss, EFvalues were linearly transformed (i.e., EF(a) multiplied by 0.236 and EF(t) by 0.406,see 2.3 in the main text for more details).

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e13151310

4. Discussion

4.1. N balance

N balances in the simplified form used here proved to be a poorindicator for possible N leaching losses for the time scale of a singleyear and a slightly better indicator for longterm data. This is inaccordance with many previous studies (e.g. Oenema et al., 2003;Sieling and Kage, 2006; De Ruijter et al., 2007). The relativelyhigh correlation for Central Iowa and Wittkoppenberg may beexplained by the fact that at these sites, several different N fertilizerrates were applied and compared in the same year and the samecrop, in contrast to most other sites. For Purdue, the high correla-tion may be due to the long duration of the experiment withcomparable management practices. In contrast, at Brandis witha similar overall duration of the experiment, the crop rotation washighly variable from year to year, and consequently the correlationcoefficients of Nbal vs Nloss are relatively low (Table 5). This indi-cates that Nbal can give an indication of the Nloss risk for specificmanagement practices and over longer periods. These findings maybe explained by several factors.

Presumably, not all relevant N input and N output factors wereincluded in the calculation of the N balance. Moreover, equilibriumbetween input and output was probably not attained at a temporalscale of only one year. N mineralization and immobilization andtherefore changes in the soil N pool were not accounted for in thecalculation of Nbal. The soil N pool can be large for fine-texturedsoils with large organic matter content. At such sites, appreciableNloss can be observed even for negative Nbal values, for instanceseveral of the lysimeters at Brandis (cf. Oenema et al., 2005). Crops,

yields, N inputs as fertilizer/manure and weather conditions werevariable from year to year for many of the sites considered in thisstudy (Tables 1 and 2). Consequently, N immobilization andmineralization were probably also variable and not negligible. Thiswould have to be included in the N balance calculation for a morereliable prediction of N losses at the temporal scale of single years.However, direct measurement of N mineralization and immobili-zation is expensive and labor demanding and an estimation is notstraightforward due to the variety of influencing factors, such as Napplication rate, soil type, crop type, climate/weather, and

Table 5Comparison of Nbal vs Nloss for separate sites (note that number of sites is lower for Nbal vs Nloss than for EF vs Nloss because not for all sites Nbal could be calculated).

Site Annual data Longterm data

Mean (min.max) n R RMSE ME RMSE ME

Ames �0.4 (�71.132) 10 0.00 81.2 �37.7 42.4 �37.7Brandis L1 6.4 (�130.84) 17 0.22 55.3 �21.3 12.0 �9.9Brandis L4 13.9 (�177.93) 17 0.28 64.8 �19.4 9.8 �4.9Brandis L5 43.3 (�114.114) 17 0.22 57.3 �.5 19.3 16.2Brandis L7 9.3 (�106.90) 17 0.08 45.3 �3.7 8.0 5.8Brandis L8 27.9 (�116.86) 17 0.20 50.2 2.1 15.5 13.9Brandis L9 �29.9 (�186.81) 17 0.06 69.9 �35.2 26.3 �25.2Brandis L10 �36.6 (�170.42) 17 0.20 91.1 �61.3 57.7 �57.3Brandis L11 4.0 (�144.63) 17 0.18 66.2 �30.5 29.2 �20.3Central Iowa 30.7 (1.124) 12 0.76 27.7 �7.3 11.7 �7.3Dummerstorf 58.0 (37.79) 8 �0.56 44.3 31.6 31.6 31.6Glencorse 20.0 (.) 1 e 13.9 13.9 e e

Hohenschulen 48.8 (�14.119) 15 �0.01 58.1 �1.6 31.7 �8.0Lamberton 23.2 (�71.126) 15 �0.29 66.8 �9.7 23.4 �21.1Purdue 100.0 (�62.201) 45 0.39 102.9 68.8 94.3 91.2Willsboro �76.9 (�136.�32) 17 0.29 94.9 �90.7 92.0 �88.9Wittkoppenberg 25.3 (�38.92) 12 0.37 43.0 �16.8 34.6 �16.8

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e1315 1311

management practices (e.g. tillage) (Engels and Kuhlmann, 1993;Vigil et al., 2002).

In this context, one has to consider that most of the fieldexperiments encompassed only one N application rate per site andyear, and no comparisons of different N application rates in parallelplots. In studies where several N rates applied in the same year andat the same site are compared there is a relatively high correlationbetween Nbal and Nloss (e.g. Central Iowa, Jaynes et al., 2001).However, Nbal is influenced by N application rates in various ways:In general, higher N application rates increase the N supply to cropsand increase yield and consequently N export by harvested crops;on the other hand, net N mineralization of soil organic matter maydecrease with increasing N application rate (Engels and Kuhlmann,1993; Sieling and Kage, 2006). It has been often described that Nleaching losses are weakly correlated with N application ratesbelow a threshold value and increase distinctly for N applicationrates exceeding a soil and crop-specific optimum (e.g. Ten Berge,2002; Sieling and Kage, 2006). Moreover, timing (Randall et al.,2003), type (slurry, manure, mineral fertilizer) (Sieling and Kage,2006), incorporation and tillage practices (Franklin et al., 2007)often exhibit a significant influence on N losses besides N appli-cation rates.

Table 6EF(a) vs Nloss for separate sites.

Site Annual data

Mean (min.max) n R

Ames 53.8 (34.7.89.3) 10 0.16Brandis L1 58.7 (3.6.162.7) 17 0.61Brandis L4 75.8 (13.4.197.2) 17 0.63Brandis L5 188.5 (103.2.409.7) 17 0.28Brandis L7 57.4 (14.0.157.3) 17 0.29Brandis L8 91.0 (31.1.218.0) 17 0.56Brandis L9 37.6 (.8.71.6) 10 0.56Brandis L10 28.9 (3.4.57.4) 8 0.14Brandis L11 88.9 (14.0.227.7) 17 0.75Central Iowa 142.4 (68.9.174.1) 12 0.74Dummerstorf 121.6 (60.9.209.9) 8 0.93Glencorse 309.2 (281.9.351.4) 5 �0.47Groß-Lüsewitz 142.9 (31.7.230.2) 13 0.58Hohenschulen 174.1 (80.5.303.4) 26 0.34Lamberton 115.2 (.0.362.2) 15 0.86No 3 field 260.0 (234.1.295.0) 3 �0.41Purdue 94.5 (35.7.173.3) 45 0.45Waseca 120.4 (.0.325.3) 13 0.88Willsboro 107.5 (32.2.210.5) 18 0.32Wittkoppenberg 132.3 (92.3.204.5) 12 0.62

The preponderance of negative ME values both for single yearand longterm data may be explained by the numerous sites witha negative Nbal considered in this study, which exhibited never-theless considerable Nloss. Especially several of the Brandislysimeters had negative N balances during several years of thestudy period, but showed considerable Nloss. This was obviouslycaused by non-equilibrium conditions (Haferkorn, 2000). Theseconditions were imposed by varying crop types, weather condi-tions and N application rates; moreover, several of the lysimeterswere taken from sites with climatic conditions differing from thoseat Brandis and consequently, the N cycle had to adjust to siteconditions of Brandis; it is not clear whether this adjustment hadbeen totally accomplished within the study period of 17 years(Haferkorn, 2000).

Another factor is atmospheric deposition which has beenaccounted for in only a part of the study sites (Table 1). At Brandis,atmospheric deposition rates were highly variable and amountedup to 69 kg N ha�1 y�1. Consequently, for this site, it was notnegligible for the overall Nbal. For the other sites with explicitatmospheric deposition data (Table 1), values are about5e15 kg N ha�1 y�1. In those cases where atmospheric depositionwas not given in the original references, we assumed that it was

Longterm data

RMSE ME RMSE ME

30.1 �24.6 27.1 �24.626.1 �13.8 18.3 �18.126.4 �15.5 21.4 �21.129.5 .6 3.7 �3.115.0 .5 5.9 �4.318.7 �4.4 9.7 �9.28.7 �.1 2.7 �1.4

57.4 �42.6 30.3 �29.932.0 �12.1 22.2 �16.810.6 �4.5 9.1 �4.55.9 2.2 5.3 5.3

54.8 52.8 52.8 52.817.6 �1.7 .0 36.129.8 �4.2 27.7 �10.821.3 �7.9 25.0 �14.224.0 21.9 21.9 21.918.2 �9.0 14.1 �13.113.4 .0 .0 3.718.4 12.3 9.3 8.224.2 �11.0 12.9 �11.0

Table 7PNCL vs Nconc for separate sites (negative PNCL values not included in analysis).

Site Annual data Longterm data

Mean (min.max) n R RMSE ME RMSE ME

Ames 58.3 (5.4.149.0) 4 �0.05 66.3 26.3 14.8 �14.8Brandis L1 41.6 (1.4.108.7) 10 �0.20 44.5 19.5 10.7 �8.9Brandis L4 40.7 (3.5.116.0) 12 �0.29 38.9 13.0 7.2 �3.6Brandis L5 40.0 (2.2.103.1) 15 �0.16 36.2 12.5 11.5 9.7Brandis L7 52.5 (.6.139.3) 11 �0.24 68.1 37.5 9.7 7.3Brandis L8 42.0 (2.6.103.7) 14 �0.05 40.0 23.7 11.1 9.9Brandis L9 1038.0 (.7.3100.0) 3 0.93 1761.0 1014.0 e e

Brandis L10 16.1 (6.8.33.3) 3 0.99 100.0 �68.4 e e

Brandis L11 45.9 (1.0.155.6) 11 0.25 47.1 16.1 26.2 �18.0Central Iowa 10.9 (.3.43.5) 12 0.69 10.6 �44.3 4.5 �2.8Dummerstorf 41.6 (12.2.84.8) 8 0.23 37.7 27.0 15.8 15.8Glencorse 3.2 (�) 1 e 2.2 2.2 e e

Hohenschulen 21.2 (1.7.56.5) 14 0.04 20.8 �1.5 12.2 �3.1Lamberton 24.8 (2.2.70.2) 7 �0.16 28.7 4.2 17.1 �14.7Purdue 90.0 (38.0.223.4) 36 0.48 80.1 69.9 61.3 56.6Wittkoppenberg 20.5 (7.7.31.5) 9 �0.25 17.3 �0.6 3.2 0.2

Table 8BEP(a) vs Nloss for separate sites.

Site Annual data Longterm data

Mean (min.max) n R RMSE ME RMSE ME

Ames 2.6 (�33.4.98.7) 10 �0.02 54.8 �34.7 38.6 �38.2

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e13151312

deemed to be negligible by the authors, and since we had no betterinformation, we used those data without correction for atmo-spheric deposition.

Ammonia volatilizationwas accounted for in some of the studiesfor freshly applied manure (e.g. Tiemeyer et al., 2008). In thesecases it had already been subtracted from the N application rates. Inother cases, for manures and fertilizers after application, estimationof ammonia volatilization losses is not straightforward anddepends on application method, fertilizer/manure type, etc. (e.g.Schröder et al., 2007; Delgado et al., 2008). Moreover, ammoniavolatilization seems to be negligible in most cases: for instance,Schröder et al. (2007) estimated N losses due to ammonia volatil-ization for injected/incorporated manure at 5% and for mineralfertilizer (Caeammoniumenitrate) at 1%. These values may behigher for grassland and for other conditions, application methodsand fertilizer types, but information about that was mostly notprovided in the original references. Moreover, grassland land usewas not included in the present study (Table 1).

Also, denitrification losses should be assessed for a complete Nbalance; although denitrification amounts typically to20e30 kg N ha�1 during the growth period of annual crops (Niederet al., 1989), it shows high variability depending primarily on soilaeration status, SOM content and nitrate content. For Germany Nlosses of agricultural areas due to denitrification have been esti-mated in a wide range of 2.6e80 kg ha�1 y�1 (Nieder et al., 1989).

The lack of correlation between Nbal and Nconc in the leachatecan be explained by the fact that Nconc is highly dependent on theleaching rate with dilution at higher seepage rates and often highconcentrations at low seepage rates (e.g. Coles and Trudgill, 1985;Haferkorn, 2000). For a given site, seepage rates are mainly influ-enced by weather and soil conditions and therefore largely inde-pendent on Nbal.

Brandis L1 7.2 (�53.6.88.8) 17 0.26 38.5 �20.5 19.6 �19.4Brandis L4 15.8 (�84.3.108.4) 16 0.30 43.8 �18.5 14.6 �13.7Brandis L5 79.3 (�120.1.261.1) 17 0.37 95.0 35.5 69.1 66.2Brandis L7 1.5 (�48.0.35.3) 17 0.25 25.2 �11.5 5.7 �5.3Brandis L8 24.8 (�39.3.114.3) 17 0.35 39.5 �1.0 10.0 8.5Brandis L9 �11.8 (�52.1.1.9) 10 �0.08 28.8 �20.7 11.1 �10.8Brandis L10 �11.5 (�79.6.8.6) 8 0.33 72.3 �60.9 36.6 �36.5Brandis L11 8.4 (�47.0.100.6) 17 0.17 54.0 �25.9 30.1 �23.0Central Iowa 48.5 (1.4.192.5) 12 0.79 48.5 10.5 17.3 5.7Dummerstorf 65.5 (32.1.107.3) 8 0.32 47.1 39.1 51.8 51.8Glencorse 70.3 1 e 64.2 64.2 41.8 41.8Hohenschulen 78.9 (�17.3.190.0) 15 0.13 78.8 28.6 42.4 16.0Lamberton 22.0 (�62.9.153.5) 11 0.24 65.4 �23.0 22.5 �17.1Purdue 126.1 (46.0.257.7) 36 0.65 101.8 92.1 86.9 82.2Willsboro �85.6 (�195.8.�20.8) 17 �0.05 114.3 �99.4 87.1 �83.0Wittkoppenberg 43.6 (�40.9.188.2) 12 0.46 66.1 1.4 42.1 �8.7

4.2. Exchange frequency of soil solution

The relatively high correlations between EF and Nloss for single-year data confirm that N leaching through the soil is highly influ-enced by soil hydrological and weather conditions. Similarly, manyprevious studies reported an increase of N leaching with increasingseepage rate (e.g. Juergens-Gschwind, 1989; Sieling and Kage,2006). The higher correlation of EF vs Nloss compared to Nbal vsNloss suggests that the use of NLI approaches which account fortransport factors alone without considering N source factors maybe justified to some extent, at least when single-year data andcomparable site conditions (especially fertilizer application rates)

are considered. The correlations of the pooled data obscure evensomewhat the fact that when separate sites are considered,correlations between EF and Nloss are in most cases higher thancorrelations for the pooled data. However, for some sites thecorrelations were notably lower: especially for the Scottish sites,negative correlation coefficients between EF and Nloss were found,which is explained by larger leaching losses in years with autumnchisel ploughing, whereas lower Nloss was observed in years whencultivation was delayed until late winter (Vinten et al., 1994).

When different sites are compared with each other, however, EFis not necessarily a good predictor of Nloss, as evidenced by thehigh variability and partially high values of RMSE and ME of EF vsNloss for separate sites.

The lower correlations between EF and Nloss for longterm dataalong with the fact that for separate sites, RMSE and ME are notconsistently smaller for longterm data than for single-year datamay be explained by the more or less erratic variation of seepagerates from year to year, depending on weather conditions. Hydro-logical conditions exert a rather short-term influence, and seepagerates in a given year have no influence on seepage rates in thefollowing year. In contrast, components of the N cycle are cumu-lative over longer periods of time (soil N pool as buffer) and Nloss inthe long run tends to approach Nbal.

Whereas both EF(a) and EF(t) yielded largely similar correlationswithNloss, the lower RMSE andME values for EF(t) vs Nloss than forEF(a) vsNloss suggest that EF(t)may be a better predictor than EF(a).

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e1315 1313

Theoretically, one would expect that EF(a) is a better predictor forNloss than EF(t), because EF(a) presumably accounts to a largerdegree for the pore space that actually participates in percolation,compared with EF(t). One would assume that pores with diameters<0.2 mm (corresponding to matric potential of the permanentwilting point at�15 000 cmwater), which are included additionallyin EF(t), contain largely immobile water which does not participatein percolation. However, this view may be too simplistic, or therelation is overshadowed by other soil-related factors which are notaccounted for in theEF, for instancemacropores. Possiblya largepartof infiltrating water and nitrate loads flows not through the porespace considered in the EF approach, i.e., pores with diameters<50 mm (corresponding to field capacity), but rather througha fraction of the larger pore space corresponding to a matricpotential above field capacity (poreswith diameters>50 mm). It hasbeen shown variously that such preferential flow events could beresponsible for the majority of water and solute fluxes to theunderground, even in soils which ostensibly contain no (visual)macropores (e.g. Buczko et al., 2006).Whereas suchmacropores arenot considered neither in tWHC nor in aWHC, differences betweenaWHC and tWHC are rather small for most soils.

Whereas Nloss increases with EF and seepage rate, Nconc isinversely proportional to EF and seepage rate. This could beexplained by dilution (attenuation) of N loads with increasingseepage (e.g. Coles and Trudgill, 1985; Tiemeyer et al., 2008).

4.3. Potential nitrate concentration in leachate

The same dilution effect as discussed for the negative correla-tion between EF and Nconc is probably the reason for the overallnegative (although not significant) correlation between PNCL andNloss (when negative PNCL values are excluded from the analysis).This concurs with the notion that PNCL is an indicator for Nconcrather than for Nloss. However, also between PNCL and Nconcvalues, correlations are not significant. Similarly as in our study, lowcorrelations between PNCL and Nconc in upper groundwater werereported by De Ruijter et al. (2007) for several sites in theNetherlands.

Presumably, the quality of PNCL as a predictor for Nloss dependsto a large degree on the quality of the N balance assessment whichwas used. The simple N balance used here might be too simplisticfor this purpose. Incorporation of mineralization and immobiliza-tion of N might improve the predictions of PNCL (Frede andDabbert, 1999), but the necessary data were not available in thisstudy, and probably this applies also to many other datasets(see Section 4.1).

These problems along with the highly erratic variations of PNCLprediction quality for different sites make the PNCL (at least in itssimple form) a poor predictor of N leaching losses. Moreover, for anenvironmental assessment of the impact of nitrate leaching, notnitrate concentration in the leachate but rather in the groundwaterare relevant. However, nitrate concentrations in groundwater aredetermined by nitrate loads transported from the root zone and thegroundwater recharge. For a specified period, N loads are theproduct of Nconc and seepage rate (cf. Meisinger and Delgado,2002). For low seepage rates, very high PNCL values may becalculated, whereas the actual impact (N load) is low due to the lowamount of seepage water. Similar conclusions were drawn by DeJong et al. (2007) in their investigation of the Indicator of Risk ofWater Contamination by Nitrogen in Canada.

4.4. Composite NLIs

The results (3.4) suggest that for annual data, BEP is a betterpredictor of Nloss than Nbal, but a poorer predictor than EF. On the

other hand, for longterm data correlations between BEP and Nlossare higher than corresponding correlations of Nbal and EF vs Nloss.Although RMSE andME values suggest that for longterm data BEP isnot necessarily a better predictor of Nloss than Nbal (if it is assumedthat the physical units are the same), the overall results indicate thatBEP gives reasonable predictions of Nloss. The appeal of BEP as anNLI consists in the fact that it includes both anN balance and soil andweather data, whereas the quality of prediction is not worse than forNbal or EF. Therefore, the effect of management options (i.e., Napplication rates) is better accounted for while the influence of soiland weather data is not neglected either.

Comparing the two different formulations, BEP(a) and BEP(t), itis conspicuous that for all data pooled, although BEP(a) showsslightly higher correlations than BEP(t) with Nloss, RMSE values forBEP(t) are distinctly lower than for BEP(a). This applies both forpooled data (Tables 3 and 4) and for separate sites (Table 8, ES2).This suggests that BEP(t) is a better predictor for Nloss than BEP(a).

4.5. General discussion

When considering the relatively simple NLI approaches inves-tigated in this study, it is important to bear in mind that the N cycleis not encompassed completely, i.e., only N application rates asfertilizer/manure, atmospheric N deposition, N uptake by harvestedcrops, and N leaching are considered. The results obtained in thiscomparative study suggest that neglecting the processes whichwere not considered probably induces some loss of predictivecapabilities of the NLI approaches.

Further, nitrate leached from the soil is effective ecologicallyonly after passing through the vadose zone as nitrate concentrationin the ground and surfacewaters. Therefore processes in the vadosezone and in the aquifer would be also of interest and the sole focuson the soil zone as in the NLIs investigated here may not be suffi-cient. From an environmental point of view, the incorporation ofparameters for the vadose zone and groundwater would hence bejustified (e.g. Shaffer and Delgado, 2002). For instance, Kersebaumet al. (2006) combined the EF approach with an indicator forgroundwater vulnerability into a groundwater pollution risk index,and groundwater vulnerability is included in several recentcomplex NLIs (e.g. Delgado et al., 2008).

Moreover, due to the lack of factors which explicitly account formanagement practices (crop type, N application rate and incorpo-ration method/timing, tillage system, fertilizer/manure type), it isnot straightforward to calculate scenarios of the effect of differentmanagement practices on Nloss. For this purpose, the use ofa composite NLI with explicit factors for N management (e.g.Delgado et al., 2008) would be necessary. The same applies to off-site factors such as tile drainage or riparian buffers.

5. Conclusions

In this study, four relatively simple NLIs were compared: Nbal, EF,PNCL and a composite NLI calculated as the product of Nbal and EF(BEP). Experimental data of Nloss comprising several datasets fromfield sites in Germany, Scotland, and North America were used fortesting the predictions of these NLI approaches (output validation).

Calculated NLI values were compared with measured Nloss andNconc using linear correlation analysis, mean errors and root meansquare errors. For single-year data, the highest correlation coeffi-cients were obtained for the comparison of EF vs Nloss. On the otherhand, when datawere averaged over several years, Nbal proved to bea better predictor of Nloss than EF. The results for BEP were betweenthose of Nbal and EF. PNCL showed no significant correlations withmeasured Nconc in leachate. Maybe a more complex formulation ofPNCL which accounts for mineralization, immobilization and

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e13151314

denitrificationwould yield better results, although the required dataare probably not available in many cases.

Consequently, EF seems to be a sufficient and simple indicatorof N leaching through soil when single-year data are consideredfor specific sites, and when the impact of management options isnot of interest. However, since a main purpose of NLIs is toinvestigate the influence of management (above all, N applicationrate) on Nloss, NLIs which account for N sources should be pref-erable. The results of this study suggest that PNCL (in its simplestform) seems not suitable as NLI. The composite indicator BEP ispreferable to Nbal, at least for single-year data, because itencompasses both source and transport components, whereas it isrelatively simple. The effect of management decisions on Nloss canbe assessed for different N application rates and N export byharvested crops. For investigating the effect of other managementpractices, more complex NLIs with explicit factors for manage-ment options could be used instead.

However, these simple NLIs have several limitations: 1. The Ncycle is not considered completely, 2. No factors for processes in thevadose zone and aquifer are incorporated, 3. No managementfactors are included, i.e., the impact of different managementpractices cannot be assessed in a straightforward manner.

Appendix. Supplementary information

Supplementary data associated with this article can be found inthe online version, at doi:10.1016/j.jenvman.2020.02.007.

References

Bach, M., 1987. Die potentielle Nitratbelastung des Sickerwassers durch die Land-wirtschaft in der Bundesrepublik Deutschland, vol. 93. Göttinger Bodenkund-liche Berichte, Göttingen, 186 pp.

Baker, J.L., Johnson, H.P., 1981. Nitrate-nitrogen in tile drainage as affected byfertilization. J. Environ. Qual. 10, 519e522.

Behrendt, H., Bach, M., Opitz, D., Pagenkopf, W.-G., Scholz, G., Wendland, F., 2003.Nutrient Emissions into River Basins of Germany on the Basis of a HarmonizedProcedure. UBA-Texte 82. Umweltbundesamt, Berlin.

Bockstaller, C., Guichard, L., Makowski, D., Aveline, A., Girardin, P., Plantureux, S.,2008. Agri-environmental indicators to assess cropping and farming systems. Areview. Agron. Sustain. Dev. 28 (1), 139e149.

Bohne, K., Gall, H., Zachow, B., 1997. Simulation von Stickstoff-Austrägen aus Lysi-metern. Z. Pflanz. Bodenk. 160, 5e13.

Buczko, U., Bens, O., Hüttl, R.F., 2006. Tillage effects on hydraulic properties andmacroporosity in silty and sandy soils. Soil. Sci. Soc. Am. J. 70, 1998e2007.

Cannavo, R., Recous, S., Parnaudeau, V., Reaug, R., 2008. Modeling N dynamics toassess environmental impacts of cropped soils. Adv. Agron. 97, 131e174.

Coles, N., Trudgill, S., 1985. The movement of nitrate fertiliser from the soil surfaceto drainage waters by preferential flow in weakly structured soils, Slapton, S.Devon. Agric. Ecosyst. Environ. 13, 241e259.

De Jong, R., Yang, J.Y., Drury, C.F., Huffman, E.C., Kirkwood, V., Yang, X.M., 2007. Theindicator of risk of water contamination by nitrate-nitrogen. Can. J. Soil Sci. 87,179e188.

De Jong, R., Drury, C.F., Yang, J.Y., Campbell, C.A., 2009. Risk of water contaminationby nitrogen in Canada as estimated by the IROWC-N model. J. Environ. Manage.90, 3169e3181.

Delgado, J.A., 1998. Sequential NLEAP simulations to examine effect of early and lateplanted winter cover crops on nitrogen dynamics. J. Soil Water Conserv. 53,241e244.

Delgado, J.A., Shaffer, M., Hu, C., Lavado, R., Cueto-Wong, J., Joosse, P., Sotomayor, D.,Colon, W., Follett, R., DelGrosso, S., Li, X., Rimski-Korsakov, H., 2008. An indexapproach to assess nitrogen losses to the environment. Ecol. Eng. 32, 108e120.

De Paz, J.M., Delgado, J.A., Ramos, C., Shaffer, M.J., Barbarick, K.K., 2009. Use of a newGIS nitrogen index assessment tool for evaluation of nitrate leaching acrossa Mediterranean region. J. Hydrol. 365, 183e194.

De Ruijter, F.J., Boumans, L.J.M., Smit, A.L., Van Den Berg, M., 2007. Nitrate in uppergroundwater on farms under tillage as affected by fertilizer use, soil type andgroundwater table. Nutr. Cycl. Agroecosyst. 77, 155e167.

Drury, C.F., Yang, J.Y., De Jong, R., Yang, X.M., Huffman, E.C., Kirkwood, V., Reid, K.,2007. Residual soil nitrogen indicator for agricultural land in Canada. Can. J. SoilSci. 87, 167e177.

Engels, T., Kuhlmann, H., 1993. Effect of the rate of fertilizer on apparent netmineralization of N during and after cultivation of cereal and sugar beet crops.Z. Pflanz. Bodenk. 154, 149e154.

Follett, R.F., Delgado, J.A., 2002. Nitrogen fate and transport in agricultural systems.J. Soil Water Conserv. 57, 402e408.

Follett, R.F., Keeney, D.R., Cruse, R.M. (Eds.), 1991. Managing Nitrogen for Ground-water Quality and Farm Profitability. SSSA, Madison, WI, p. 357.

Franklin, D., Truman, C., Potter, T., Bosch, D., Strickland, T., Bednarz, C., 2007.Nitrogen and phosphorus runoff losses from variable and constant intensityrainfall simulations on loamy sand under conventional and strip tillage systems.J. Environ. Qual. 36, 846e854.

Frede, H.-G., Dabbert, S. (Eds.), 1999. Handbuch zum Gewässerschutz in der Land-wirtschaft, second ed. Ecomed, Landsberg, p. 451.

Goodlass, G., Halberg, N., Verschuur, G., 2003. Input output accounting systems inthe European community e an appraisal of their usefulness in raising aware-ness of environmental problems. Eur. J. Agron. 20, 17e24.

Haferkorn, U., 2000. Größen des Wasserhaushaltes verschiedener Böden unterlandwirtschaftlicher Nutzung im klimatischen Grenzraum des MitteldeutschenTrockengebietes e Ergebnisse der Lysimeterstation Brandis. Diss. Univ. Göt-tingen, 157 pp.

Havlin, J., 2004. Impact of management systems on fertilizer nitrogen use efficiency.In: Mosier, A.R., Syers, J.K., Freney, J.R. (Eds.), Agriculture and the Nitrogen Cycle,vol. 65. International Council of Scientific Unions/Scientific Committee onProblems of the Environment, Scope, pp. 167e178.

Hölting, B., Haertlé, Th., Hohberger, K.-H., Nachtigall, K., Villinger, E., Weinzierl, W.,Wrobel, J.-P., 1995. Konzept zur Ermittlung der Schutzfunktion der Grund-wasserüberdeckung. Geologisches Jahrbuch C 63, 5e24.

Intergovernmental Panel on Climate Change (IPCC) Climate Change, 2007. ThePhysical Science Basis Contribution of Working Group I to the Fourth Assess-ment Report of the IPCC.

Jansons, V., Buismanis, P., Dzalbe, I., Kirsteina, D., 2003. Catchment and drainagefield nitrogen balances and nitrogen loss in three agriculturally influencedLatvian watersheds. Eur. J. Agron. 20, 173e179.

Jaynes, D.B., Colvin, T.S., Karlen, D.L., Cambardella, C.A., Meek, D.W., 2001. Nitrateloss in subsurface drainage as affected by nitrogen fertilizer rate. J. Environ.Qual. 30, 1305e1314.

Juergens-Gschwind, S., 1989. Ground water nitrates in other developed countries(Europe) e relationships to land use patterns. In: Follett, R.F. (Ed.), NitrogenManagement and Ground Water Protection. Elsevier Science, Publishers,Amsterdam. The Netherlands, pp. 75e138.

Kersebaum, K.C., Matzdorf, B., Kiesel, J., Piorr, A., Steidl, J., 2006. Model-basedevaluation of agri-environmental measures in the Federal State of Brandenburg(Germany) concerning N pollution of groundwater and surface water. J. PlantNutr. Soil Sci. 169, 352e359.

Kladivko, E.J., Frankenberger, J.R., Jaynes, D.B., Meek, D.W., Jenkinson, B.J.,Fausey, N.R., 2004. Nitrate leaching to subsurface drains as affected by drainspacing and changes in crop production system. J. Environ. Qual. 33 (5),1803e1813.

Knappe, S., Haferkorn, U., Meissner, R., 2002. Influence of different agriculturalmanagement systems on nitrogen leaching: results of lysimeter studies. J. PlantNutr. Soil Sci. 165, 73e77.

Makowski, D., Tichit, M., Guichard, L., Van Keulen, H., Beaudoin, N., 2009. Measuringthe accuracy of agro-environmental indicators. J. Environ. Manage. 90 (Suppl. 2),S139eS146.

Meisinger, J.J., Delgado, J.A., 2002. Principles for managing nitrogen leaching. J. SoilWater Conserv. 57 (6), 485e498.

Miegel, K., Zachow, B., 2006. Abbau von Stickstoff-Bilanzüberschüssen durchmodell-gestützte Düngeempfehlung. ((Reduction of nitrogen surpluses byusing of a nitrogen balance model to determine fertilizer recommendations)).Rostocker Meeresbiol. Beitr. 15, 49e60.

Müller, U., 2004. Auswertungsmethoden im Bodenschutz e Dokumentation zurMethodenbank des Niedersächsischen Bodeninformationssystems (NIBIS),seventh ed. Schweizerbart, Stuttgart, Germany.

Nieder, R., Schollmayer, G., Richter, J., 1989. Denitrification in the rooting zone ofcropped soils with regard to methodology and climate: a review. Biol. Fertil.Soils 8, 219e226.

Oenema, O., Kros, H., de Vries, W., 2003. Approaches and uncertainties in nutrientbudgets: implications for nutrient management and environmental policies.Eur. J. Agron. 20, 3e16.

Oenema, O., Van Liere, L., Schoumans, O., 2005. Effects of lowering nitrogen andphosphorus surpluses in agriculture on the quality of groundwater and surfacewater in the Netherlands. J. Hydrol. 304, 289e301.

Pamperin, L., 2002. Nitratverlagerung in Abhängigkeit von der Bodennutzung, denStandorteigenschaften und der Grundwasserneubildung eines stau-wasserbeeinflussten Grundmoränenstandortes. Horizonte e HerrenhäuserForschungsbeiträge zur Bodenkunde, p. 7.

Pervanchon, F., Bockstaller, C., Bernard, P.Y., Peigné, J., Amiaud, B., Vertès, F.,Fiorelli, J.L., Plantureux, S., 2005. A novel indicator of environmental risks due tonitrogen management on grasslands. Agric. Ecosyst. Environ. 105, 1e16.

Powlson, D.S., Addisott, T.M., Benjamin, N., Cassman, K.G., De Kok, T.M., VanGrinsven, H., L'hirondel, J.L., Avery, A.A., Van Kessel, C., 2008. When does nitratebecome a risk for humans? J. Environ. Qual. 37, 291e295.

Randall, G.W., Huggins, D.R., Russelle, M.P., Fuchs, D.J., Nelson, W.W., Anderson, J.L.,1997. Nitrate losses through subsurface tile drainage in conservation reserveprogram, alfalfa, and row crop systems. J. Environ. Qual. 26, 1240e1247.

Randall, G.W., Vetsch, J.A., Huffman, J.R., 2003. Nitrate losses in subsurface drainagefrom a corn-soybean rotation as affected by time of nitrogen application anduse of nitrapyrin. J. Environ. Qual. 32, 1764e1772.

U. Buczko et al. / Journal of Environmental Management 91 (2010) 1305e1315 1315

Rankinen, K., Salo, T., Granlund, K., 2007. Simulated nitrogen leaching, nitrogenmass field balances and their correlation on four farms in south-westernFinland during the period 2000e2005. Agr. Food Sci. 16, 387e406.

Schröder, J.J., Neeteson, J.J., 2008. Nutrient management regulations in TheNetherlands. Geoderma 144, 418e425.

Schröder, J.J., Scholefield, D., Cabral, F., Hofman, G., 2004. The effects of nutrientlosses from agriculture on ground and surface water quality: the position ofscience in developing indicators for regulation. Environ. Sci. Policy 7, 15e23.

Schröder, J.J., Aarts, H.F.M., van Middelkoop, J.C., Schils, R.L.M., Velthof, G.L.,Fraters, B., Willems, W.J., 2007. Permissible manure and fertilizer use in dairyfarming systems on sandy soils in The Netherlands to comply with the nitratesdirective target. Eur. J. Agron. 27, 102e114.

Shaffer, M.J., Delgado, J.A., 2002. Essentials of a national nitrate leaching indexassessment tool. J. Soil Water Conserv. 57, 327e335.

Sieling, K., Kage, H., 2006. N balance as an indicator of N leaching in an oilseedrape e winter wheat e winter barley rotation. Agric. Ecosyst. Environ. 115,261e269.

Sieling, K., Günther-Borstel, O., Hanus, H., 1997. Effects of slurry application andmineral nitrogen fertilization on N leaching in different crop combinations.J. Agric. Sci. 128, 79e86.

Sogbedji, J.M., van Es, H.M., Yang, C.L., Geohring, L.D., Magdoff, F.R., 2000. Nitrateleaching and nitrogen budget as affected by maize nitrogen rate and soil type.J. Environ. Qual. 29, 1813e1820.

Strock, J.S., Porter, P.M., Russelle, M.P., 2004. Cover cropping to reduce nitrate lossthrough subsurface drainage in the northern US Corn Belt. J. Environ. Qual. 33,1010e1016.

Ten Berge, H.F.M., 2002. A Review of Potential Indicators for Nitrate Loss fromCropping and Farming Systems in The Netherlands. Plant Research Interna-tional B.V, Wageningen, The Netherlands, 168 pp.

Tiemeyer, B., Lennartz, B., Kahle, P., 2008. Analysing nitrate losses from an artifi-cially drained lowland catchment (north-eastern Germany) with a mixingmodel. Agric. Ecosyst. Environ. 123 (1e3), 125e136.

Townsend, A.R., Howarth, R.W., Bazzaz, F.A., Booth, M.S., Cleveland, C.C., Collinge, S.K., Dobson, A.P., Epstein, P.R., Holland, E.A., Keeney, D.R., Mallin, M.A., Rogers, C.A., Wayne, P., Wolfe, A.H., 2003. Human health effects of a changing globalnitrogen cycle. Front. Ecol. Environ. 1, 240e246.

Van Eerdt, M.M., Fong, P.K.N., 1998. The monitoring of nitrogen surpluses fromagriculture. Environ. Pollut. 102, 227e233.

Vigil, M.F., Eghball, B., Cabrera, M.L., Jakubowski, B.R., Davis, J.G., 2002. Accountingfor seasonal nitrogen mineralization: an overview. J. Soil Water Conserv. 57,464e469.

Vinten, A.J.A., Vivian, B.J., Wright, F., Howard, R.S., 1994. A comparative study ofnitrate leaching from soils of differing textures under similar climatic andcropping conditions. J. Hydrol. 159, 197e213.

Wehrmann, J., Scharpf, N.C., 1979. Der Mineralstickstoffgehalt des Bodens alsMaßstab fur den Stickstoffdüngerbedarf (Nmin � Methode). Plant. Soil. 52,109e126.

Wendland, F., Albert, H., Bach, M., Schmidt, R., 1993. Atlas zum Nitratstrom in derBundesrepublik Deutschland. Springer, Heidelberg, 96 pp.

Yang, J., Greenwood, D.J., Rowell, D.L., Wadsworth, G.A., Burns, I.G., 2000. Statisticalmethods for evaluating a crop nitrogen simulation model, N_ABLE. Agric. Syst.64, 37e53.