84 A.J.J. Van Rompaey et al. Agriculture, Ecosystems and Environment 83 2001 83–94
behaviour and decision rules in response to land use change drivers has only recently been recognised
Veldkamp and Fresco, 1996. A good example of this principle is the Common
Agriculture Policy of the European Union EU-CAP. Since the reform of the EU-CAP in 1992, the support
system for European farmers has thoroughly changed. The new CAP includes regulations on set-aside in
order to control the level of agricultural production and to secure the income of farmers. Therefore, these
EU-CAP rules can be seen as a land use change driver. The farmers are financially supported by EU
relative to the area of set-aside land on their farm. Set-aside land is land that is either fallowed or used for
non-food crops, e.g. energy crops. In the new support arrangements the farmer has to set aside a minimum
percentage of his land, based on his area of arable land in 1991. EU changes the minimum set-aside per-
centage from year to year depending on the market situation. Since 1992 the minimum set-aside percent-
age has varied between 5 and 15 see Table 1. The take-out period, during which production for agri-
cultural purposes is forbidden, runs from 15 January until 31 August for a given year. Protection by vege-
tation of the fallow fields with selected fallow-species is obligatory: however, the fallow species are specific
for each EU-member state Sibbesen, 1997.
As the potential for surface runoff and soil ero- sion is very much affected by land cover, these CAP-
regulations potentially have a very strong impact on the average soil erosion rates in agricultural regions.
A permanent vegetation cover protects the soil from direct raindrop impact, crusting and sealing which
reduces the amount of surface runoff till almost zero.
Table 1 Minimum set-aside percentages since the reform of the Common
Agriculture Policy of the European Union Year
Minimum set-aside percentage
1992 15
1993 15
1994 15
1995 12
1996 10
1997 5
1998 5
1999 10
De Ploey 1989 estimated that soil erosion rates on unprotected fields may be 100–1000 times higher than
on fields with permanent vegetation cover. Therefore, it may be expected that a fallow percentage of, for
example, 10 will reduce the total amount of soil erosion by 10 since soil erosion on the protected
fields may be reduced to 0 Mg ha
− 1
per year. This is indeed true if farmers select at random fields for
set-aside. However, if farmers prefer to take steep and erodible fields out of production the reduction
of the total erosion rate will be much higher since the average soil erosion risk of the remaining fields
will be lower. Thus, insight in the decision rules that farmers use in response to the CAP-regulations is
necessary to estimate the magnitude of the effects of the new policy on soil erosion in agricultural
regions.
The objectives of this study are: 1 to determine what field properties do farmers take into account
when selecting parcels for set-aside; 2 to model and simulate the spatial pattern of set-aside fields under
different scenarios of the EU-CAP, and 3 to quan- tify the impact of different set-aside scenarios on the
regional soil erosion risk.
2. Materials and methods
2.1. Landscape characteristics of the study area The study area 850 km
2
consists of 10 municipali- ties situated in the southern part of the province Flem-
ish Brabant in central Belgium Fig. 1. The study area is a part of the Belgian loess belt, which is composed
of a strongly incised plateau reaching up to 100 m above sea level. Incision of the valleys can reach 60 m
Fig. 2.
The region is characterised by a semi-continuous loess cover. The topsoils generally have a very high
silt content 750 g kg
− 1
, a clay content between 100 and 200 g kg
− 1
and a sand content below 150 g kg
− 1
. According to the FAO FAO, 1998 the majority of
the soils can be classified as Luvisols with different drainage types, depending on the landscape position.
The depth of the loess cover is highly variable but is of the order of some decimetres to some meters
Goossens, 1993. Locally outcrops of tertiary sandy deposits occur with very weak soil development
A.J.J. Van Rompaey et al. Agriculture, Ecosystems and Environment 83 2001 83–94 85
Fig. 1. Location of the study area in grey.
Arenosols. On the colluvium in at the footslopes Regosols can be found.
A reclassified LANDSAT satellite image allowed distinguishing different land use classes Fig. 3;
Gulinck et al., 1996. The 50 of the land in the study area consists of arable land, 17 is urban zone
or infrastructure while 18 is under pasture and 13 is covered with forest.
The gradual mechanisation of the agriculture since the second world war changed the landscape structure
Fig. 2. Topography of the study area. The plateau reaches up to 100 m above sea level.
thoroughly Desmet et al., 1999. Larger fields and the disappearance of field boundaries increased the
soil erosion susceptibility Van Oost et al., 2000. The average field size in the study area is now ca. 2.5 ha.
The arable land is especially vulnerable to erosion in spring and early summer when typical summer
crops like sugar beet Beta vulgaris L., maize Zea mays
L., potatoes Solanum tubersorum L. and chicory Chicory intybus L. have a very low vegeta-
tion cover. Moreover, this period is characterised by
86 A.J.J. Van Rompaey et al. Agriculture, Ecosystems and Environment 83 2001 83–94
Fig. 3. Land use classes in the study area Classified LANDSAT; Gulinck et al., 1996.
the occurrence of heavy convective showers Vandaele and Poesen, 1995. The erosion and surface runoff on
arable land on the plateau edges causes yearly prob- lems of flooding and muddy floods in the villages in
the valleys Verstraeten and Poesen, 1999. Further- more, the aquatic environment suffers from pollution
because of transfers of nutrients and pesticides to the river system by leaching and surface runoff.
An erosion risk map Fig. 4 of the study area was compiled by Van Rompaey et al. 2000 using the
universal soil loss equation USLE Wischmeier and Smith, 1978. The USLE is an empirical model that
predicts the annual average, long-term water erosion rate as the product of a rainfall erosivity factor R,
a soil erodibility factor K, a slope length factor L, a slope steepness factor S, a crop factor C and
Fig. 4. Predicted average long-term soil erosion rates in the study area Mg ha
− 1
per year.
a management factor P. Average erosion rates of more than 10 Mg ha
− 1
per year are considered as very serious and a threat for the sustainability of the arable
land. The erosion map shows that the mean annual erosion rates exceed 10 Mg ha
− 1
in one-third of the study area. On local hotspots soil erosion rates can
exceed 50 Mg ha
− 1
per year. The average soil erosion rate on arable land in the study area is ca. 6.0 Mg ha
− 1
per year. 2.2. Statistical analysis of land use patterns
2.2.1. Questionnaire In order to find out what criteria farmers use when
taking fields out of production a questionnaire was designed. During the summer of 1997, 31 farmers in
A.J.J. Van Rompaey et al. Agriculture, Ecosystems and Environment 83 2001 83–94 87
Table 2 Parcels of the 31 questioned farmers in summer 1997
Category
a
Frequency Relative frequency
I 57
5.9 II
195 20.0
III 722
74.1 Total
974 100
a
Category I: fallow at the moment of the inquiry; Category II: no fallow at the moment of the inquiry but a possibly next year,
and Category III: no fallow and non-considerable in the future.
the study area were asked to locate their fields on aerial photographs. The farmers were asked to classify their
fields in three categories: fallow at the moment of the inquiry Category 1; no fallow at the moment of the
inquiry, but a possibly in the near future Category 2, or no fallow and none considered in the future
Category 3. In total a database of 974 fields was acquired Table 2.
The mean criteria farmers mentioned for taking fields out of production are listed in Table 3. It is
remarkable that soil erosion susceptibility is never mentioned and slope gradient only once. Drawing di-
rect conclusions from Table 3 is dangerous since it is rather difficult to analyse the driving factors of one’s
own decisions and behaviour. Therefore, the farmer’s responses may be rather vague e.g. the criterion poor
quality, incomplete or even biased. Moreover, for many fields farmers could not mention an exact rea-
son as to why the field was taken out of production.
Table 3 Criteria for set-aside fields mentioned by farmers results of the
questionnaire Criteria
Times mentioned Poor quality general
13 Surface fits with obliged percentage
11 Geometric shape of the parcel
7 Bad drainage
7 Near to forest
5 Crop of last year
4 Too small
4 Too far from farm
3 Difficult access with machinery
3 Possible expropriation
2 Clay texture
2 Too steep
2 Stoniness
Soil erosion
It appears that the decision mechanisms of farmers with respect to set-aside are often based on personal
experience and knowledge, which they find very hard to translate in objective and isolated criteria.
2.2.2. Statistical analysis of slope A second possibility to identify possible factors re-
lated to set aside is to extract some objective field characteristics and to analyse the possible differences
between the two field categories fallow and no fal- low. Therefore, statistical tests were carried out to test
whether or not certain field characteristics are signifi- cantly different for fallow and non-fallow fields.
The locations of all the fields of the questioned farmers were digitised on aerial photographs resulting
in a digital field-file in vector format. Digital elevation data for the study area were available from the Belgian
National Geographic Institute NGI. The dataset con- sists of regular grid data sampled every 1
′′
in latitude and every 2
′′
in longitude from scanned topographic maps at a scale of 1:50 000. Interpolation and filtering
with a low-pass filter resulted in a raster digital ele- vation model DEM representing the topography as
a continuously varying surface. The accuracy of this DEM was evaluated with more accurate DEMs de-
rived from 1:10 000 topographic maps for eight test areas. The overall root mean square error RMSE of
the altimetric differences between the two DEMs is 3.1 m. The correspondence is better in the flat valleys
and worse on the incised plateau. Slopes were calcu- lated using the algorithm of Zevenbergen and Thorne
1987. A regression between the calculated slopes and the slopes from the test areas provided a correction for
the systematic underestimation of the slopes caused by the low resolution of the input data Van Rompaey
et al., 1999. Via an overlay of the field-file and slope map, it was possible to extract the mean slope for each
field.
In order to find out to what degree the slope gradient of a field determines the probability of being taken
out of production, the dataset of the questionnaire was split in two parts: the fallow and possibly fallow fields
and the non-fallow fields. If slope does not play a role in the selection process of the fallow fields, then there
should be no difference between the mean slope of the fallow and possibly fallow fields, and the mean
slope of the non-fallow fields. This hypothesis was tested via a Student t-test Wonnacott and Wonnacott,
88 A.J.J. Van Rompaey et al. Agriculture, Ecosystems and Environment 83 2001 83–94
1972. All statistical tests were carried out using SAS procedures SAS Institute, 1985. The procedure first
tests whether the variances of the two populations are equal or not by means of an F-test. Depending on
the case equal variances or non-equal variances the appropriate T-value is calculated which can be used
to calculate the probability of the averages of the two groups in this case the slope gradients of set-aside
versus arable fields being equal.
2.2.3. Statistical analysis of soil texture and soil drainage
Although the soils developed on the loess deposits mainly belong to the Luvisol group of the FAO soil
classification FAO, 1998, there were some variations in soil texture and soil drainage. In the northern part
of the study area, the eolian deposits have a higher sand content. In the southern part of the study area, the
variation in soil texture is mainly caused by outcrops of tertiary deposits that have a higher silt or clay con-
tent. A soil texture map was compiled on the basis of soil cores of 1 m depth with a density of 1–2 drillings
ha
− 1
. Map purity estimates indicate that ca. 35 of the observations were classified in a textural class just
above or below the correct textural class Van Meir- venne, 1998. Soil textures were classified using the
Belgian texture classification Ameryckx et al., 1985 that does not correspond to any international classi-
fication. Seven soil texture classes are distinguished: loam A, sandy loam P, light sandy loam L, loamy
sand S, sand Z, clay E and heavy clay U. Soil drainage class was also mapped. Three classes were
distinguished: well-drained soils, soils with a moder- ate drainage and soils with a poor drainage. Via an
overlay of the field-file and the soil texture and soil drainage map, it was possible to extract for each field
the soil texture class and the soil drainage class. The original texture classes were reduced to four main tex-
ture classes: loam A+P, sandy loam L+S, sand Z and clay E+U. This was done because some of the
original texture classes contained very few members. This could make further analysis unreliable.
Since soil properties texture and drainage are qual- itative parameters, a Student t-test is not possible.
Therefore, a χ
2
-test was used for testing whether soil properties of fallow fields are not significantly differ-
ent from the non-fallow fields. A χ
2
-test compares the expected frequency distribution over classes with the
observed distribution and calculates the probability of the two distributions being equal. The χ
2
-value is cal- culated as follows Wonnacott and Wonnacott, 1972:
χ
2
=
n
X
i= 1
O
i
− E
i 2
E
i
1 where O
i
is the observed frequency in class i, E
i
the expected frequency in class i, and n is the number of
classes. The expected distribution of the fallow fields over
the different texture classes was calculated as the to- tal distribution of all the fields in the database times
the relative frequency of the fallow fields in the total amount of fields 25.9, see Table 2. The same pro-
cedure was carried out for soil drainage classes. The expected distribution over three main drainage classes
well drained, moderate drainage and poor drainage was compared with the observed distribution.
2.3. Simulation of set-aside patterns under different scenarios
The information from the limited questionnaire was extrapolated over the whole study-area and used
for simulation of the effect of different EU-CAP sce- narios. First the transition probability for each field
in the study area is estimated based on its specific slope and soil characteristics. A transition probability
is the probability that a field will be converted from arable land into fallow land. For a single factor e.g.
the texture class, slope class or the drainage class the transition probability can be estimated as the relative
frequency of fallow fields in that category. For exam- ple, the questionnaire results show that in the slope
class 0–5, 20.5 of the fields was fallow or possibly fallow. Therefore, the transition probability for this
class can be estimated as 20.5. In the slope class 5–10 the relative frequency of fallow fields and
therefore the corresponding transition probability is higher 32.5. Such a single factor transition proba-
bility can be calculated for each class of the different field characteristics slope, texture and drainage.
However, more than one field characteristic is known. The overall transition probability can then be
calculated using the theorem of Bayes P
event|A
i
∩ B
i
= P
event|A
i
∗ P event|B
i
P event
2
A.J.J. Van Rompaey et al. Agriculture, Ecosystems and Environment 83 2001 83–94 89
where A
i
is the ith class of variable A e.g. the class ‘10–15’ of slope and B
i
is the ith class of variable A
e.g. the class ‘sand’ of texture. This means that the probability of ‘an event’ in our case the set-aside
of a field given conditions A
i
and B
i
is equal to the probability of the event under condition A
i
times the probability of the event under condition B
i
divided by the overall probability of the event. For example, the
transition probability of a field for which slope and texture are known can be calculated as
P fallow|slope class : ‘ 15’ ∩ texture
class : ‘sand’ =
P fallow|slope class : ‘ 15’
∗ P
fallow|texture class : ‘sand’ P
fallow 3
For n factors in this case field characteristics the theorem can be extended.
P fallow|A
i
∩ B
i
∩ · · · N
i
= P
fallow|A
i
∗ P fallow|B
i
∗ · · · P fallow|N
i
P fallow
n− 1
4 In this case the average probability for fallow
Pfallow is known since this is the minimum set-aside percentage of the EU-CAP see Table 1.
Using Eq. 4 the spatial pattern of transition prob- ability was calculated for a range of set-aside per-
centages. An example of such a transition probability map for a minimum set-aside percentage of 10 is
shown in Fig. 5. It is important to point out that the transition probabilities calculated using the procedure
described above are independent of the distribution of observations in the sample used to derive the tran-
sition probabilities. As the estimated single factor transition probabilities are the outcome of discrete
events fallow or non-fallow their accuracy can be assessed using the theory of binomial distributions.
The standard deviation of the sample mean, being the estimated single factor transition probability calcu-
lated from the observed relative frequencies, can then be estimated as follows for samples having more than
20 observations Conover, 1971
σ = r
p × 1 − p
n 5
where n is the number of observations, and p is the relative frequency of transitions observed. Thus, the
reliability of the transition probability is dependent only on the number of observations in each class as
well as the number of transitions observed. Based on these final transition probabilities, stochastic simula-
tions of possible set-aside patterns in the study area. This was done for different EU-CAP scenarios. The
study area was divided in square blocks of 5 km×5 km since these spatial units correspond more or less with
spatial distribution of the fields of one farmer and is therefore the level at which set-aside simulations have
to be carried out. For each of these blocks set-aside volumes of 5, 10, 20 and 25 were simulated.
A simulation was carried out as follows. A field of arable land in a 5 km×5 km block was selected
at random. Next a random number between 0 and 1 was chosen. This number was then compared with
the transition probability of the selected field. If the random number was less than the transition probability
then the field was accepted for set-aside; if not the field was left in production. This procedure was re-iterated
until the desired simulated percentage of set-aside land was reached. This way of field selection implies that
fields with a higher transition probability have a higher probability of being selected for set-aside.
As mentioned in the introduction fallow fields have to be protected with selected fallow crops and can thus
be considered as land units with no erosion. Therefore, new soil erosion risk maps were compiled for the dif-
ferent EU-CAP set-aside scenarios with soil erosion rates of 0 Mg ha
− 1
per year on the spots correspond- ing with a simulated fallow field.
3. Results and discussion
