196 A. Huber et al. Agriculture, Ecosystems and Environment 80 2000 191–204
would only be dependent upon application rates. How- ever, the probability of occurrence of watercourses at
field edge is the decisive factor with regard to the re- gional importance of spray drift as a pathway for pes-
ticide input into surface water. In general frequency of surface waters per unit area can be described by
drainage density in km km
− 2
Horton, 1932. In this study drainage density was first computed on the ba-
sis of the digital surface water net 1:200.000; HAD, 1997 using GIS techniques. Due to cartographic gen-
eralization the digital surface water net is represented incompletely, therefore real values were extrapolated
by regression analysis taking measured drainage den- sities in 147 different agroecological zones Huber,
1998.
2.5. Other non-point sources of surface water contamination with pesticides
Some non-point sources were not modeled because they are not considered to be relevant on a national
scale such as atmospheric deposition or wind erosion. For the following two reasons sediment-bound pesti-
cide loss was also neglected: 1. The spatial databases required for erosion calcula-
tion are not available for Germany, so modeling of pesticide input is limited to the dissolved phase.
2. With Germany’s prevailing weather, soil and land use conditions, surface runoff occurs frequently
causing no or hardly any soil erosion, e.g. 5 to 15 times per year according to model calculations.
Heavy rainstorms including higher erosion only oc- cur once or twice per year, mainly effecting row
crops. In this respect the situation differs signifi- cantly from the ones in other countries, especially
in North America. Thus pesticide transport in so- lute phase is assumed to be the dominant runoff
component.
Burgoa and Wauchope 1995 also stated that sed- iment constitutes such a small fraction of runoff fol-
lowing even highly erosive storms, that it is likely that the bulk of active ingredient will be lost in the water
phase.
Some authors reported a notable pesticide flux from groundwater to surface water Squillace and Thurman,
1992. However, concentrations in groundwater are generally very low Leistra and Boesten, 1989. In
this study it was therefore not assumed that pesticide concentrations in surface water, which are usually
higher than those in groundwater, are caused to a considerable extent by groundwater discharge.
3. Results and discussion
Model results for pesticide loss via surface runoff, tile drains and spray drift are shown in Table 2 and
Fig. 3a–c. To account for parameter uncertainty inher- ent to all regional scale models, computed loads were
given with ranges of variation both in Table 2 and in generated maps Fig. 4a–d. Magnitude of variation
Fig. 3. Mean solid line and variation bars of modeled pesticide loss to surface waters for transport with a surface runoff, b
through tile drains and c spray drift no. of active ingredients refer to Table 2.
A. Huber et al. Agriculture, Ecosystems and Environment 80 2000 191–204 197
Table 2 Active ingredients considered in the study, their chemical properties and their predicted annual losses into surface waters via different
pathways in Germany cumulated over all treatments, reference year 1994
e
No. Active ingredient
common name K
oc a
l kg
− 1
Dt
50
day
a
Treated crops
Runoff kg Tile drains kg
Spray drift kg Total kg
Mean Range
b
Mean Range
c
Mean Range
d
Mean Range 1. Metamitron
H 156
21 Sugar beet
2423 629 –4571
11 2–15
2433 441–5138 2. Isoproturon
H 100
12 Grain
1278 361–2540
947 1–6680
20 4–33
2245 5–9565 3. Ethofumesate
H 131
51 Sugar beet
1030 475–2193
78 0–1658
1 0.2–2
1109 178–3853 4. Terbuthylazine
H 301
88 Corn, potatoes
881 585–1662
14 0–4255
3 0.5–6
898 171–6123
5. Dichlorprop-P H
80 14
Grain 631
223–1129 0–36
8 1–13
639 224–1482
6. Metolachlor H
200 39
Corn, legumes 10
284–1029 0–47
1 0.2–2
511 96–1123
7. Chloridazon H
205 43
Sugar beet 408
238–919 0–109
0.3 0–0.6
408 76–1029
8. Dichlofluanid F
10 18
Grapevine 385
215–532 25
2–28 410
217–560 Orchards
42 21–67
105 26–405
388 48–756
535 95–1228
9. Bentazone H
28 18
Grain, potatoes 298
94–382 103
0.8–894 0.7
0.3–1 402
95–1276 10. Chlortoluron
H 235
27 Grain
284 155–666
0–339 2
0.5–5 286
155–1005 11. Simazine
H 201
57 Grain, corn
130 57–318
33 0–504
163 57–822
12. Mecoprop-P H
139 9
Grain 118
37–283 7
1–13 125
37–283 13. Metabenzthia-
zuron H
200 20
Grain, legumes 101
32–198 0–25
0.6 0.3–1
102 32–223
14. Propineb F
3 8
Grapevine 96
31–187 11
107 31–187
Orchards 2
33–158 119
19–593 1221
155–2363 1342 174–2956 15. MCPA
H 62
15 Grain, potatoes
87 0–263
0.4 0.1–0.8
87 33–421
16. Phenmedipham H
870 34
Sugar beet 85
48–132 0.2
0.04–0.4 85
48–132 17. Metazachlor
H 80
6 Rape
69 9–239
1 0.2–2
70 9–239
18. Prometryn H
400 60
Grain, legumes 40
17–86 40
17–86 19. 2,4-D
H 38
10 Grain, corn
25 5–68
25 5–68
Grapevine 2
2 20. Triallate
H 2400
66 Sugar beet,
corn, legumes 24
7–44 0.08
24 7–44
21. Mancozeb F
886 15
Orchards 3.5
4–53 754
90–1462 758
90–1462 Grapevine
21 21
2–24 42
6–77 Potatoes
0.02 0.5
1 22. Methamidophos
I 5
9 All field crops
18 92
11–293 0.3
110 11–293
23. Dithianon F
2490 24
Orchards 1
536 66–1040
537 66–1040
Grapevine 9
14 23
24. Metiram F 50000
3 Grapevine
0.01 40
3–45 40
3–45 Orchards
19 22–376
19 22–376
Potatoes 0.01
0.5 1
25. Oxydemeton- methyl
I 10
2 Orchards
28 4–54
28 4–54
All field crops 0.01
0.2 1
26. Maneb F
1000 42
Potatoes 1
9 2–14
10 2–16
27. Pendimethalin H 16000
126 Grain, corn,
legumes 12
2 0.5–5
14 3–32
28. Bromoxynil H
184 4
Grain, corn 11
0.2 0.03–0.3
11 2–25
29. Parathion- methyl
I 451
22 Orchards
0.01 10
1–25 10
2–33 Grapevine
4 1
5 Corn, rape,
legumes 30. Tebuconazole
F 1008
60 Grapevine
5 3
1–11 8
2–26 Grain, rape
2 4
6 31. Triadimenol
F 1000
110 Orchards
0.2 8
0.1–18 8
0–19 Grapevine
0.3 0.4
1 Grain
1 1
32. Diuron H
900 68
Orchards 7
7 4–14
33. Parathion I
1316 18
Orchards 0.01
7 0.8–14
7 1–14
Grapevine 0.1
0.4 1
All field crops 0.04
1 34. Anilazine
F 349
22 Grain
0.2 6
3–11 6
3–11 35. Fenpropimorph
F 4178
31 Grain
0.1 4
0.8–8 4
1–8
198 A. Huber et al. Agriculture, Ecosystems and Environment 80 2000 191–204
Table 2 Continued
No. Active ingredient
common name K
oc a
l kg
− 1
Dt
50
day
a
Treated crops
Runoff kg Tile drains kg
Spray drift kg Total kg
Mean Range
b
Mean Range
c
Mean Range
d
Mean Range 36. Prochloraz
F 1000
60 Grain, rape
1 2
0.4–4 3
1–6 37. Bifenox
H 6450 7
Grain 0.01
1 0.2–2
1 0.2–2
38. Glyphosate H 12800
27 Grapevine
1 1
0.01–4 2
0.3–8 All field crops
0.8 0.6
1 Orchards
0.02 0.3
1 39. Trifluralin
H 8500 155
Rape 0.9
1 0–2
40. Dimethoate I
20 7
All field crops 0.01
0.04 0–1 1
0–1 41. Endosulfan
I 12400 50
Potatoes, rape, legumes
0.01 1
42. Linuron H
400 60
Grain 0.01
1 0–0.2
Sum 9060 1500–19,300 1490 50–16,000 3350
400–6300 13,900 2000–42,000
a
Values extracted from the chemical properties data base at the German Federal Environmental Agency unpublished.
b
Confidence ranges in parenthesis of pesticide losses resulting with different values for sorption coefficient K
OC
, +100−50, disappearance time Dt
50
, +100−50 and curve number CN-value+25−25.
c
Confidence ranges in parenthesis with different values for K
OC
, Dt
50
and tile drain density +100−50.
d
Confidence ranges in parenthesis with various distances from surface water field crops: 2, 5, 10 m; orchards, vineyards: 5, 10, 20 m.
e
H=herbicide, F=fungicide, I=insecticide.
for pesticide loss is different for every single pathway and was based on the sensitivity analysis performed
for each model Huber, 1998; Huber et al., 1998.
3.1. Surface runoff According to model results with a total amount
of ca. 9000 kg a.i. surface runoff is the dominant non-point source pathway for pesticide input into sur-
face waters. It becomes clear that chemical transport with surface runoff is a more frequent phenomenon
than losses via tile drains or spray drift, particularly in regions with a high spraying intensity like in the loess
areas of Central Germany, along the Lower Rhine and the Danube river Fig. 4a. Sugar beets Beta
vulgaris
L. form an important part of crop rotation in those areas. Like other row crops sugar beets require
thorough herbicide treatments in late spring to ensure crop development; in the same time the probability of
significant rainfalls grows with increasing tempera- ture. That means a considerable amount of herbicides
are available in a period of frequent rainstorms and low crop cover. The same applies for spring barley
Hordeum vulgare L. which is sown in April and does not develop sufficient crop canopy before early
summer Fig. 1.
The highest annual dissipation rate 5 g ha
− 1
was modeled for vineyards in the Mosel region and along
the upper and middle Rhine. In contrast vineyards where also fungicides are subject to runoff transport,
pesticide runoff in field crops consists nearly exclu- sively of herbicides. Cumulated losses after all calcu-
lated treatments vary between 0.42 of application rate for ethofumesate to less than 0.01 for stronger
bound or rapidly degradable active ingredients. In the case of isoproturon, the most frequently used her-
bicide in German agriculture, calculated runoff loss was about 0.06 of the application rate.
It is important to note that the approach which was developed for calculating runoff loss differs signifi-
cantly from erosion modeling as neither relief infor- mation nor precipitation volume are crucial model
parameters. Runoff volume is only dependent upon infiltration capacity of soil, which is a function of
porosity, water content, susceptibility to crusting or crop cover. In this way it becomes clear that higher
runoff losses were calculated for steep and therefore shallow soils in vineyards, while deeper soils in flat
areas do not show the same runoff susceptibility un- less there exists a nonpermeable layer in the upper
part of the profile. Precipitation volume is a crucial parameter in the model only with regard to its prob-
ability of occurrence. The more frequent significant rainstorms occur the shorter the time is for pesticide
degradation. Therefore small rainfall events, which are enough to produce runoff, outweigh extreme rain-
storms in relative importance for total pesticide losses because they are by far more frequent.
A. Huber et al. Agriculture, Ecosystems and Environment 80 2000 191–204 199
Fig. 4. Spatial distribution of pesticide loss to surface waters on different pathways in Germany.
3.2. Leaching and transport through tile drains For most parts of the country only neglectable
amounts of pesticides are leached to 0.8 m accord- ing to model results. However, when discussing
predicted pesticide leaching it has to be emphasized that PELMO shows a high sensitivity to chemi-
cal properties and soil organic carbon content. The
200 A. Huber et al. Agriculture, Ecosystems and Environment 80 2000 191–204
estimated mean loss of ca. 1500 kg a.i. into sur- face waters via tile drainage for the entire area of
Germany exhibits therefore a high variation for some active ingredients Fig. 3b, especially when sensitive
parameter values like half-life Dt
50
or sorption coef- ficient K
OC
are changed. In some cases no loss was predicted using standard parameter values as given
in Table 2; however, a certain loss resulted when considering parameter variation Fig. 3b.
Predicted pesticide leaching rates are shown in Fig. 4d, discharge from tile drains in Fig. 4c. PELMO
calculations gave high leaching rates in Northern Lowland areas where sandy soils prevail and Autumn
applications to winter grain, particularly treatments with isoproturon, are more common than in other
parts of the country. But only in a few regions a con- siderable leaching potential intersects with a high tile
drain density. Thus, in most regions the surface water pollution is much lower than the pesticide leaching
potential.
A considerable limitation of accuracy of predicted leaching loss must be attributed to spatial variability
of organic carbon content of arable land which could not be considered adequately on the regional scale. To
reduce the number of factor combinations to a man- ageable amount, PELMO-runs were conducted with
five different soil scenarios ranging from a coarse sand to a silt loam, and mean organic carbon contents
in the upper soil layer from 7 to 24 g kg
− 1
. Soil sce- narios were selected according to representativity for
cultivated soils, nevertheless it may be possible that the model predicted erroneous pesticide leaching rates
due to a too coarse classification of soils in one of the defined soil scenarios. However, even without regroup-
ing soil types, the modeler has to be aware of the fact that soil classified under the same taxonomic hierar-
chy offers a great variation of soil properties Loague et al., 1996.
Another limitation of the model is that PELMO does not account for preferential flow. This may
constitute a considerable drawback when the model is applied to heavier soils Klein, 1994; Thorsen
et al., 1998. It is evident, that highly permeable and thus leaching susceptible soils are drained
to a lower extent than heavier soils. However, it is possible that frequently drained soils in loess
areas, which do not exhibit a high leaching poten- tial after PELMO runs Fig. 4d would be classi-
fied differently when considering solute transport in macropores.
3.3. Spray drift Frede et al. 1998 stated in Europe that no field
studies exist determining exclusively spraydrift in- put into surface water bodies under natural condi-
tions. All monitoring studies on a catchment scale analysed a mixture of pesticide inputs from at least
two or more diffuse sources, often superimposed by accidental inputs and spills. Experimental data on the
fraction of total pesticide pollution of surface wa- ter originating from spray drift is restricted to worst
case assessments undertaken for registration purposes Ganzelmeier et al., 1995. Applying the median val-
ues of basic spray drift losses Table 1, real crop acreage, application rates and drainage density mod-
eling results demonstrate that for applications in field crops, spray drift contribution to total surface water
pollution is negligible at a total amount of 90 kg a.i. for Germany as a whole. Thus spray drift may con-
tribute locally or temporarily to the contamination of aquatic systems but its contribution to the total annual
pesticide river load is small on river basin or national scale.
This conclusion does not apply to orchards where air blast application leads to a substantially higher
fraction of pesticide transported to the field edge than following boom sprayer application to field crops.
Furthermore, the main apple tree growing area in Ger- many is located on marshland west of Hamburg where
drainage densities reach up to 100–200 km km
− 2
. Natural conditions and agricultural practices interact
in a way that spray drift input amounts to ca. 3000 kg a.i. according to model estimate. This input pathway
outweighs by far other non-point sources of pesticide pollution in this area. Moreover, this is the only path-
way where fungicides contribute to a higher extent to water pollution than herbicides in Germany Fig. 3c,
Table 2.
3.4. Comparison of modeled loss to measured pesticide loads in different catchments
3.4.1. Problem of model validation In general it is difficult to compare modeled pesti-
cide losses on a single pathway with experimental data
A. Huber et al. Agriculture, Ecosystems and Environment 80 2000 191–204 201
Table 3 Catchment studies used for the comparison of measured pesticide river loads with predicted pesticide losses for geographic location of
catchments see Fig. 5 No. Catchment no. of
analysed pesticides Size
km
2
Sampling period
Sampling frequency
Isoproturon Source
a
of experi- mental data
Measured load kg
Modeled input kg
1. Stever 5 585
1994 entire year 7 days
60.8 40.85
Gelsenwasser AG 1997 2. Nidda 5
1619 4181994–5251994
3 days 14.0
0.70 Seel et al. 1996
3. Schwarzach 6 170
4171996–421997 30 days
3.286 0.48
Wasserwirtsch.amt Würzburg 1997 4. Selz 14
363 4111995–9201995
10 days 1.413
0.308 Pflanzenschutzd. Rheinl.-Pfalz 1997
5. Frischofsbach 10 32
12281993–131995 7 days
1.173 2.935
Umweltamt Münster 1997 6. Lumda 11
129 4241993–691993
24 h 0.55
0.08 Bach and Frede 1996
7. Thierbach 6 152
4151996–1171996 30 days
0.333 0.332
Wasserwirtsch.amt Würzburg 1997 8. Wieseck 11
80 4241993–691993
24 h 0.068
0.119 Bach and Frede 1996
9. Hess. Bach 5 6
4161994–5181994 24 h
0.024
b
0.118 Fischer 1996
10. Gerrenbach 4 6
1994 entire year 24 h
0.01 0.037
Skark and Zullei-Seibert 1997 11. Lettebach 4
5 1994 entire year
24 h 0.005
0.023 Skark and Zullei-Seibert 1997
12. Blument. Aue 9 34
4201995–7161995 48 h
City of Bremen 1997 13. Körsch 6
127 1241996–11271996 30 days
No data No data
Honnen et al. 1997
a
Sources cf. Huber 1998.
b
Only pesticide load which originated from surface runoff, tile drains and spray drift.
because plots and even small catchments studied in a short period may not be representative for an entire
region. In this way only long term catchment studies should be used to prove reliability of predicted losses.
Additionally, experimental data should cover the main application periods and the most frequently applied
active ingredients. At present, only data sets from 13 German catchments Table 3, Fig. 5 fulfill these
requirements.
A major limitation of regional scale modeling is that results can never be validated in a strict sense
since measured loads also include a certain amount of pesticide which did not enter the surface water system
via modeled pathways. Additional inputs could occur via point sources from accidents or spills from sprayer
handling on farmyards. Spills were found to be an important source of surface water pollution Frank
et al., 1982. In some parts of Germany where a large number of farms is connected to public sewage plants,
farmyard runoff outweighs even non-point source discharge Bach, 1996; Fischer, 1996; Frede et al.,
1998. Finally, the model developed in this study estimated pesticide loss to surface waters while ex-
perimental data is load at catchment outlet. Between point of entry and sampling station transformation
and sorption processes may occur, hence it is possible
Fig. 5. Location of catchments used for comparison of modeled pesticide losses and measured loads in surface water.
202 A. Huber et al. Agriculture, Ecosystems and Environment 80 2000 191–204
Fig. 6. Comparison between modeled non-point source inputs and measured loads of isoproturon for catchment characteristics see
Table 3.
that measured loads do not truly reflect total pesticide input.
3.4.2. Modeled loss versus measured load Comparison of loads of isoproturon is probably
most reliable as it was measured in all catchments
Fig. 7. Comparison between modeled non-point source inputs and measured loads of five frequently analysed active ingredients in 13 German catchments Table 3.
except one, for that reason respective loads are shown separately in Fig. 6. Values for other herbicides which
were considered in at least five catchment studies are shown in Fig. 7. Considering the limitations of
the performed comparison Figs. 6 and 7 nevertheless permit several conclusions:
1. In most cases model results seem to match the
magnitude of pesticide pollution in surface waters; however, it is not possible to validate prediction ac-
curacy for a single pathway because no study, ex- cept Fischer 1996, listed loads for each pathway
separately.
2. Fig. 7 shows that the model underestimated loads in those catchments where measured pesticide
pollution exceeded about 100 g a
− 1
. A possible explanation is that the existence of point sources
is more probable in larger catchments, while in small catchments point sources are often excluded
by selection of monitoring sites.
3. Especially in small catchments it is likely that errors due to generalization of spatial input data
distort model results. For example, spatial res- olution of soil data used in this study is prob-
ably too coarse to estimate pesticide runoff or leaching in catchments with an area of less than
50–100 km
2
.
A. Huber et al. Agriculture, Ecosystems and Environment 80 2000 191–204 203
4. Conclusions