Content uploaded by Meynard Jean-Marc
Author content
All content in this area was uploaded by Meynard Jean-Marc on Dec 04, 2021
Content may be subject to copyright.
Model-based explorations to support development of sustainable
farming systems: case studies from France and the Netherlands
W.A.H. Rossing
a,
*, J.M. Meynard
b
, M.K. van Ittersum
a
a
Department of Theoretical Production Ecology, Wageningen Agricultural University, P.O. Box 430, 6700 AK Wageningen, The Netherlands
b
Unite
´d’Agronomie INRA-INA PG, F-78850 Thiverval-Grignon, France
Accepted 13 June 1997
Abstract
Sustainable land use requires development of agricultural production systems that, in addition to economic objectives,
contribute to objectives in areas such as environment, health and well-being, rural scenery and nature. Since these objectives
are at least partially conflicting, development of sustainable farming systems is characterized by negotiation about acceptable
compromises among objectives. Four phases can be distinguished in the course of farming systems development: diagnosis,
design, testing and improvement, and dissemination. During the last decade an approach coined ‘prototyping’ has emerged as
a promising method for empirical farming systems development in Western Europe. Limitations of the approach include: (1)
the limited number of systems that can be evaluated, resulting in a lack of perspective on conflicts among objectives, and (2)
the expertise-based nature of rules used during systems design which unduly narrows the range of available options and
obscures understanding of systems behaviour. In the paper, explorative studies based on transparent models of agronomy and
management are put forward to supplement empirical prototyping and to remedy its shortcomings. To illustrate the potential
of model-based explorations, two case studies are presented. The first case study deals with diagnosis and design of wheat-
based rotations in the Paris Basin of France, aimed at alleviating tactical problems of poor resource-use efficiency within the
constraints imposed by existing crop rotations. The second case study addresses design of sustainable bulb-based farming
systems in the Netherlands with the purpose of investigating strategic options at crop rotation and farm level to resolve
conflicts between economic and environmental objectives. In the discussion, methodological elements of model-based
explorations and interaction with stakeholders are addressed, and opportunities for enhanced development of sustainable
farming systems are identified. 1997 Elsevier Science B.V.
Keywords: Sustainable agriculture; Farming systems; Cropping systems; Prototyping; Model-based learning; Participatory
research
1. Introduction
In many parts of Europe, arable farmers have been
very successful in increasing yields per unit area dur-
ing the last decades. However, the production techni-
ques that were utilized have resulted in negative side
effects: emissions of pesticides and plant nutrients,
(in)organic waste, high energy consumption. Public
concern is reflected in a suite of national and interna-
tional policy statements that call for more sustainable
agricultural farming systems.
European Journal of Agronomy 7 (1997) 271–283
1161-0301/97/$17.00 1997 Elsevier Science B.V. All rights reserved
PII S1161-0301(97)00042-7
* Corresponding author. Tel.: +31 317 484766; fax: +31 317
484892; e-mail: walter.rossing@staff.tpe.wau.nl
For operational purposes, sustainability can be
defined as a combination of socio-economic, ecologi-
cal and agro-technical objectives of agricultural pro-
duction (WRR, 1995). The weight attributed to these
objectives is value-driven and varies among interest
groups. Because the objectives are usually at least
partially conflicting, development of sustainable
farming systems is characterized by negotiation
about acceptable compromises among objectives by
various stakeholders. Actors include farmers, agri-
cultural industry, consumers and the public sector.
Agricultural research contributes to this process by
developing methodology to demonstrate conse-
quences of alternative options.
During the last decade, a promising empirical meth-
odology for developing sustainable farming systems
has been elaborated, coined ‘prototyping’ (Vereijken,
1994, 1997). Prototyping involves application-ori-
ented design and testing of farming systems in colla-
boration with commercial farmers or at experimental
farms, according to a methodical approach. Four
phases can be distinguished: diagnosis, design, testing
and improvement and dissemination.
In the diagnostic phase the objectives of agri-
cultural production and the value-driven weights
attached to them by various interest groups are estab-
lished, and problems caused by the current system
design are identified. The diagnostic phase should
result in a strategic alliance of stakeholders with a
common motivation to design alternative ways of
agricultural production. In the next phase, farmers
and researchers set out to design new production sys-
tems that better meet the objectives. The result of the
design phase is a number of promising theoretical
prototypes of sustainable production systems. Follow-
ing implementation of these theoretical prototypes,
monitoring their performance in terms of objectives
provides the basis for iterative prototype improve-
ment. This phase of testing and improvement, exe-
cuted on experimental or commercial farms, results
in practical prototypes which have demonstrated
acceptable performance in all objectives. Capitalizing
upon the insights gained during the first three phases,
a larger farmer audience can be addressed during the
final phase, which is aimed at dissemination of the
prototypes within the farming community.
Prototyping suffers from two shortcomings. Only
few theoretical prototypes can be tested, resulting in
a lack of information on trade-off among objectives,
and systems design is based on expertise summarized
in simple rules which unduly narrows the range of
available options and obscures understanding of sys-
tems behaviour. Model-based explorations can rem-
edy the limitations of empirical prototyping. Models
are used because they represent devices for combining
detailed information on system components and creat-
ing system designs that meet objectives of the various
actors involved in farming systems development. The
models are used in an explorative, as opposed to pre-
dictive, fashion. Rather than aiming at predicting
which farming systems are plausible, explorations
focus on designs which are possible in relation to
the objectives of the actors involved. As a conse-
quence, results of explorative studies are presented
as options rather than recipes. While different techni-
ques may be used, including continuous simulation,
rule-based simulation or optimization techniques,
explorative models are mechanistic and integrate
components to create designs at next higher aggrega-
tion levels. The mechanistic approach enables eluci-
dation of causes of calculated system behaviour based
on insight in component behaviour and provides the
opportunity to enhance understanding of systems
behaviour.
To optimally utilize their potential, model-based
explorations should be conducted at different levels
of aggregation of agricultural systems. Objectives and
constraints at the various levels differ, resulting in
conflicting options between levels. For instance, sus-
tainability from a regional perspective may lead to
farming systems that are economically not viable.
The time horizon adopted in a study determines the
type of options available, longer time horizons leading
to designs with more futuristic farm strategies, alter-
native crops or production techniques.
This paper focuses on the contribution of model-
based explorations at the level of field, crop rotation
and farm to design of sustainable farming systems.
Studies aimed at exploring options at regional and
global levels are described by de Wit et al. (1988),
Veeneklaas (1990), van de Ven (1994), Rabbinge et
al. (1994) and Penning de Vries et al. (1995). In the
next section, two case studies are described, with
emphasis on the role of model-based explorations in
different phases of farming systems development. The
first case study illustrates the role of explorations dur-
272 W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
ing diagnosis and design of wheat management sys-
tems at field and crop rotation levels, with a time
horizon of 1–5 years. The second case study exem-
plifies model-based design at the crop rotation and
farm level, with a time horizon of 10–15 years. In
the discussion, methodological elements related to
model-based explorations and their delivery are dis-
cussed and opportunities for enhanced support of
development of sustainable farming systems are iden-
tified.
2. Explorations at the level of field, crop rotation
and farm: two case studies
2.1. Opportunities for improving wheat-based
systems in the Paris Basin of France
In the Paris Basin, farmers often have more than
half their farm land under cereals (mainly wheat and
barley), in rotation with sugar beet, potato, oilseed
rape, fodder peas and sometimes vegetables for indus-
try. High input levels of pesticides and nutrients con-
stitute a threat to environmental sustainability of these
farming systems. A series of studies were carried out
at the aggregation levels of field and crop rotation,
aimed at improving wheat management in the short
term, i.e. with a time horizon of 1–5 years. A combi-
nation of model-based explorations and empirical
research was used to diagnose current practices in
wheat cultivation and design improved systems. In
the explorations, regression models and rule-based
simulation models were used, their nature and role
differing among phases.
2.1.1. Diagnosis
Causes of variation in input efficiency and yield
between wheat crops grown on similar soils and as
part of similar crop rotations were diagnosed. Data on
actual wheat management were obtained through sur-
veys. In each surveyed field, data were collected on
crop (total biomass, yield components, crop nutri-
tional status, crop health), soil (available macro-nutri-
ents, structure of the arable layer) and environmental
conditions (temperature, radiation and precipitation).
Furthermore, information was gathered on whole-
farm management, i.e. crop and soil management
practices (timing of operations, labour requirement
and equipment used) for all crops in the rotation
(Meynard, 1985a; Dore´ et al., 1997).
In the Picardie region, the actual yield ranged from
4 to 9 t grain/ha. Variation in yield appeared to be
more related to variations in ear and grain density
than to variation in weight per grain (Meynard,
1985a). To identify the factors limiting each of these
yield components, a yield gap analysis was carried
out. Potential size of yield components were explored
using a set of linked empirical regression models that
related the size of a yield component in the absence of
limiting or reducing factors, to critical crop variables.
For instance, potential ear density was estimated by
the regression model of Masle (1985) in which bio-
mass of the shoot at the start of stem elongation and
variety are used as input variables, while potential
grain density was described by the regression relation
of Boiffin et al. (1981) that uses shoot biomass at the
start of flowering as input. Next, the yield gap, defined
as the ratio of actual level and potential level ofa yield
component, was correlated to soil and environmental
variables. Correlation and categorical analyses
revealed significant association between yield gap
and two factors: compacted soil structure and belated
applications of nitrogen fertiliser that caused nitrogen
deficiency during the first part of stem elongation
(Meynard and David, 1992). The same factors
decreased nitrogen fertiliser efficiency. For example,
uptake of N on compacted soils was reduced by about
30–50 kg/ha compared to soils without compaction,
which increased risk of leaching of nitrogen.
Belated fertilizer application and soil compaction
appeared correlated to aspects of whole-farm work
organisation. Sowing of sugar beet was demonstrated
to receive priority over simultaneously required ferti-
lizer application in wheat. Ploughing to alleviate soil
compaction, generally caused in crops preceding
wheat, competed for labour with sugar beet harvest-
ing, and was replaced by less time consuming, but
also less effective, shallow tillage. The scale of
these problems varied among farms depending on
the relative importance of the competing crops and
the labour-power and equipment available (Meynard,
1985a; Aubry, 1995).
Yield reduction caused by aphids and diseases was
found to be positively correlated to early sowing
(before mid October), high nitrogen fertilizer input
(about 200 kg/ha), high seed rates (about 300 grains/
273
W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
m
2
for early sown crops) and the level of susceptibility
of the cultivar, except when fungicides and aphicides
had been applied twice (Meynard, 1991). Survey data
were summarized in a multiple regression model that
predicted yield loss relative to yield of a ‘healthy’
crop that had received two pesticide applications
(Chevallier-Ge´rard et al., 1994). Input variables in-
cluded attainable yield, i.e. crop yield in absence of
pests and diseases, year, location, sowing date, varie-
tal resistance spectrum and preceding crop. The model
was used to explore the economic returns on pesticide
application for different combinations of target yield,
sowing dates and cultivar resistance, using weather
data from 1978 to 1991. The results demonstrated
that pesticide application was justified in the vast
majority of years for a susceptible cultivar such as
The´se´e that was sown early with a high yield target.
In contrast, positive returns on pesticide input were
obtained in four out of 14 years only when sowing
occurred after mid November and a more resistant
variety such as Renan was selected in combination
with a lower target yield.
2.1.2. Design
The results of the diagnostic phase stimulated the
design of low-input wheat management systems that
were environmentally friendly and provided eco-
nomic margins that were at least equal to those of
intensive systems. Three major constraints of prevail-
ing systems should be overcome: soil compaction,
belated fertilizer application and disease risk. Work-
ing hypothesis during design of these new systems
was that by adopting a target yield below the level
in conventional systems, it would pay off to sow
later, to reduce sowing rates and nitrogen input, and
to adopt varieties that, although lower yielding than
popular varieties, were more disease resistant. As a
consequence, risks of lodging and infestation by dis-
eases and aphids were expected to be lower and costs
of growth regulators and pesticides would decrease,
while at the same time less labour would be required
(Meynard, 1985a). A range of alternative systems was
assessed by model-based explorations. In the study,
the set of linked regression models that were used
during the diagnostic phase was extended to account
for the entire growing season. For instance, Shinosaki
and Kira’s model, modified by Willey and Heath
(1969) and Meynard (1985b), was added to calculate
aerial biomass at the start of stem elongation using
plant density and sowing date as inputs. This model’s
output was input to calculation of ear density accord-
ing to the model of Masle (1985). The balance sheet
approach (Re´my and He´bert, 1977) was used to cal-
culate the effect of nitrogen fertilizer application on
yield. The complete set of models is described by
Meynard (1985a). Results of the study indicated that
Table 1
Comparison of characteristics of a prototype integrated wheat management system and a conventional intensive system for the Picardie region
in the Paris Basin of France
Aspect Integrated system Intensive system
Target yield range (t/ha) 6.5–7.5 8.0–9.0
N requirement (kg/ha) 195 240
Sowing rate (grain/m
2
)
Sowing before 25 October 180 260
Sowing after 11 November 250 490
Nitrogen fertilizer
First application:
Date 15 February ±10 days
Rate (kg/ha) 40 70
Second application:
Date Beginning of stem elongation
Rate (kg/ha) According to balance sheet method: crop N-requirement minus estimated soil supply
Growth regulator No CCC at start of stem elongation
Fungicides According to damage threshold Two fixed applications: at heading
and 4 weeks before
Prototype design is based on an explorative model-based study by Meynard (1985a).
274 W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
the ‘integrated system’ described in Table 1 would
provide the largest returns.
Widespread application of these design principles
to the diversity of constraints of specific farmers, has
been facilitated by the development of the interactive
software tools ‘De´cible´’ (Aubry et al., 1992; Cheval-
lier-Ge´rard et al., 1994) and ‘Otelo’ (Attonaty et al.,
1993). De´cible´ simulates the effects of crop manage-
ment on wheat yield, gross margin, protein content,
and soil mineral nitrogen at harvest for specific fields
characterized by cropping history, soil type and
weather (Fig. 1). Crop management is described by
a set of decision rules, representative for a farmer or
proposed by researchers and extensionists. In the deci-
sion rules environmental and agronomic conditions
are related to actions. For example, a possible rule
would be: ‘if the wheat crop is in development stage
30 and calculated trafficability of the soil is sufficient,
then apply nitrogen dressing calculated according to
the balance sheet method’. These generic decision
rules are made specific for a particular crop in a par-
ticular year by a ‘decision simulator’. Simulated crop
management is combined with modules of crop
growth and development in a ‘crop simulator’ which
contains the agronomic relations described earlier.
These modules can easily be tested and adapted to
new varieties and different areas. When run with up
to 30 years of historical weather data, De´cible´ enables
exploration of different sets of decision rules, thus
providing information to the farmer for selection of
the set that is most desirable for his specific objectives
regarding grain quality, work organization, or eco-
nomic and environmental goals.
While De´cible´ is used to explore options at the crop
and field level, opportunities for improvement of work
organization at the crop rotation and farm level can be
explored with the interactive software tool Otelo
(Attonaty et al., 1993). Otelo is a rule-based system
which simulates consequences of work prioritization
on dates of tillage, sowing, harvest, etc. for a given
farm. Otelo enables the farmer to explore different
ways to reduce competition among activities and
assess the possible contribution of changing machin-
ery, manpower, or cropping plan. During exploration,
the farmer specifies his decision rules in the same
format as described for De´cible´, for all crops and
activities on the different fields of the farm. These
decision rules are input to Otelo. The farmer then
simulates the dates of the various operations using
weather data from the last two or three years. The
comparison between simulated and actual dates is a
validation of the representation of the farmer’s deci-
sion rules. Such validation determines the quality of
the ensuing explorations, and increases the farmer’s
confidence in the model. Similar to De´cible´, risks
associated with the farmer’s decision rules can be
assessed by running Otelo with weather data for a
period up to 30 years. Risk may be expressed as prob-
ability of exceedence of a threshold value and can be
estimated from simulated frequency distributions for
sowing date, lateness in fertilization, or soil compac-
tion at sowing or harvest. Various sets of decision
rules, cropping patterns or equipment can be com-
pared to those of the farmer. Both in Otelo and in
De´cible´, the agronomist and the farmer iteratively
identify the best organization pattern, integrating the
characteristics of the farm.
Combination of surveys and model-based explora-
tions during diagnosis resulted in a perspective on
bottlenecks in existing wheat-based systems at the
level of wheat crop and crop rotation. In the design
phase, models were used to assess alternative solu-
tions with respect to objectives pertaining to eco-
Fig. 1. General lay-out of De´cible´, a software tool for interactive
design of wheat management systems (after Aubry et al., 1992).
275
W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
nomic returns, environment and labour availability.
Empirical evaluation of the performance of the inte-
grated system that emerged as most promising (Table
1) will be addressed in Section 3.1.
2.2. Opportunities for improving flower bulb based
farming systems in the Netherlands
Current systems of flower bulb production in the
Netherlands use considerable amounts of nutrients
and pesticides per unit area. High prices of product
and land, relatively low input prices and a defensive
attitude among growers towards environmental issues
are among the causes for these high input levels. Leg-
islation is aimed at reducing negative environmental
side-effects, particularly addressing pesticides and
nutrients. To support design of environmentally
more acceptable production systems by an association
of growers and environmentalists, an explorative
study was carried out (Rossing et al., 1997). In the
study, fragmented agronomic information was synthe-
sized in a database and a linear programming optimi-
zation model was used to explore technical options for
flower bulb production with a time horizon of 10–15
years. The choice of time horizon was reflected in the
choice of farm sizes (in terms of labour and area, both
treated as exogenous variables) and in the choice of
production techniques. The study focused on farms
located on coarse sandy soils in the west of the Neth-
erlands, allowing rent of land for bulb production on
heavier soils further away from the farm.
In accordance with the operational definition of
sustainability proposed by WRR (1995), a distinction
was made between value-driven objectives and fact-
driven agronomic information. One economic and
two environmental objectives were formulated in
interaction with the association of stakeholders. The
economic objective was represented by maximization
of farm gross margin. The environmental objectives
were minimization of pesticide input expressed in kg
active ingredient (a.i.) averaged over the cropped area
and minimization of nitrogen surplus calculated as
nitrogen not taken up by the crop and not transferred
to a subsequent crop, averaged over the cropped area.
Important value-driven constraints that were formu-
lated comprised farm size, the possibility to rent addi-
tional land free of soil-borne pests and diseases, and
the variety of crops that could be grown.
Agronomic information was synthesized to define
management systems for four bulb crops, i.e. tulip,
narcissus, hyacinth and lily, and for one break crop,
i.e. winter wheat, which has positive effects on soil
structure and soil health. Crop management systems
were characterized by soil type and soil health, crop-
ping frequency, crop protection regime and nutrient
regime. The characteristics were chosen such that a
wide array of crop production techniques could be
defined that varied distinctively in terms of the objec-
tives of flower bulb production. In addition, inter-crop
management systems were defined, such as soil fumi-
gation, inundation, and prevention of wind erosion
with straw. Congruent with the time horizon of 10–
15 years, attention was focused on production techni-
ques still in an experimental stage and techniques
derived from other crops, rather than on current prac-
tices only. For all specified crop and inter-crop man-
agement systems inputs and outputs were formulated
using empirical information, expertise and production
ecological theory (Rabbinge, 1993; de Koning et al.,
1995; van Ittersum and Rabbinge, 1997).
Crop and inter-crop management systems were
combined to rotations in a multiple goal linear pro-
Fig. 2. Calculated maximum farm gross margin (index, see Table
2) associated with combinations of farm-based average pesticide
input (kg active ingredient/ha) and farm-based average nitrogen
surplus (kg N/ha) for a farm with 15 ha sandy soils, three full
time labour equivalent, and optional rent of land. Optional crops
on sand: tulip, hyacinth, narcissus, lily and winter wheat; on clay:
tulip and narcissus. Points of equal farm gross margin are con-
nected (iso-lines). Each combination of pesticide input and nitro-
gen surplus for which maximum farm gross margin is calculated is
indicated as x. Arrows indicate development paths (see text and
Tables 2 and 3). (Rossing et al., 1997; reprinted with permission of
Kluwer Academic).
276 W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
gramming approach (de Wit et al., 1988; Schans,
1996) to allow evaluation of objectives. By maximis-
ing farm gross margin at increasingly tighter con-
straints on the environmental objectives, the trade-
off between market and environment was explored.
The reference situation represents a production sys-
tem which just meets the (anticipated) governmental
targets with respect to pesticide input and nitrogen
surplus for the year 2000. Two development paths
were assessed, representing gradually reduced pesti-
cide use and N-surplus, respectively (Fig. 2). The
development path for pesticide reduction (Table 2)
shows that in the first step a substantial reduction in
pesticide input may be achieved with relatively little
loss of farm gross margin. This is achieved mainly by
substituting soil fumigation by inundation and adop-
tion of new low-dosage fungicides in tulip production.
No changes in cropping sequence or area rented land
occur. Further reduction in pesticide input (step 2) is
most economically accomplished by abolishing the
use of mineral oil for virus control in lily. The asso-
ciated yield loss in lily causes a reduction in farm
gross margin. Again, no changes in cropping fre-
quency occur. The third step, zero pesticide input,
causes major changes: the rotation changes from
tulip-lily-wheat to narcissus-wheat and farm gross
margin becomes negative. In all steps, the rented
land, free of soil-borne pests and diseases remains
approximately 11 ha. On this rented area tulip is
grown with a relatively moderate pesticide input of
12 kg (a.i.)/ha.
In contrast, the results for the development path for
nitrogen surplus reduction show that with the defined
techniques, N-surplus reduction is only possible at the
expense of a considerable reduction in income. A
decrease in N-surplus of 30% beyond the levels antici-
pated for 2000 is associated with a 40% decrease in
farm gross margin (Table 3). In the cropping sequence
lily is replaced by narcissus, which has a much lower
gross margin but higher N-efficiency. Experiences on
two experimental farms and current trends in the sec-
tor support the conclusion that reducing pesticide use
affects farm income less than N-surplus reduction.
Remedy may be sought in development of new tech-
nologies, aiming at more precise application of nutri-
ents in time and space, or in re-evaluation of strategic
choices, such as the current use of alluvial sandy soils
for growing the bulk of nutrient-inefficient flower
bulb crops. The sensitivity of results to farm size,
range of crops, prices and assumptions on input-out-
Table 2
Exploration of flowerbulb production systems under increasingly tighter constraints on pesticide input (kg active ingredient/ha) for a farm of
15 ha sandy soils, three full time labour equivalent, and possibility to grow tulip, narcissus, lily, hyacinth and winter wheat
Reference
a
Step 1
b
Step 2 Step 3
Objectives
Farm gross margin (indexed) 100 97 77 −4
Average pesticide input (kg a.i./ha) 50 30 10 0
Average nitrogen surplus (kg N/ha) 140 140 140 140
Production techniques
Fraction area per crop (%)
Tulip 33 33 33 –
Lily 33 33 33 –
Narcissus – – – 50
Winter wheat 33 33 33 50
Pesticide input per crop (kg a.i./ha)
Tulip 18 12 12 0
Lily 86 78 18 –
Narcissus – – – 0
Winter wheat 0 0 0 0
Area fumigated (ha) 1.2 0 0 0
Area rented (ha) 11 11 10 11
a
The reference farming system, equivalent to point A in Fig. 1, just meets the anticipated governmental targets regarding pesticide input and
nitrogen surplus for the year 2000. The associated farm gross margin has index value 100. Zero gross margin has index value 0.
b
Step 1 results in point E in Fig. 1, step 2 in point F, step 3 in point G.
277
W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
put relations is reported elsewhere (Rossing et al.,
1997).
The approach of separating objectives and bio-phy-
sical options was much appreciated by the association
of growers and environmentalists and resulted in brid-
ging the gap between the two parties involved. The
existing polarization appeared to be caused by diver-
gent views on objectives, rather than by disagreement
on bio-technical relations. The perspective on the
trade-offs among all objectives focused the discussion
on preferred development pathways. While the a
priori outlook of growers was especially focused on
tactical decision making, the study increased aware-
ness of the importance of strategic choices over tac-
tical choices (Rabbinge and Zadoks, 1989). In
particular, the importance of introducing a soil health
restoring break crop in the rotation, such as winter
wheat, and renting healthy land proved to be impor-
tant strategic options for mitigating the decrease in
farm gross margin associated with less pesticide
input and lower nitrogen surplus. Based on the results
of the study, participating farmers actively promoted
research on ecology of soil-borne pests at their experi-
mental station in response to the lack of knowledge
that had become apparent during the explorations.
Despite uncertainty in a number of the agronomic
relations, the results were deemed sufficiently robust
for testing and improvement on commercial farms. A
major project was formulated and is anticipated to
start in 1998. The project envisages continuous train-
ing of selected farmers and extensionists and efforts
are currently focused on adapting and extending the
exploratory design tools for this educational purpose.
3. Discussion
3.1. Methodological aspects of model-based
explorations
In the introduction of this paper, model-based
explorations were put forward to supplement empiri-
cal prototyping and to remedy its shortcomings: the
limited number of production systems that can be
evaluated and the rules of thumb used during the
design process. The case studies demonstrated the
capacity of models to explore large numbers of alter-
native production systems and to enhance understand-
ing of systems behaviour due to the transparency of
model components. The case studies differed with
respect to model types and aggregation levels.
In the Dutch case study, input-output relations
stored in databases that were linked to a linear pro-
gramming model were used to address strategic
changes in flower bulb production systems needed
to resolve conflicts between economic and environ-
mental objectives. In the French case study, a combi-
nation of regression models and rule-based simulation
models were used for tactical exploration of wheat
management systems aimed at adjustment of bottle-
Table 3
Exploration of flowerbulb production systems under increasingly tighter constraints on nitrogen surplus for a farm of 15 ha sandy soils, three
full time labour equivalent, and possibility to grow tulip, narcissus, lily, hyacinth and winter wheat
Reference
a
Step 1
b
Step 2 Step 3
Objectives
Farm gross margin (indexed) 100 62 38 2
Average pesticide input (kg a.i./ha) 50 50 9 8
Average nitrogen surplus (kg N/ha) 140 100 90 55
Production techniques
Fraction area per crop (%)
Tulip 33 33 50 –
Lily 33 16 – –
Narcissus – 16 – 50
Winter wheat 33 33 50 50
Area rented (ha) 11 11 11 12
a
The reference farming system just meets the anticipated governmental targets regarding pesticide input and nitrogen surplusfor the year 2000.
The associated farm gross margin has index value 100. Zero gross margin has index value 0.
b
Step 1 results in point B in Fig. 1, step 2 in point C, step 3 in point D.
278 W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
necks within the constraints imposed by existing crop
rotations. In the regression models agronomic infor-
mation was summarized to assess potential production
during diagnosis or target production levels during
design. The rule-based models were used to mimic
farm management decisions, both of a particular
farmer and in an explorative sense. In principle,
results of these tactical explorations provide input-
output relations for strategic optimization studies.
Establishing such link constitutes an important re-
search area to be developed, as it would improve the
coherence and consistency of explorative studies at
different time and spatial scales.
The case studies indicate that relevant answers
require explorative studies at different aggregation
levels. Only by combining the opportunities at the
crop rotation level with those at the crop level, soil
and nutrient management in wheat could be improved
without sacrifices in other crops caused by labour con-
straints. For flower bulbs, a study at the sectoral level
would be desirable to explore the implications of var-
ious options identified at the farm level, because
prices are largely determined by the production vol-
ume realised in the Netherlands.
The case study in the Netherlands demonstrated the
usefulness of sensitivity analysis to reveal gaps in
knowledge relevant to the problem. The consequences
of uncertainty in agronomic knowledge were revealed
by varying single parameters or parameters in a single
relation and assessing the resulting change in model
output in the conventional way (cf. Janssen, 1994).
However, sensitivity in linear programming models
represents a special case, because, typically, uncer-
tainty in agronomic knowledge may have little effect
on the realization of objectives, but leads to very dif-
ferent optimal production systems (Scheele, 1992;
Hijmans and van Ittersum, 1996). New mathematical
techniques are needed to reveal the range of produc-
tion systems that results in similar levels of satisfac-
tion of objectives.
By definition, models are simplified representations
of reality, targeted at capturing the essential elements
of system dynamics. To be relevant, statements based
on model calculations should be accompanied by an
indication of their quality. Quality assessment may
address model components at the field level, such as
a single regression relation or an input-output relation,
and compare it to reality. Concepts for model valida-
tion or evaluation at the field level have been
described by various authors (e.g. Teng, 1981; Ros-
sing, 1991). A similar approach to quality assessment
of models at the level of crop rotations or farms is less
useful because the large number of uncontrollable
variables impedes classical experimental design.
Quality may then be interpreted as the degree to
which a model-based systems design that emerged
as potentially successful, performs better than an
existing system. In such output-oriented assessment
of model quality, causality requires attention: the
model results must be better for the right reasons.
Approaches to output-oriented assessment of model
quality that were developed in the French case study
include (1) cropping system experiments based on
decision rules, and (2) monitoring of farms that have
adopted the prototype systems. In a cropping system
experiment alternative wheat management systems
are evaluated on-farm for their effects on a set of
objectives. Each management systems consists of a
specific set of decision rules that emerged as promis-
ing from the design phase and was further refined by
discussion among farmers, advisers and researchers
(Meynard et al., 1996). Variation in production situa-
tions among farms is taken into account by executing
the experiment in a network of farms. This approach
was adopted by Meynard (1985a, 1991) for evaluating
the simulated design for integrated wheat manage-
ment (Table 1). During 4 years, 28 on-farm experi-
ments were executed in which this integrated system
was compared to the conventional system. Outputs of
the integrated system that were compared to the con-
ventional system comprised mean gross margin, yield
variability, and risk of nitrate leaching. The integrated
system appeared better than the conventional system
from both the economic and environmental point of
view (Table 4). An important spin-off of farmers
being responsible for execution of the experiments
in their fields, was the increased credibility of the
results to the farming community.
In a survey of farms that had adopted integrated
wheat production systems, Aubry (1995) showed
that farmers modified the underlying decision rules
to simplify their decision-making tasks. Farmers clas-
sified their large number of different wheat fields in
groups which could be treated in similar ways and
monitored crop development and diseases in only
one field. Further improvement of the relevance of
279
W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
systems design may be expected when during design
the set of fields on the farm sown to a given crop, i.e. a
decision-making level intermediate between the crop
and the crop rotation, is taken into account.
3.2. Interaction with stakeholders
As was argued in the introduction to this paper,
development of sustainable farming systems implies
negotiations about change and has an important social
dimension. Contributions by model-based explora-
tions to this social aspect can be assessed in terms
of ‘product’, ‘process’ and ‘instrument’.
The contribution of explorative studies may be
assessed in terms of their envisaged product, i.e.
change in perceptions and/or change in actions of
actors in the agricultural knowledge chain. Such
impact assessment may help to improve design and
delivery of explorative studies, but few retrospective
studies of this sort have been carried out and reported
(Sebillotte, 1996). The funding body informally eval-
uated the case study on flower bulb systems by asking
a journalist to interview the participants. Counter-
intuitive results of the study were reported with
respect to the importance of introducing a soil health
restoring break crop in the rotation, and with respect
to renting healthy land. In contrast to these strategic
options, the a priori attention of the participants had
been focused on improving management of pests, dis-
eases and nutrients. The study prompted a range of
activities by the association of farmers and environ-
mentalists. After completion of the study, the associa-
tion continued discussions, involved other parties
from the flower bulb industry, and formulated a pro-
posal for testing and improvement of prototype sys-
tems of integrated flower bulb farming that was
widely supported.
Apart from its outcome, the process, i.e. the execu-
tion of an explorative study in itself may contribute to
development of sustainable farming systems by sti-
mulating discussions among stakeholders based on
scientific information. In both case studies, partici-
pants considered communication and reflection on
sustainability to be improved as a result of the ex-
plorative approach in diagnosis and design. Essential
elements in this respect are the clear separation of
value-driven objectives and fact-driven options, the
quantitative nature of model results that enabled dis-
cussion on acceptable trade-offs among objectives,
and the transparency of underlying information
which improved understanding of system behaviour.
Effective contribution to the process of design by sta-
keholders necessitates research planning in which
communication between researchers and stakeholders
is explicitly taken into account. For instance, during
the flower bulb case study researchers interacted with
a delegation of the association once every 6 weeks to
discuss general progress, and the association orga-
nized two workshops to formulate the value-driven
objectives that were used to evaluate options during
Table 4
Performance of a prototype integrated wheat management system and a conventional intensive system in 28 on-farm experiments during 4
years, on different soil types and with different previous crops in the Picardie region in the Paris Basin of France
Aspect Integrated system Intensive system
Actual yield
Mean (t/ha) 7.5 8.0
Highest mean yield (cases out of 28 experiments) 4 24
Gross margin
Highest mean gross margin (cases out of 28 experiments) 21 7
Standard deviation of yield (t/ha) 1.10 1.14
N-recovery
Highest mean N-recovery (cases out of 28 experiments) 22 6
Fungicide treatment
a
Average number 1.4 2.0
Agronomic details of the systems are described in Table 1.
Soil tillage, sowing date and herbicide application were farmer-specific. In the systems, the same variety was used.
a
In additional experiments with a more disease resistant variety the average number of fungicide treatments was 0.7.
280 W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
the study. These frequent social interactions may well
have paved the way for the projects ‘product’
described in the previous paragraph.
The instruments, i.e. the models that were used
during the explorations served as computer-aided
learning tools for stakeholders (Leeuwis, 1993; Cerf
et al., 1994; Papy, 1996). Characteristically, models
were used to answer ‘what-if’ questions, with
researchers acting as intermediates between model
and stakeholders (in the flower bulb case) or process
facilitators (in the wheat case). Fast response to ‘what-
if’ questions is necessary to maintain process momen-
tum, requiring further improvement of the linear pro-
gramming model concerning user-friendliness,
flexibility and methodology for sensitivity analysis.
While in the flower bulb case the model itself was
meant to be used by researchers only, the tools
Otelo and De´cible´ were designed for interactive appli-
cation by farmers and extensionists to a specific farm.
Two aspects were found to influence the success of the
application (Mousset et al., 1997). Firstly, the users
should be able to gain confidence in the tool. Farmers
were usually able to recall their agronomic decisions
of the preceding three years, which enabled assess-
ment of model quality by comparison of simulation
results with actual data. Secondly, the amount of time
spent in creating input for the model should be com-
mensurate to the expected value of information of
model output. Especially for Otelo, the amount of
farm-specific input data is considerable, and the tool
is used only for complex situations, requiring innova-
tive solutions.
3.3. Model-based explorations and empirical
prototyping
The case studies in this paper demonstrated how
model-based explorations supplement empirical pro-
totyping during the first two phases of sustainable
farming systems development: diagnosis and design.
The diagnostic surveys of the prototyping approach
were complemented by modeling studies exploring
production potential. Their combination enabled iden-
tification of constraints in current practices and
assessment of opportunities for improvement. Oppor-
tunities were elaborated during the model-based
design phase to reveal trade-offs among objectives.
Although not substantiated by the case studies, sup-
plementation of prototyping by model-based explora-
tions may also be expected during the last two phases
of development of sustainable farming systems: test-
ing and improvement, and dissemination. Promising
options emerging from the design phase are put to the
test empirically. During testing and improvement of
these prototypes, explorations can reveal yield gaps or
trends in slow processes such as soil organic matter
turn-over. At the start of dissemination, empirical pro-
totyping has resulted in prototype systems that have
proven their value in practice, while essential ele-
ments of production systems are synthesized in mod-
els at different levels of aggregation to facilitate
extrapolation to new conditions. During all phases
model-based explorations may indicate gaps in
knowledge of researchers, extensionists and farmers,
thus contributing to learning about the system by all
actors.
The pivotal role of learning appears prominently in
farmer-oriented projects that have been successful in
stimulating more judicious use of resources (Zadoks,
1989; Kenmore, 1991; Sebillotte, 1996). Two impor-
tant characteristics of learning are (1) cyclic iteration
of experimentation, action, observation and reflection,
and (2) repeated switching between aggregation
levels, time periods, knowledge domains and farm
types. Model-based explorations in interplay with
prototyping have the potential of contributing to
such non-formal education. To realize this potential
requires a research approach in which equal attention
is devoted to creating and synthesizing relevant agro-
nomic knowledge and to creating settings in which
learning can take place (Chatelin et al., 1994; Okali
et al., 1994; Ro¨ling, 1996; Somers, 1997).
References
Attonaty, J.M., Chatelin, M.H., Poussin, J.C. and Soler, L.G., 1993.
Advice and decision support systems in agriculture: new issues.
In: Farm Level Information Systems, Woudschoten, Zeist, The
Netherlands.
Aubry, C., 1995. Gestion de la solel d’une culture dans l’exploita-
tion agricole. Cas du Ble´ d’Hiver en Grande Culture dans la
Re´gion Picarde. Doctoral Thesis, Institut National Agronomique
Paris-Grignon, 277 pp.
Aubry, C., Chatelin, M.H., Poussin, J.C., Attonaty, J.M., Meynard,
J.M., Ge´rard, C. and Robert, D., 1992. De´cible´: a decision sup-
port system for wheat management. In: Book of Abstracts, 4e`me
Congre`s d’Informatique agricole, Versailles, SAF, Paris.
281
W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
Boiffin, J., Caneill, J., Meynard, J.M. and Sebillotte, M., 1981.
Elaboration du rendement et fertilisation azote´e du ble´ d’hiver
en Champagne Crayeuse. I. Protocole et me´thode d’e´tude d’un
proble`me technique re´gional. Agronomie, 1: 549–558.
Cerf, M., Papy, F., Aubry, C. and Meynard, J.M., 1994. Agronomic
theory and decision tools. In: J. Brossier, L. de Bonneval and E.
Landais (Editors), Systems Studies in Agriculture and Rural
Development. INRA, Paris, pp. 343–356.
Chatelin, M.H., Mousset, J. and Papy, F., 1994. In: B.H. Jacobsen,
D.E. Pedersen, J. Christensen and S. Rasmunsen (Editors), Farm-
ers’ Decision-Making – A Description Approach. Proc. from the
38th EAAE seminar, Institute of Agricultural Economics and the
Royal Veterinary and Agricultural University, Copenhagen, pp.
369–381.
Chevallier-Ge´rard, C., Denis J.B. and Meynard J.M., 1994. Perte de
rendement due aux maladies cryptogamiques sur ble´ tendre d’hi-
ver. Construction et validation d’un mode`le de l’effet du syste`me
de culture. Agronomie, 14: 305–318.
De Koning, G.H.J., van Keulen, H., Rabbinge, R. and Janssen, H.,
1995. Determination of input and output coefficients of cropping
systems in the European community. Agric. Syst., 48: 485–
502.
De Wit, C.T., van Keulen, H., Seligman, N.G. and Spharim, I.,
1988. Application of interactive multiple goal programming
techniques for analysis and planning of regional agricultural
development. Agric. Syst., 26: 211–230.
Dore´, T., Sebillotte, M. and Meynard, J.M., 1997. A diagnostic
method for assessing regional variations in crop yield. Agric.
Syst., 54: 169–188.
Hijmans, R.J. and van Ittersum, M.K., 1996. Aggregation of spatial
units in linear programming models to explore land use options.
Neth. J. Agric. Sci., 44: 145–162.
Janssen, P.H.M., 1994. Assessing sensitivities and uncertainties in
models: a critical evaluation. In: J. Grasman and G. van Straten
(Editors), Predictability and Non-Linear Modelling in Natural
Sciences and Economics. Kluwer Academic, Dordrecht, pp.
344–361.
Kenmore, P.E., 1991. How rice farmers clean up the environment,
conserve biodiversity, raise more food, make higher profits –
Indonesia’s IPM – a model for Asia. FAO Inter-Country Pro-
gramme for Integrated Pest Control in Rice in South and South-
east Asia, Manila, Philippines, 56 pp.
Leeuwis, C., 1993. Of computer, myths and modelling. The social
construction of diversity, knowledge, information and commu-
nication technologies in Dutch horticulture and agricultural
extension. Wageningen Studies in Sociology, Pudoc, Wagenin-
gen, Vol. 36, 468 pp.
Masle, J., 1985. Competition among tillers in winter wheat: con-
sequences for growth and development of the crop. In: W. Day
and R.K. Atkin (Editors), Wheat Growth and Modelling. Plenum
Press, New York, pp. 33–54.
Meynard, J.M., 1985a. Construction d’itine´raires techniques pour
la conduite du ble´ d’hiver. Doctoral Thesis, Institut National
Agronomique Paris-Grignon. Paris, 258 pp.
Meynard, J.M., 1985b. Les besions en azote du ble´ d’hiver jusqu’au
de´but de la montaison. Agronomie, 5: 579–589.
Meynard, J.M., 1991. Pesticides et itine´raires techniques. In: P.
Bye, C. Descoins and A. Deshayes (Editors), Phytosanitaires,
Protection des Plantes, Biopesticides. INRA, Paris, pp. 85–
100.
Meynard, J.M. and David, G., 1992. Diagnostic sur l’e´laboration du
rendement des cultures. Cahiers Agric., 1: 9–19.
Meynard, J.M., Reau, R., Robert, D. and Saulas, P., 1996. Evalua-
tion expe´rimentale des itine´raires techniques. In: G. Urbano et al.
(Editors), Experimenter sur les Conduites de Culture. DERF/
ACTA, Paris, pp. 63–72.
Mousset, J., Aslahe´, C., Billa, Ph., Boiffin, J., Chatelin, M.H.,
Chopplet, M., Franc¸ois, M., Gandon, H., Groell, F., His, M.,
Hopquin, J.P., Klein, D., Papy, F., Quie´vreux, D. and Soler,
L.G., 1997. Le conseil Agro-Equipement en Picardie: MECA-
GRO. Colloque ‘Aide a`lade´cision et choix de strate´gies dans les
entreprises agricoles’, Laon (France), 10–11/12/1996, INRA, pp.
195–206.
Okali, C., Sumberg, J. and Farrington, J., 1994. Farmer Participa-
tory Research. Rhethoric and Reality. Intermediate Technology
Publications, London, 159 pp.
Papy, F., 1996. Modelling the farm as a business to help farmers
make their decisions: a summary review. In: J.P. Colin, E. Craw-
ford and C. Fillonneau (Editors), Research Methodology for
Agricultural Systems Analysis. In press.
Penning de Vries, F.W.T., van Keulen, H. and Rabbinge, R., 1995.
Natural resources and limits to food production in 2040. In: J.
Bouma, A. Kuyvenhoven, B.A.M. Bouman and J.C. Luyten
(Editors), Eco-Regional Approaches for Sustainable Land Use
and Food Production. Kluwer Academic, Dordrecht, The Nether-
lands, pp. 65–87.
Rabbinge, R., 1993. The ecological background of food production.
In: D.J. Chadwick and J. Marsh (Editors), Crop Protection and
Sustainable Agriculture. Ciba Foundation Symposium 177,
Wiley, Chichester, pp. 2–29.
Rabbinge, R., van Diepen, C.A., Dijsselbloem, J., de Koning,
G.J.H., van Latesteijn, H.C., Woltjer, E. and van Zijl, J., 1994.
‘Ground for choices’: A scenario study on perspectives for rural
areas in the European community. In: L.O. Fresco, L. Stroosnij-
der, J. Bouma and H. van Keulen (Editors), The Future of the
Land: Mobilising and Integrating Knowledge for Land Use
Options. Wiley, New York, pp. 95–121.
Rabbinge, R. and Zadoks, J.C., 1989. Disease and pest control. In:
J.C. Zadoks (Editor), Development of Farming Systems. Pudoc,
Wageningen, pp. 32–39.
Re´my, J.C. and He´bert, J., 1977. Le devenir des engrais azote´s dans
le sol. Comptes Rendus de l’Academie d’Agriculture de France,
63: 700–710.
Ro¨ling, N.G., 1996. Towards an interactive agricultural science.
Eur. J. Agric. Educ. Ext., 2: 35–48.
Rossing W.A.H., 1991. Simulation of damage in winter wheat
caused by the grain aphid Sitobion avenae. 3. Calculation of
damage at various attainable yield levels. Neth. J. Plant Pathol.,
97: 87–103.
Rossing, W.A.H., Jansma, J.E., de Ruijter, F.J. and Schans, J.,
1997. Operationalizing sustainability: exploring options for
environmentally friendly flower bulb production systems. Eur.
J. Plant Pathol., 103: 217–234.
Schans, J., 1996. MGOPT_CROP, a Multiple Goal Linear Pro-
282 W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
gramming Model for Optimisation of Crop Rotations, version
1.0. AB-DLO, Wageningen, Report 69, 48 pp.
Scheele, D., 1992. Formulation and characteristics of GOAL. Tech-
nical Working Document (W64), Netherlands Scientific Council
for Government Policy, The Hague, The Netherlands, 64 pp.
Sebillotte, M., 1996. Les Mondes de l’Agriculture: Une Recherche
pour Demain. INRA, Paris, 258 pp.
Somers, B.M., 1997. Learning for sustainable agriculture. In: M.K.
van Ittersum and S.C. van de Geijn (Editors), Perspectives for
Agronomy – Adopting Ecological Principles and Managing
Resource Use. Elsevier Science, Amsterdam, The Netherlands,
pp. 353–359.
Teng, P.S., 1981. Validation of computer models of plant disease
epidemics: a review of philosophy and methodology. Z. Pflan-
zenkr. Pflanzenschutz, 88: 49–63.
Van de Ven, G.W.J., 1994. A mathematical approach for compar-
ison of environmental and economic goals in dairy farming at
the regional scale. In: L.’t Mannetje and J. Frame (Editors),
Grassland and Society. Proceedings of the 15th General Meeting
of the European Grassland Federation (EGF), Wageningen,
6–10 Juni 1994, Wageningen Press, Wageningen, pp. 453–
457.
Van Ittersum, M.K. and Rabbinge, R., 1997. Concepts in produc-
tion ecology for analysis and quantification of agricultural input-
output combinations. Field Crops Res., 52: 197–208.
Veeneklaas, F.R., 1990. Dovetailing technical and economic ana-
lysis. Doctoral Thesis, Erasmus University Rotterdam, The
Netherlands, 159 pp.
Vereijken, P., 1994. Designing prototypes. Progress report 1. Pro-
gress reports of research network on integrated and ecological
arable farming systems for EU and associated countries. AB-
DLO, Wageningen, 87 pp.
Vereijken, P., 1997. A methodical way of prototyping integrated
and ecological arable farming systems (I/EAFS) in interaction
with pilot farms. Eur. J. Agr. 7. 235–250.
Willey, R.W. and Heath, S.B., 1969. The quantitative relationships
between plant population and crop yield. Adv. Agron., 21: 281–
331.
WRR, 1995. Sustained risks, a lasting phenomenon. Reports to the
Government No. 44, Netherlands Scientific Council for Govern-
ment Policy, The Hague, The Netherlands, 205 pp.
Zadoks, J.C., 1989. EPIPRE, a computer-based decision support
system for pest and disease control in wheat: Its development
and implementation in Europe. In: K.J. Leonard and W.E. Fry
(Editors), Plant Disease Epidemiology, Volume II. Macmillan,
New York, pp. 3–29.
283
W.A.H. Rossing et al. / European Journal of Agronomy 7 (1997) 271–283
