Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Environmental Modelling & Software 24 (2009) 359–370

Contents lists available at ScienceDirect

Environmental Modelling & Software


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

Fuzzy knowledge-based model for soil condition assessment in Argentinean


cropping systems
Diego O. Ferraro*
IFEVA, Cátedra de Cerealicultura, Facultad de Agronomı́a, Universidad de Buenos Aires/CONICET, Argentina

a r t i c l e i n f o a b s t r a c t

Article history: A knowledge-based system (KBS) for assessing soil condition in agroecosystems is presented. The KBS
Received 3 September 2007 was built through expert opinion elicitation and available scientific data using fuzzy logic. The system is
Received in revised form 25 July 2008 structured into three main elements: (1) input variables that represent the physical domain of soil
Accepted 28 July 2008
condition assessment and are related to environmental and crop management conditions; (2) primary
Available online 10 October 2008
modules that describe the fuzzy nature of the soil indicators and; (3) secondary modules that represent
the elicited knowledge on soil condition assessment from an expert panel. The application of the KBS on
Keywords:
data on crop fields from Inland Pampa (Argentina) indicated that soil nitrogen depletion poses a hazard
Fuzzy assessment
Sustainability for soil health as no crop was able to accomplish more than 50% of the sustainability criteria elicited for
Knowledge-based systems soil nitrogen extraction from the system. Conversely, soil carbon and physical conditions exhibited values
Soil indicators closer to the desirable scenarios elicited by the fuzzy if-then rules, with values of 0.84, 0.71 and 0.74 for
Argentina maize, soybean and wheat, respectively, where higher indicator values reflect better soil condition
assessment. No significant differences were observed in the overall soil degradation module between
crops, with values of 0.64 for maize and wheat and 0.67 for soybean. The KBS developed in this work
provided an alternative modeling tool for assessing agroecosystem condition when knowledge regarding
long-term assessment is imprecise and uncertain.
Ó 2008 Elsevier Ltd. All rights reserved.

1. Introduction et al., 1994). However, as there are many conflicting frameworks to


develop indicators, it is unclearly the best way to collect data (Reed
An important number of negative externalities of modern et al., 2006) or reach consensus not only in qualitative but also in
agriculture models have been extensively documented (Matson the quantitative issues (Farshad and Zinck, 1993). Also, the infer-
et al., 1997). The growing dependence on non-renewable resources, ence of long-term soil condition and the complexity in system
the loss of biodiversity, the loss of land through soil erosion and the assessment requires a specific framework analysis that is able to
heavy reliance on chemical fertilizers and pesticides are among evaluate extremely divergent parameters, like science-based,
them. The acknowledgement of these negative effects has coined quantitative data and normative settings (like political threshold
the agricultural sustainability concept (Mebratu, 1998). Briefly, values or goals of landscape protection) which are mainly qualita-
a sustainable agroecosystem has to be able to guarantee an tively described and depend strongly on actual socio-political
economic profit, a social equitability, and the improvement of conditions (von Wirén-Lehr, 2001). In order to overcome these
natural capital in a long-term perspective (WCED, 1987). However, difficulties, the use of expert knowledge may improve the reliability
this conceptual approach needs to be complemented with clear of modeling (Yamada et al., 2003). Expert opinion elicited from
operational definitions in order to develop indicators for testing a specialist panel is a much more straightforward way to define
long-term attainment of a desired condition (Lara and Stancu- system behavior rules than multivariate statistic-selected indica-
Minasian, 1999). tors because it can be combined with data regarding variables on
In agricultural sustainability assessment, soil condition is a main which no data exist or is it very difficult to obtain (Marcot et al.,
issue since it is linked not only to food and fiber productivity but 2001; Andrews et al., 2004). Moreover, the expert-based indicators
also to ecosystems function and the maintenance of local, regional, have performed equally well in describing the whole system vari-
and global environmental quality (Glanz, 1995). Traditionally, soil ation (Andrews et al., 2002; Martin et al., 2005).
quality assessments are aimed to identify soil properties that are Knowledge-based systems (synonymous with expert systems)
more sensitive to change in management than others (Gregorich are models that facilitate the integration of quantitative and qual-
itative reasoning methods (Saunders et al., 2005). They are able to
* Tel./fax: þ54 11 4524 8053x44. provide a formal logical specification for the interpretation of
E-mail address: ferraro@agro.edu.ar information and store both the knowledge and experience of

1364-8152/$ – see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.envsoft.2008.07.006
360 D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370

human experts and the available scientific and empiric data in way as well as to make a quantitative environmental assessment
a specific area of interest (Giarratano and Riley, 1989). Particularly, focused on the soil component, the KBS was applied using data
the development of knowledge-based systems is considered as from crop fields from Inland Pampa (Argentina). The KBS developed
a suitable approach for assessing the environmental condition of is meant to provide an assessment tool, especially to policy makers,
agroecosystems (Davis, 1995; Girard and Hubert, 1999). Regarding local stakeholders and research workers for evaluating current
the long-term assessment of soil condition, a KBS may offer a flex- management practices and comparing alternative management
ible and consistent analysis framework by dealing successfully with practices.
multiple sources of information, complex interactions between
system components and the essential translation of system 2. Materials and methods
condition in descriptions related to management decisions (Girard
and Hubert, 1999). 2.1. Model description
Knowledge-based systems generally contain two major
Model structure contains indicators regarding soil carbon and nitrogen stocks
components, a knowledge base (i.e. expert opinion) and an infer- resulting from crop management (i.e. tillage, harvest, yields, and fertilization) as
ence mechanism for drawing conclusions (Paterson et al., 2008). well as the soil physical degradation and the toxicity effects of pesticides on soil
Fuzzy logic has been used meaningfully in knowledge-based biota. Briefly, the model structure was organized into three main elements: (1) the
systems for both the knowledge representation and inference input variables that represent the physical domain of the soil condition assessment
and are related to both environmental and crop management conditions; (2) the
mechanisms (Zimmermann, 1996; Tan, 2005). It is a very flexible
primary modules that describe the fuzzy nature of the indicators and; (3) the
framework that allows integration of different types of information secondary modules that are based on the elicited knowledge from an expert panel
to formalize conclusions, and has already been applied to (details about the expert panel skills and selection procedure are given in Appendix
ecosystem assessment in agriculture (van der Werf and Zimmer, A.1). The model was hierarchically structured in a bottom-up manner by combining
1998; Metternicht and Gonzalez, 2005); forestry (Ducey and Lar- input variables in 11 primary modules (Table 1), and these ones in 6 secondary
modules (Fig. 1). Input variables are used for primary module calculation through
son, 1999; Iliadis, 2005) and grassland management (Pearson and the membership functions illustrated in Table 1 (see Technical appendix A for
Ison, 1997). Also, fuzzy logic is capable to handle the goals ambi- calculation details). They were based on the results of a literature review and
guity and is well-suited for eliciting the expert knowledge when reproduce the impact of these input variables on the selected key components of soil
data is lacking or there is no full agreement in the desirable level of integrity. Knowledge review was useful for defining the input variable values for full
membership of both fuzzy subsets (i.e. Min and Max). Linear membership functions,
the system key attributes (Cornelissen et al., 2001; Fleming et al.,
with two thresholds assigning each parameter value to a ‘‘Min’’ and/or ‘‘Max’’ class,
2007). were chosen due to their simplicity (Mowrer, 2000). Also, a solution obtained by
In a previous work (Ferraro et al., 2003) a set of fuzzy logic- using a linear membership function is often of the same quality as the solution
based indicators were developed to evaluate the effects of pesti- obtained using a complicated non-linear membership function (Sahinidis, 2004).
cides and tillage on soil condition. In this paper, both indicators Monotonic (such as linear) transitions from one state to another show continuous
changes of the variable under study, from a minimum to a maximum level. Soil
were modified and integrated with other soil indicators in order to
nitrogen balance was an exception to this rule because it uses a triangular
develop a more comprehensive model for assessing soil condition membership function that defines the full membership values of fuzzy Max subset
in agroecosystems also based on a fuzzy set approach (Zadeh, within the transition range. These values show an increase on one end of the range
1965). It is considered a goal-oriented model, as it is based on the and a decrease on the other end (Table 1). Primary and secondary module values are
combined through rules elicited from the expert panel for calculating soil condition
formulation of a desired, sustainable state and distinct aims which
indicators (Table 2). The first secondary module (i.e. SM1) assesses both the
are characterized by a set of selected indicators (von Wirén-Lehr, management and the environmental effects on the decomposition rate of residue
2001). In order to show the application of the model in a detailed biomass. Then, this module is incorporated into the second one (SM2) in order to

Table 1
Input (IV) and membership function variables (MF variable) for calculating the primary modules (PM)

Input variable (IV) MF variable Primary module (PM) Fuzzy subset PM(1) Source

Min Max
*
Crop C:N ratio Substrate 25 200 Min Guerif et al., 2001
decomposition (1)
*
Mean daily precipitation Water availability (2) 0 4 Max Stroo et al., 1989; Schreiber, 1985;
during fallow [mm] Savabi and Stott, 1995
STUa (stubble incorporation) *
Crop residue 0 2 Max Douglas and Rickman, 1992; Karlen
incorporation (3) et al., 1994; Guérif et al., 2001
*
Mean air temperature Temperature (4) 4 32 Max Stroo et al., 1989; Steiner et al., 1999
during fallow [ C]
*
(1)  (2)  (3)  (4) Abiotic 0 1 Max Thorburn et al., 2001
decomposition
*
Insect toxicity units [TUi] Biotic 0 1.90 Min Ferraro et al., 2003
decomposition
Crop yield [t/ha]l Crop residue mass [t/ha]/annual net primary HANPP (Harvest 0 1 Min Haberl and Geissler, 2000
Mean annual production (ANPP) of potential vegetation [t/ha] of ANPP)
precipitation [mm]
Fertilization rate [kg/ha] Nitrogen output [kg/ha]/nitrogen input [kg/ha] Soil nitrogen 0–2 1 Max Raun and Johnson, 1999; Neeteson,
Crop yield [t/ha] balance 1990
*
SOM content [%] Soil fertility 0 4 Max
Tool impactb *
Tillage impact 0.18 1.84 Max Ferraro et al., 2003
K factorc *
Soil erodibility 0 0.75 Max Renard et al., 1997

Min and Max value for each MF variable define the linear interval [0,1] of the membership function. PM(1) indicates which one of the fuzzy subsets corresponds to membership
value ¼ 1 (i.e. PM ¼ 1). Asterisk indicates the same variable for both IV and MF variable.
a
STU value represents the relative magnitude of crop residue incorporation after tillage (i.e. one moldboard plow/year: STU ¼ 1, for intermediate values see Table A1).
b
Tool impact value represents the cumulative effect of tillage on both soil structure and stubble incorporation (i.e. one moldboard plow/year: Tool Impact ¼ 1.84, for
intermediate values see Table A1).
c
K factor [t/h ha1 MJ mm1] is the soil erodability factor form the Revised Universal Soil Loss Equation (RUSLE) (Renard et al., 1997) with larger values corresponding to
higher erosion risk.
D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370 361

Overall soil deradation (SM6)

Soil physical degradation (SM5) Soil chemical degradation (SM4)

Tillage impact (Timp) Soil nitrogen depletion (SM3)

Tool impact Soil Nitrogen balance

Soil erodibility Nitrogen input (kg/ha)


Crop yield (kg/ha)
K Factor
Soil fertility

SOM content (%)

Soil carbon depletion (SM2)

HANPP

Mean annual precipitation (mm)


Crop yield (kg/ha)

ANPP decomposition (SM1)

Abiotic decomposition

Crop C:N ratio


Mean daily precipitation during fallow [mm]
Stubble incorporation (STU)
Mean daily temperature during fallow [°C]

Biotic decomposition

Insect toxicity units [TUi]

Fig. 1. Diagram of the knowledge-based system for soil condition assessment. Secondary and primary modules are shown in solid lines and input variables in dotted lines.

assess the status of carbon stock at field scale. The calculations from this module main crops’ species were used during the cropping period in the selected fields:
have the objective of estimating changes in the remaining system energy (i.e. maize (Zea mays L.), and wheat (Triticum aestivum L.)/soybean (Glycine max L.)
biomass) by combining the resulting conditions from SM1 and the human appro- succession. A sequence of continuous cropping using these species is maintained for
priation of aboveground net primary production (HANPP) (Haberl and Geissler, a number of years, ending in a sowing pasture, and returning to row crop rotation.
2000). Subsequently, the secondary module (SM3) is created to estimate the
depletion of soil nitrogen stock. Soil chemical degradation module (SM4) evaluates Table 2
the partial effects of crop management and the environmental conditions on soil Results from the elicitation process of the conclusion rules from each secondary
carbon and nitrogen stocks (Fig. 1). Data about tillage impact and soil erodibility module
were combined by means of the elicited fuzzy if-then rules for estimating soil
Secondary Antecedents Rule Conclusion rules
physical degradation (SM5). Finally, an overall soil degradation assessment (SM6)
module
was obtained by combining chemical (SM4) and physical (SM5) assessments. IF AND THEN
Technical details of knowledge elicitation process, the fuzzy set theory, and the
A1 A2 A1 A2 Starting Weighted-
building of fuzzy sets and rules are presented in Technical appendix (Sections A.1–
value average score
A.3).
(WAS)
2.2. Model evaluation ANPP Abiotic Biotic 1 1 1 1.00
decomposition decomposition decomposition 1 0 0.8 0.72
A sensitivity analysis was performed to assess the relative importance of input (SM1) 0 1 0.2 0.20
variable variation in secondary modules outcome. The sensitivity of an independent 0 0 0 0.00
variable x with respect to the dependent variable y is quantified by the sensitivity
Soil carbon SM1 HANPP 1 1 0 0.14
index I (Lenhart et al., 2002).
depletion 1 0 0.6 0.56
  (SM2) 0 1 0.4 0.42
ðy  y1 Þ=y0 
I ¼  2   100 (1) 0 0 1 0.94
ð2Dx=x0 Þ 

with y0 the value of the dependent variable y using the initial variable x0. The value Soil nitrogen Soil nitrogen Soil fertility 1 1 1 1.00
of this initial variable changes by Dx yielding x1 ¼ x0Dx and x2 ¼ x0 þ Dx with depletion balance 1 0 0.6 0.65
corresponding values y1 and y0; Dx value was 25% of the initial value x0 of the (SM3) 0 1 0.3 0.20
respective independent variable x. Since only one independent variable was modi- 0 0 0 0.00
fied at a time the interactions among independent variables are not taken into Soil chemical SM3 SM2 1 1 1 1.00
account. The indexes were ranked into four classes to compare the calculated degradation 1 0 0.35 0.31
sensitivity: low sensitivity (0 ¼ I < 5), medium sensitivity (5 ¼ I < 20), moderately (SM4) 0 1 0.65 0.71
high sensitivity (20 ¼ I < 60), and high sensitivity (60 ¼ I < 100). 0 0 0 0.00

2.3. Site analyzed Soil physical Tillage impact Soil erodibility 1 1 0 0.00
degradation 1 0 0.3 0.30
The study was conducted in the Inland Pampa (Argentina) (34–35 S; 61–63 W). (SM5) 0 1 0.7 0.70
This region comprises a land area of about 4.5 million ha in the center of Argentina. 0 0 1 1.00
The most frequently cropped soils in the region are Mollisols which developed from
Overall soil SM5 SM4 1 1 1 1.00
eolian sediments of the Pleistocene era, with dominantly udic and thermic water
degradation 1 0 0.65 0.65
and temperature regimes, respectively (Moscatelli et al., 1980). The average annual
(SM6) 0 1 0.55 0.55
rainfall decreases from about 900 mm in the east to 750 mm in the west part of the
0 0 0 0.00
study area (Soriano et al., 1991). The Inland Pampa is a sub-region of a fertile plain
covered by grasslands, which during the 1900s and 2000s was transformed into an Higher values within the interval [0,1] for the conclusion rule scores indicate better
agricultural land mosaic by grazing and farming activities (Hall et al., 1992). Three scenarios in terms of the sustainability assessment.
362 D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370

Different tillage systems (from no-tillage to conventional tillage) were used in the the WAS values for the intermediate rules were 0.65 and 0.55 (Table
crop fields, and different combinations of tillage tools were utilized as well (e.g. 2). The expert panel denoted an intermediate knowledge level
moldboard, chisel and disk ploughing; disk and spike-tooth harrows; row cultiva-
tors; and seed drills). At the farm level, the study was conducted by using data from
regarding the model components. Average values of knowledge
106 farms for maize, wheat and soybean provided by the AACREA (Argentine quality (Kq) elicited ranged from levels 4–6 of the knowledge level
Association of Agricultural Experimentation Consortia) farmers association. Vari- scale of 0–0 (Fig. 2). The effect of both tillage impact and soil
ables in the database were (1) yield per field; (2) pesticide dose applied in each field erodibility on soil physical degradation (i.e. SM5) obtained the
(herbicide, insecticides, and fungicides); (3) type and number of tillage operations;
lower score in knowledge quality from the expert panel. The
(4) monthly weather variables: air temperature and precipitation; (5) fertilization
rate; and (6) soil texture and organic matter content from soil series description coefficient of variation (CV) of the conclusion rule scores obtained
following Soil Taxonomy (Soil-Survey-Staff, 1999). The final data set consisted of 360 in each secondary module evaluates the level of disagreement
maize, 274 soybean and 483 wheat fields. among the expert panel members (Fig. 2b). The level of disagree-
ment ranged from 0 (SM5 and SM6) to 55% (SM2: soil carbon
depletion module).
3. Results

3.1. Knowledge elicited 3.2. Model testing

A total number of 24 if-then fuzzy rules for secondary module Model performance was evaluated through sensitivity analyses
calculations were elicited from the expert panel and they represent using the sensitivity index I with values ranging from low (0 ¼ I < 5)
the consensus reached through the applied aggregation method to high (60 ¼ I < 100) (Table 3). Baseline scenario (Table 4) for
(Table 2). The first secondary module (SM1) comprises a set of sensitivity calculations represents the mid-range value delimited
conclusion rules that show a higher effect of abiotic over biotic by the Min and Max fuzzy subset values (Table 1) of each input
decomposition in the overall biomass decomposition model. This variable. Mean daily temperature during fallow was the most
effect could be detected after inspecting the weighted-average important input variable for defining final SM1 value with I values
value (WAS) for the intermediate rules (see Appendix for WAS ranging from 72 to 78 for the three crops analyzed (Table 3). The
calculation). When abiotic decomposition reached its maximum others’ variables showed I values ranging from 20 to 46. All input
value and the biotic value reached its minimum, the WAS value for variables related to SM1 calculations showed I values lower than 10,
the conclusion rule was 0.72 (i.e. 72% of the maximum value for when calculating SM2. This module was mainly affected by crop
SM1). On the contrary, when abiotic decomposition reached the yield (CY) and mean annual precipitation (MAP) (Table 3). However,
lowest value and biotic decomposition reached its highest value, sensitivity analyses of SM2 calculation revealed that the model
the WAS value for this rule was 0.20 (i.e. 20% of the highest ANPP structure showed differences among crops. For example, SM2 did
decomposition scenario). Following the same pattern of analysis, not change due to CY and MAP changes when maize crop scenario
Table 2 reveals a higher influence of soil nitrogen balance than soil was analyzed. Oppositely, CY and MAP changes resulted in
fertility on soil nitrogen depletion module (SM3). When assessing moderately high sensitivity I values (20 ¼ I < 60) when soybean
soil chemical depletion (SM4), the expert panel gave more impor- and wheat were analyzed. This variability among crops is explained
tance to soil carbon than soil nitrogen depletion. This standard can by the specific differences in the crop residue biomass after harvest,
be envisaged in the two intermediate rules for SM4 calculations: and the resulting influence on the HANPP calculation (Table 1).
when SM3 ¼ 1 and SM2 ¼ 0 (i.e. the best scenario for soil nitrogen SOM content showed a range of I values from 14 to 20 associated
balance and the worst one for soil carbon depletion) the WAS value with soil nitrogen depletion module (SM3) calculation. After
for SM4 is closer to 0 (i.e. the worst condition) than when it examining the effect of input variables effect on soil chemical
happens the opposite condition (i.e. if SM3 ¼ 0 and SM2 ¼ 1; then degradation module (SM4), it was also possible to detect differ-
SM4 ¼ 0.71; Table 2). An intermediate conclusion rule for soil ences among crops. As a result of MF parameters and the elicited
physical degradation (SM5) reached a WAS value of 0.3 (i.e. 30% of fuzzy if-then rules, maize crop showed low and marginally medium
the most desirable scenario for soil physical condition) when tillage sensitivity values for all the input variables considered. Soybean
impact is maximum and no erodibility conditions exist. Conversely, and wheat showed I values of 42 and 25, respectively, in SM4
when soil erodibility reaches its highest value and there is no calculation when MAP changes were considered. Sensitivity anal-
noticeable tillage impact effect on soil condition the SM5 value is ysis also showed that the soil physical degradation (SM5) module
0.7 (i.e. 70% of most desirable scenario for soil physical condition). was more affected by the tillage impact than by the erosion risk
Finally, the elicited knowledge from the expert panel showed that itself (i.e. K factor). Sensitivity analysis was able to detect the
the physical and chemical components of the overall soil degra- relative influence of all input variables on the overall soil degra-
dation module (SM6) showed similar effects on soil indicators, as dation (SM6) module. Crop yield (CY) changes resulted in

ALL
Secondary module

SM6
SM5
SM4
SM3
SM2
SM1 a b
0 1 2 3 4 5 6 7 8 9 10 0 10 20 30 40 50 60 70 80 90 100
Knowledge level [0-10] Disagreement between experts (CV)

Fig. 2. Knowledge quality (Kq) (a) and disagreement level (b) of the expert panel. Disagreement between experts is the coefficient of variation of the conclusion rule values
elicitated (0 ¼ maximum agreement; 100 ¼ maximum disagreement). Symbols denote the average of the four conclusion rules for each secondary module and for all together (ALL).
Ranges show the lowest and highest values that were elicited.
D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370 363

Table 3 matter (SOC) content in wheat crop was also significantly higher
Results of sensitivity index I calculations for each secondary module than in soybean and maize fields. The erosion risk, denoted by K
Secondary module Crop Base Input variables factor, showed a slightly higher value in maize field than in the
1 2 3 4 5 6 7 8 9 10 11
other two crops. Results for the environmental primary modules
* * * * * * showed the most pronounced difference among crops in the biotic
SM1 M 0.17 33 41 41 73 35
S 0.22 20 46 46 79 39 * * * * * * decomposition module (Table 5). Soybean exhibited the worst
W 0.23 30 47 47 74 37 * * * * * * condition for biomass decomposition mediated by the soil biolog-
SM2 M 0.88 1 3 3 5 4 0 0 * * *
*
ical components, while maize showed an intermediate and wheat
S 0.76 2 4 0 7 3 57 54 * * * * the highest (i.e. most desirable) value. The abiotic decomposition
W 0.85 3 5 5 8 4 32 29 * * * *
conditions, mediated by climate and stubble soil incorporation due
SM3 M 0.42 * * * * *
44 *
43 14 * * to tillage operations, were far from the full membership of the
S 0.62 * * * * *
51 *
37 20 * * maximum fuzzy subset in all the three crops analyzed (Table 5).
* * * * * * * *
W 0.54 86 71 20 Consequently, none of the crops analyzed showed values higher
SM4 M 0.70 2 3 3 6 4 6 0 6 2 * * than 0.09. As all the membership functions are linear, those values
S 0.65 1 3 3 6 2 11 42 7 4 * * represent a biomass decomposition rate mediated by the abiotic
* *
W 0.71 3 5 5 8 4 14 26 12 3 environment that is equivalent to 10% of the maximum decompo-
SM5 M 0.50 * * * * * * * * *
40 17 sition scenario in all the crops analyzed. Soil fertility and erodibility
S 0.50 * * * * * * * * *
40 17 showed values that ranged from 0.41 to 0.50 and they followed the
* * * * * * * * *
W 0.50 40 17 same pattern as SOM content and K factor respectively. Regarding
SM6 M 0.61 1 1 1 2 2 3 0 2 1 9 4 the variability within the field crops analyzed, the highest CV values
S 0.59 0 1 1 2 1 12 15 2 1 9 4 corresponded to abiotic decomposition (average CV ¼ 154.1%) and
W 0.61 1 2 2 3 2 5 9 5 1 9 5 the lowest to soil erodibility (average CV ¼ 17.6%).
Input variables are: (1) Crop C:N ratio; (2) Mean daily precipitation during fallow
[mm]; (3) Stubble incorporation (STU); (4) mean daily temperature during fallow 3.3.2. Management input variables and primary modules
[ C]; (5) insect toxicity units [TUi]; (6) crop yield [t/ha]; (7) mean annual precipi-
Management input variables showed a higher variability among
tation [mm]; [8] fertilization rate [kg/ha]; (9) SOM content [%]; (10) tool impact; (11)
K factor. Base column indicates the secondary module value in the base scenario of crops than the environmental variables (Table 6). Regarding the
Table 4. Asterisk indicates that input variable is not used for the secondary module effects of tillage operations, soybean showed the lowest value in
calculation. stubble soil incorporation (i.e. STU). Soybean also showed the lowest
Secondary modules: SM1 (ANPP decomposition); SM2 (Soil carbon depletion); SM3 tillage effect on the soil physical attributes being evaluated (i.e.
(soil nitrogen depletion); SM4 (soil chemical degradation); SM5 (soil physical
degradation); SM6 (overall soil degradation).
lowest value of tool impact). No significant differences in these
Crops: M (maize); S (soybean); W (wheat). tillage-related variables were observed between maize and wheat.
Sensitivity index I values: low (0 ¼ I < 5), medium (5 ¼ I < 20), moderately high Input variables related to nitrogen management showed the highest
(20 ¼ I < 60), and high (60 ¼ I < 100) in bold. value for nitrogen input (i.e. fertilization rate) and output (i.e. soil
nitrogen balance) in soybean fields, while maize and wheat fields
showed intermediate and the lowest values for both nitrogen vari-
a sensitivity I value of 12.2 for soybean. Mean annual precipitation ables, respectively (Table 6). Pesticide toxicity caused by crop
(MAP) showed medium I values of 15 and 9.5 for soybean and protection products was significantly higher in soybean, and in turn
wheat, respectively (Table 3). Finally, tool impact had also medium affected both mammals and insects as well (Table 6). Crop yield
sensitivity (I ¼ 9) when analyzing final SM6 values for the three showed no difference between wheat and soybean, both repre-
crops under analysis. senting roughly 38% of the maize crop yield (Table 6). These differ-
ences in crop yield are related to the other two secondary modules,
3.3. Model application as maize showed the lowest values for human appropriation of
aboveground net primary production (HANPP) and soil nitrogen
3.3.1. Environmental input variables and primary modules balance (Table 6). Finally, tillage impact assessment showed the
Results from the database showed that wheat crop experienced highest value for maize and wheat, reaching almost 30% of the most
different environmental conditions than the other two crops desirable scenario determined by the MF functions of Table 1.
analyzed (Table 5). Both daily precipitation and air temperature
during the fallow period were significantly higher for wheat than 3.3.3. Secondary modules
for soybean and maize (Kruskal–Wallis test, p < 0.05). Soil organic Primary module combination through the elicited fuzzy if-then
rules determined final values for each one of the secondary
modules (Table 2, Fig. 3). A higher value for each of these indicators,
Table 4
in the 0–1 range, represents higher membership grade to the
Input variables allowed varying in the sensitivity analysis and their respective values
for the baseline scenario sustainability criteria elicited from the expert panel. Soybean crop
showed the lowest value for aboveground net primary production
Input variable Baseline value
(ANPP) decomposition (SM1 ¼ 0.19). Regarding the partial indica-
Crop C:N ratio 70.0 (maize), 40 (soybean), tors of soil chemical condition, soybean and wheat exhibited the
65 (wheat)
Mean daily precipitation during fallow [mm] 2
lowest value for the soil carbon status indicator (SM2 ¼ 0.71 and
STU (stubble incorporation) 1 0.74, respectively). Soybean crop exhibited the best condition
Mean air temperature during fallow [ C] 18 related to soil nitrogen depletion (SM3 ¼ 0.45). When these two
Insect toxicity units [TUi] 0.95 indicators were combined through the soil chemical indicator
Crop yield [t/ha] 8.0 (maize), 3.0 (soybean and
(SM4), soybean and wheat resulted in lower values than maize. The
wheat)
Mean annual precipitation [mm] 800 assessment of soil physical degradation (SM5) showed statistically
Fertilization rate [kg/ha] 50 significant differences among crops, ranging from 0.81 for maize to
SOM content [%] 2 0.87 for soybean (Fig. 3). Overall soil degradation (SM6), which was
Tool impact 1.01 calculated by integrating both chemical and physical degradation
K factor 0.375
assessments, showed no statistically significant difference between
364 D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370

Table 5
Environmental input variables (IV) and primary modules (PM) results from the crop fields analyzed

Maize Soybean Wheat

Mean Min Max Cv [%] Mean Min Max Cv [%] Mean Min Max Cv [%]
Input Mean daily precipitation during fallow [mm] 2.48a 1.25 5.45 46.33 2.64a 1.25 5.45 48.9 2.87b 1.24 5.45 48.71
variables Mean air temperature during fallow [ C] 14.65a 13.73 18.98 6.47 14.21a 13.14 15.51 4.49 20.28b 17.78 21.63 3.47
Mean annual precipitation [mm] 981a 796 1709 29.7 1022a 797 1710 32.97 1071a 797 1710 33.32
SOM content [%] 1.67a 1.30 4.00 49.18 1.69a 1.30 4.00 50.67 1.93b 1.30 4.00 55.56
K factor 0.28b 0.19 0.51 41.09 0.26a 0.19 0.51 42.51 0.27a 0.16 0.51 42.74

Primary Abiotic decomposition 0.02a 0.00 0.15 136.2 0.04a 0.00 0.73 198.18 0.09b 0.01 0.45 127.81
modules Biotic decomposition 0.82b 0.00 1.00 29.78 0.65a 0.00 1.00 58.30 0.95c 0.00 1.00 13.88
Soil fertility 0.41a 0.32 1 49.19 0.42a 0.32 1 50.67 0.48b 0.32 1 55.56
Soil erodibility 0.50b 0.31 0.57 17.95 0.48a 0.31 0.57 16.01 0.49a 0.31 0.70 18.76

Primary module values range from 0 to 1 and denote the membership grade to the fuzzy subsets Max of Table 1.
Means in the same row followed by the same letter are not significantly different (multiple comparisons after Kruskal–Wallis test, p < 0.05).

all the three crops analyzed (SM6 ¼ 0.64 for maize and wheat and sustainability-based decisions (i.e. ex-ante), but they could also
SM6 ¼ 0.67 for soybean). Thus, all three values are showing have a role in post-monitoring of environmental policy impact
a similar degree of membership to the sustainable condition (Quinn and Hall, 1996).
elicited and represented through the model. Model validation constitutes another very important step in the
modeling process (Yaman, 1996). Validity of a model is always
4. Discussion provisional and based on the availability of field data and knowl-
edge of the real system against which the model can be tested
A successful soil condition assessment in agroecosystems (Nguyen and de Kok, 2007). In a knowledge-based system this
involves advances in understanding the multiple interactions performance depends on the quality of the scientific and the expert
between human and natural systems. This complexity implies knowledge elicited. Knowledge of soil biodiversity and its specific
a number of modeling constraints relating to scaling mismatches, contribution to ecosystem function are today limited (Barrios,
synthesis of non-homogeneous information, uncertainty in causal 2007). In this model, the biological effect on the decomposition
relationships, assessment of trade-offs, and validation (Bellamy and model is affected by the potential consequence of pesticides on the
Lowes, 1999). The knowledge-based system developed in this work soil fauna. Research on soil quality assessment regarding chemical
attempts to deal with these constraints by identifying the highest and physical properties is much more developed but the identifi-
consensus among experts in order to gather relevant knowledge for cation of key indicators and their threshold values are still a matter
designing sustainable soil management practices. Although it is of discussion (Arshad and Martin, 2002). Likewise, the elicited
recognized that this approach is highly sensitive to the quality of knowledge from the expert panel also showed variability in both
the acquired expert knowledge, it is also accepted that a fuzzy quality level and final score for the conclusion rules. This is
expert system offers great flexibility (Metternicht and Gonzalez, a common problem when eliciting expert opinion because experts
2005). Hence, the KBS developed here could be easily adapted to provide their assessments by considering all relevant information
different conditions as experts can re-define fuzzy rules to adapt available as well as their own experiences. For example, soil carbon
them to local conditions or include new ones for additional issues depletion was the most controversial issue when assessing soil
(e.g. soil phosphorus status or soil compaction). condition. A possible explanation for these differences of opinions
Measurability and policy relevance are identified in several could be the differential perceptions about the importance of the
quality criteria for ecosystem indicators (Yli-Viikari et al., 2007). intermediate components of this indicator (i.e. crop residue
Despite the fact that assessments based on impact criteria usually decomposition and crop residue harvest from the system). Clearly,
are constrained by data limitations (Levitan et al., 1995), the the search of guidelines for long-term assessment of soil condition
system developed in this work uses simple and easily available resulted in inaccurate and uncertain knowledge. Therefore, the
data. On-farm tillage and fertilizer operations are usually regis- current approach is one of the best ways to clearly show the
tered, and climate and soil data are available for all targeted end- strengths and weaknesses of a model to its intended users, as
users, from many sources. Moreover, the knowledge base elicited a method of validation (Beck and Chen, 2000). Anyhow, final users
from the expert panel proposes concrete guidelines for policy- are able to detect the knowledge areas where more research should
related work (Maud et al., 2001) since these indicators not be conducted in order to validate the elicited knowledge regarding
only have a role at the farm level, enabling farmers to make long-term soil sustainability.

Table 6
Management input variables (IV) and primary modules (PM) results from the crop fields analyzed

Maize Soybean Wheat

Mean Min Max Cv [%] Mean Min Max Cv [%] Mean Min Max Cv [%]
Input variables STU (stubble incorporation) 0.75b 0.00 4.90 132.6 0.54a 0.10 6.30 175.5 0.70b 0.10 4.10 134.9
Insect toxicity units [TUi] 0.63b 0.00 17.13 311.0 1.65c 0.01 20.72 177.2 0.11a 0.00 3.85 314.0
Crop yield [t/ha] 8.42b 2.47 14.2 27.81 3.08a 0.88 5.0 27.21 3.30a 1.0 6.21 29.13
Fertilization rate [kg/ha] 26.4b 0.00 110.4 85.06 85.19c 24.37 137.5 27.20 19.71a 0.00 92.00 110.6
Tool impact 0.79b 0.00 4.38 112.4 0.60a 0.18 5.38 145.9 0.72b 0.18 3.78 118.6

Primary modules HANPP 0.04a 0.00 0.40 10.05 0.27b 0.00 0.88 30.32 0.16b 0.00 0.75 24.87
Soil nitrogen balance 0.16a 0.00 1.05 91.71 0.50b 0.00 1.01 76.17 0.76b 0.00 1.21 73.17
Tillage impact 0.31b 0.00 1.00 59.11 0.19a 0.00 1.00 45.17 0.27ab 0.00 1.00 52.92

Primary module values range from 0 to 1 and denote the membership grade to the fuzzy subsets Max of Table 1.
Means in the same row followed by the same letter are not significantly different (multiple comparisons after Kruskal–Wallis test, p < 0.05).
D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370 365

SM1 SM2
1.00 1.00

a
0.75 0.75
c
0.50 0.50 b
b
a a
0.25 0.25

0.00 0.00
Maize Soybean Wheat
Maize Soybean Wheat

SM3 SM4
1.00 1.00
b a
a
0.75 0.75

0.50 0.50

b a
0.25 a 0.25

0.00 0.00
Maize Soybean Wheat Maize Soybean Wheat

SM5 SM6
1.00 1.00
a c
a a a
0.75 b 0.75

0.50 0.50

0.25 0.25

0.00 0.00
Maize Soybean Wheat Maize Soybean Wheat

Fig. 3. Distribution of the values for the secondary modules 1–6 resulting from different crops analyzed. The vertical line shows the range from minimum to maximum value, the
box contains 50% of the values, excluding the lowest and the highest 25%. The horizontal line inside the box shows the median value. Secondary modules are SM1 ¼ ANNP
decomposition, SM2 ¼ soil carbon depletion; SM3 ¼ soil nitrogen depletion; SM4 ¼ soil chemical degradation; SM5 ¼ soil physical degradation; and SM6 ¼ overall soil degradation.
Crops with the same letter are not significantly different (multiple comparisons after Kruskal–Wallis test, p < 0.05).

Sustainability criteria for agroecosystems management and the appeared as the most susceptible component in the crop fields
development of an unbiased way to test them are two underde- analyzed. SM3 values were never higher than 0.5 which indicates
veloped research areas in Argentina. In the study area, agricultural that no crop was able to accomplish more than 50% of the
soils have been intensively used for conventional crop production sustainability criteria for soil nitrogen extraction from the system.
(Maddonni et al., 1999; Martı́nez-Ghersa et al., 2000; de la Fuente The intensification process in the study area entailed the adoption
et al., 2006) in the last two decades. However, there is neither of conservation tillage systems (Fabrizzi et al., 2005). These
enough knowledge on environmental issues, nor clear perception soil conservation systems (e.g. no-tillage) reduced continuous soil
of environmental alterations across space and time due to inten- cultivation, thus preventing the deterioration of physical soil
sification (Viglizzo et al., 2003). Particularly, soil condition assess- properties (Arshad, 1999; Ferreras et al., 2000). However, results
ments in the study area are frequently restricted to system tillage from KBS application indicated that the adoption of conservation
effect evaluation on a limited set of soil properties (Alvarez et al., tillage systems proved useful to reduce the risk of soil erosion, but
2001; Aon et al., 2001; Fabrizzi et al., 2005; Giuffré et al., 2006; nutrient deficiency problems could still be present. Nutrient
Steinbach and Alvarez, 2006). With the KBS developed here, it was depletion in soils adversely affects soil quality and reduces crop
possible to obtain not only a set of quantitative indicators of yield and, consequently, poses a potential threat to global food
different soil components but also a more comprehensive risk security and agricultural sustainability. In developing countries, as
assessment using both environmental and crop management a result of unbalanced fertilization with surplus N, nitrate can be an
aspects for assessing the potential environmental impact of three important environmental contaminant due to its high potential for
different crop production scenarios. In general, there were no leaking into surface or ground waters. However, results from KBS
significant differences when assessing the potential impact on soil application in agricultural systems of Inland Pampa did not reveal
condition among crops. However, the intermediate indicators did a significant risk of groundwater contamination due to over-
show some heterogeneity among crops. Results from the KBS fertilization. Moreover, it seems that the more important problem
application indicated that the soil physical condition (i.e. SM5) related to the nitrogen use in the cropping systems analyzed is the
showed values higher than 0.80 in all crops analyzed, which means risk of avoiding adequate soil fertility replenishment.
that the potential impact caused by tillage intensity and erodibility Further long-term scientific research linking crop management
conditions was relatively low. However, soil nitrogen condition and soil functionality could substantially improve the overall
366 D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370

performance of KBS like the one laid out in this work. Meanwhile, are from Argentina, but they also take part of several international
the assessment of trade-offs between different ecosystem functions scientific projects. The structured questionnaire contained a start-
through expert opinion can complement traditional analyses for ing value as a conclusion for each fuzzy rule of the secondary
detecting a sustainable trajectory based on the measurement of soil modules (Table 2). Firstly, all the experts were asked to consider, on
properties. Moreover, the use of the KBS developed in this work as their own, and in a numeric way, each one of the conclusions, using
priors for ecological models could be a useful cost-effective the [0,1] interval. This numeric interval reflects the magnitude of
assessment tool for making confident predictions about the effect each scenario (i.e. rule) for achieving a sustainable trajectory of the
of crop management on soil condition. system. Also, as the various experts may specialize in different areas
of the sustainability assessments, it is necessary to define the
5. Conclusion knowledge quality (Kq) of the expert panel. Therefore, they were
also asked to rate their own knowledge of each one of the
The knowledge-based system developed in this work proposed secondary module assessment topics using a 0–10 scale. Finally, the
an operational framework for soil condition assessment when experts’ assessments were combined in a single weighted aggre-
knowledge is inaccurate and uncertain. It combines scientific and gated assessment. Mathematically, the weighted-average score of
expert knowledge. Primary modules are built based on available the fuzzy rule conclusion can be calculated as follows (Goossens
scientific literature and secondary modules rely on the elicited et al., 1998; Van Der Fels-Klerx et al., 2002):
knowledge about soil condition from an expert panel. Soil condi- X
tion can be broadly defined as the capacity of a living soil to Weighted-average scoreðWASÞ ¼ WeEAe (A1)
function (Doran et al., 1996), and the assessment of management e

and environmental effects on soil condition should be considered


as the primary indicators of sustainable management (Doran,
2002). The analytical framework used in this work is able to Ke
Weighting factorðWeÞ ¼ P (A2)
incorporate not only management and environmental effects but Ke
also other impact aspects (i.e. economic, social) and it may increase e

the strength for assessing the impact of agricultural practices on the


whole system sustainability. Also, the fuzzy logic approach used in X
this work is very useful to develop a continuous process of e ¼ 1; 2; .; E with We ¼ 1 and We  0
ecosystem monitoring by the inclusion of new rules for weighting e
indicators’ scores in different situations. Further work must define
where, WAS: weighted combination of EA1, EA2,., EAE; Ke [e ¼ 1,
scaling procedures for applying the indicators developed at the
2,., E]: the knowledge quality of the experts 1, 2, ., E for assessing
field and farm level to the regional level, in terms of classification of
a fuzzy if-then rule conclusion; EAe [e ¼ 1, 2,.,E]: the expert
farms, vulnerability of the environment, or fate of pollutants,
assessment of the experts 1, 2, .,E for a fuzzy if-then rule
according to the available data at the regional scale.
conclusion.
The simple method adopted in this paper assumes that experts
Acknowledgements
are unbiased and consistent (i.e. calibration is considered unnec-
essary). Knowledge representation of the KBS was implemented
This material is based upon work supported by the University of
through the NetWeaver logic engine (Reynolds, 1999). NetWeaver
Buenos Aires (UBA); the National Council for Scientific Research
knowledge-based system is composed of an engine and a graphic
(CONICET); the National Agency for Science Promotion (ANPCyT) of
user interface for knowledge-based developers, which provides
Argentina; and the U.S. National Science Foundation under Bio-
controls for designing, editing and interactively evaluating knowl-
complexity in the Environment Grant No. 0410348. D.O. Ferraro
edge bases. By means of the final step, it is possible to implement
was supported by a post-doctoral fellowship from CONICET.
the conceptual model in a definite computerized system.

A.. Technical appendix

A.1. Knowledge acquisition and representation A.2. Knowledge-based system development

Knowledge acquisition for KBS building was obtained from both Primary modules use fuzzy logic to normalize the quantitative
knowledge elicitation from an expert panel and available scientific information from data variables (i.e. input variables) in order to
data. For knowledge elicitation, structured questionnaires were obtain a value that expresses the effect of these variables on the soil
used to collect information with regard to potential effects of condition assessment. This value is calculated through mathemat-
agricultural management on soil condition. It has been suggested in ical functions called membership functions (MF). A membership
the literature that it is not necessary to have a large number of function fA(x) defines the fuzzy nature of the indicators that
experts because of diminishing marginal returns associated with describe the degree to which an event occurs, but not whether it
a big group of experts, and the suggestion is that the biggest benefit occurs. In all cases, except for soil nitrogen balance where a trian-
is reached with the first three to four experts (Clemen and Winkler, gular MF was used, linear MFs were used since the input value that
1999; Winkler and Clemen, 2004). Hence, the objective was to determines the full membership (i.e. fA(x) ¼ 1) of the primary
select experts with different opinions rather than to seek a large module was a maximum (Max) or a minimum (Min) (Table 1). In
number of individuals with the same/similar opinions (Soll, 1999). both cases the shape of each MF was defined by the Max and Min
Finally, six expert researchers from universities contributed with values that determine the domain of the input variable for the
their scientific knowledge as well as with their expertise in lineal interval of the membership m(x) [0,1]. Secondary modules
understanding agricultural farming systems. The experts’ back- were based on the elicited knowledge from the expert panel. They
grounds varied from practice to research, over half of them had are represented by fuzzy rules of the form IF (antecedent)-THEN
recent experience in both these fields. Most of them had specialized (consequent). These rules organized the modules in a hierarchical
in soil science but also had knowledge on crop and landscape way up to the overall soil condition assessment. One fuzzy rule
ecology, as well as some normative expertise. All expert affiliations consists of K arguments in the form of fuzzy sets Ai,j with
D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370 367

membership functions mAj and one response variable r with decomposition based on the initial C:N ratio of the different resi-
a numerical conclusion Bi: dues of the crop rotation.
Water availability: Soil water availability is also considered when
IF ðx1 is Ai; 1Þ AND ðx2 is Ai; 1Þ AND ðxK is Ai; KÞ THEN r is Bi (A3) estimating residue decomposition (Parton et al., 1987). As well as
where, Ai,j are the fuzzy subsets (e.g. acceptable or unacceptable) temperature effect, soil water effect is assessed through
corresponding to a partitioned domain of the input variable xj (j ¼ 1, a membership function that demonstrates the relative rate of
., K) and r is output variable. In addition, the IF-THEN rules include microbial activity with increasing water availability. There are
the logical connective AND among the antecedents, representing various forms of functions used to describe the influence of water
the intersection of both fuzzy sets through the min-operator: content on decomposition. The simplest way to describe this
influence is a linear relationship in the soil moisture range
wBiðx1; x2; xKÞ ¼ min ðmAi; 1; mAi; 2; mAi; KÞ (A4) considered. Mean daily precipitation during the fallow period was
used as an indirect variable for assessing soil water availability for
where, wBi (x1;x2;xK) is the final membership value of the i rule. residue decomposition soil moisture estimator (Steiner et al., 1999;
This final value depends on the minimal value of the antecedents, Quemada, 2005).
and represents a precautionary view when assessing ecosystem Crop residue incorporation: Crop residue decomposition may
integrity. Conventionally, the conclusion of each one of the rules vary according to the surface residues that are incorporated, or
has also linguistic values (e.g. acceptable, unacceptable). Never- removed, into the soil using the tillage action. The incorporation
theless, the logical inference allows the replacing of these conclu- primary module assesses the effect of the residue incorporation on
sions by numerical values (c) in the interval [0,1] (Zadeh, 1965). the biomass decomposition process. The input variable STU (Fer-
Consequently, in all fuzzy rules used, the numerical conclusion is raro et al., 2003) is a synthetic indicator of the effect of different
graded between 0 and 1. Then, fuzzy rules are combined in each tillage operations on the retention of crop stubble on the soil
module through a fuzzy rule system. A fuzzy rule system consists of surface (Table A1). Basically, no-tillage represents the lower
i ¼ 1, ., M rules and can be represented in the form of a matrix with decomposition scenario while conventional tillage represents the
positive integer values Ai,j and Bi: most favorable scenario for residue decomposition.
0 1
A1;1 . A1;K B1 Table A1
R ¼ @ . . . .A (A5) Potential impacts on soil condition of the main tillage tools (modified from Ferraro
AM;1 . AM;K BK et al., 2003)a

Tillage tool STU STA Tool impact


Finally, rules were aggregated (i.e. defuzzification process) with
the purpose of transforming the set of numeric conclusions from Moldboard plow 1 0.8 0.92
Disk plow 0.9 0.8 0.86
each module into a single crisp value. There are several defuzzifi- Heavy tandem disk harrows 0.7 1 0.82
cation methods, such as the weighted-average, maximum Disk harrow 0.7 0.8 0.74
membership, average maximum membership, and center of Field cultivator 0.65 0.1 0.43
gravity. In this study, the weighted-average defuzzification method Chisel plow 0.4 0.2 0.32
Rotary rod 0.4 0.1 0.28
(Takagi and Sugeno, 1985) was employed.
Spike-tooth harrow 0.4 0.1 0.28
Vibrocultivator 0.4 0.1 0.28
A.3. Model parameters Rotary mower 0.4 0 0.28
Single sweep row cultivator 0.3 0.1 0.28
Driller 0.1 0.3 0.18
A.3.1. Secondary module 1 (SM1): aboveground net primary Bed shaper 0.2 0.1 0.18
productivity decomposition (ANPP decomposition) Rod weeder 0.2 0.1 0.16
(a). Abiotic decomposition. This primary module analyzes the Paraplow 0.2 0 0.12
Aerial sprayer 0 0 0
effect of four main components of the decomposition process. The
Terrestrial sprayer 0 0 0
final value of this module is calculated as follows:
a
Characteristic of each tillage operation: STU: relative magnitude of crop residue
Abiotic decomposition ¼ substrate decomposition incorporation after tillage; STA: relative effect on stability of soil aggregates after
tillage (0 ¼ minimum effect, 1 ¼ maximum effect). Tool impact ¼ 0.6STA þ 0.4STU.
 water availability
 crop residue incorporation Temperature: This module describes the effect of temperature on
residue decomposition rate. The input variable is the average air
 temperature (A6) temperature during the fallow period (Stroo et al., 1989).
where abiotic decomposition is the maximum (or potential)
decomposition rate related to abiotic factors and substrate decom- (b). Biotic decomposition. This module was built in order to eval-
position, water availability, crop residue incorporation and tempera- uate the effect of pesticides on the biotic component of the crop
ture are factors, scaled from 0 to 1, accounting for the limitations to residue decomposition process. The input variable is represented
decomposition imposed by residue C:N ratio, moisture, residue– by the amount of insect toxicity units (Ferraro et al., 2003). This
soil contact and temperature, respectively. variable brings together two other variables that describe the
Substrate decomposition: The kinetics of crop residue decom- toxicity and the amount of active ingredients utilized in each field:
position and their mineralization are largely influenced by the (1) contact acute Lethal Dose 50 for bees (mg active ingredient/bee)
quality of the plant materials (Ma et al., 1999). The N content, or and (2) the dose applied (mg active ingredient/ha) with each
their C:N ratio often controls the early decomposition of crop pesticide application:
residues, particularly residues having a high initial C:N ratio such as X
mature cereals (Douglas and Rickman, 1992). Therefore, most
Insect toxicity units ½TUins ¼ ðD=LD 50bÞ (A7)
i:n:
models have developed a function (e.g. C:N factor), which is to
reduce the decay rate when microbial N demand for decomposing where, D ¼ Dose applied (g formulated product/ha); LD
micro flora is higher than the N supply (Gijsman et al., 2002). In this 50b ¼ contact acute lethal dose 50 for bees (mg formulated product/
model, the C:N module assesses the litter quality effect on residue bee). Threshold values for Min and Max fuzzy subsets (Table 1)
368 D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370

come from a previous pesticide impact assessment (Ferraro et al., Table 2). Full membership to the Max fuzzy subset (i.e. the most
2003) and are equivalent to no pesticide used (i.e. TUi ¼ 0) and to an desirable scenario of soil nitrogen management) is reached when
applied dose of Chlorpyrifos equal to 0.33 of the manufacturer’s the calculated ratio is equal to 1.
label dose (i.e. TUi ¼ 1.903), respectively. Nitrogen output: Nitrogen output (Nout) was calculated as the
amount of nitrogen harvested from the crop grain. It depends on
A.3.2. Secondary module 2 (SM2): soil carbon depletion the specific nitrogen seed concentration (NSC) and the crop yield
The fuzzy rule system of this secondary module integrates two (Janssen, 1998):
partial modules: ANPP decomposition (SM1) and harvest of
aboveground net primary production (HANPP) (Fig. 1). Nout ¼ crop yield½t=ha  NSC½% (A13)
where, NSC ¼ 1.6 (maize); 5.5 (soybean); 2.45 (wheat).
(a). Human appropriation of aboveground net primary production Nitrogen input: Nitrogen input was calculated by the addition of
(HANPP). fertilizers to cropland (N applied) and by biological fixation (N
fixation) from crop cultivation. The amount of biological fixation is
HANPP ¼ crop residue mass=above ground net
important in leguminous plants (soybean) assuming that soybeans
primary production of potential vegetation ðA8Þ obtain half their N uptake from biological fixation (Sheldrick et al.,
2003).
The primary module HANPP is calculated by dividing two partial
input variables. (b). Soil fertility. This module considers the potential availability of
Crop residue mass: Crop residue mass (RM) represents the nitrogen for crop uptake. It assumes that quantity of N available
annual input of biomass due to crop growth, and its value is esti- from natural sources is closely related to soil organic matter
mated from crop yield and the average harvest index for each crop content (Van Keulen, 1995).
species (Steiner, 1994; Schomberg et al., 1996; Steiner et al., 1999):
A.3.4. Secondary module 4 (SM4): soil chemical degradation
RM maizeðt=haÞ ¼ Y  ð1:51YÞ þ 3:7 (A9) Final score of SM4 is determined by the fuzzy if-then rules eli-
cited from the expert panel. In terms of soil condition evaluation,
SM4 rules are built to obtain a single conclusion through the partial
RM wheat and soybeanðt=haÞ ¼ Y  ð2:56YÞ þ 0:4 (A10)
effects of crop management on soil nitrogen and carbon balance.
where Y: crop yield (t/ha).
Aboveground net primary production (ANNP) of potential vegeta- A.3.5. Secondary module 5 (SM5): soil physical degradation
tion: Potential vegetation is defined as the maximum biomass that (a). Tillage impact. Soil tilling was characterized by the tillage tools
could be established in a given area assuming a lack of human used in each field. The impact of each one of these tools was
influence (Rey, 1999). Final ANNP of potential vegetation corre- assessed through their effect on two of the soil properties most
sponds to an ecologically defined reference condition. The most affected by tillage: (1) the quantity of stubble remaining after tillage
productive agroecosystems in Argentina are located in the Pampa operations and (2) the stability of soil aggregates. Data from
region, a fertile plain originally covered by grasslands, which during a previous impact assessment (Ferraro et al., 2003) was used to
the 1900s and 2000s was transformed into an agricultural land rank the relative impact of each tillage tool (Table A1). The input
mosaic by grazing and farming activities. Therefore, ANNP of variable for the calculation of this module is:
potential vegetation was calculated through the algorithms from
one empirical-based model that considered local rainfall as the   X
main control of primary production in grasslands (Lieth, 1975; Sala Tool impact Timp ¼ 0:6STUi þ 0:4STAi (A14)
1.n
et al., 1988; Paruelo et al., 1999):
where, STU is the relative magnitude of crop residue incorporation
ANNP potential vegetationðt=haÞ after tillage and STA is relative effect on stability of soil aggregates
¼ ð49:4 þ 0:64  MAPÞ=10 ðA11Þ after tillage. The use of two moldboard plow operations (i.e.
Timp ¼ 1.84) for one crop, and only one drill operation (i.e. Timp ¼ 0),
where MAP ¼ mean annual precipitation (mm). This model defined the threshold values for Min and Max fuzzy subsets (Table
considered the precipitation use efficiency (PUE) value to account 1), respectively.
for the ANNP variability (i.e. structural model) and it was calibrated
for the 200–1200 mm AP range (Paruelo et al., 1999). (b). Soil erodibility. This module aims at assessing the soil erosion
vulnerability of agricultural fields. The soil properties used are
A.3.3. Secondary module 3 (SM3): soil nitrogen depletion reflected in the soil erodibility factor (K) in the Revised Universal
The fuzzy rule system of this secondary module integrates two Soil Loss Equation (RUSLE) (Renard et al., 1997). The K factor is an
partial modules: soil nitrogen balance and soil fertility. empirical measure of soil erodibility as affected by intrinsic soil
properties, and was calculated using the following formula
(a). Soil nitrogen balance. This module calculates the potential (Wischmeier and Smith, 1978):
unbalance in soil nitrogen status due to either overfertilization or
insufficient soil nitrogen replenishment. It was the only primary 2:1  104 ð12  SOMÞM 1:14 þ 3:25ðS  2Þ þ 2:5ðP  3Þ
module that was calculated through a triangular MF (see Table 2). K ¼
100
The MF variable for soil nitrogen balance calculation is the nitrogen (A15)
output/input ratio:
where, SOM is soil organic matter content, M is (%silt þ %very fine
Soil nitrogen balance : sand)(100  %clay), S is soil structure code and P is permeability
nitrogen output½kg=ha=nitrogen input½kg=ha ðA12Þ class. The unit of K factor is t/h ha1 MJ mm1. Soil structure code
from RUSLE is (1) very fine granular, (2) fine granular, (3) med or
Full membership to the Min fuzzy subset is reached when the coarse granular and (4) blocky, platy, or massive. Seven perme-
calculated ratio is equal to 0 or 2 (i.e. the base of the triangular MF, ability classes are defined: (1) excessively drained, (2) somewhat
D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370 369

excessively drained, (3) well drained, (4) moderately well drained, Girard, N., Hubert, B., 1999. Modelling expert knowledge with knowledge-based
systems to design decision aids: the example of a knowledge-based model on
(5) somewhat poorly drained, (6) poorly drained, and (7) very
grazing management. Agricultural Systems 59, 123–144.
poorly drained. Once the soil profile variables were defined, the K Giuffré, L., Romaniuk, R., Conti, M., Bartoloni, N., 2006. Multivariate evaluation by
factor was used as the input variable for calculating the soil erod- quality indicators of no-tillage system in Argiudolls of rolling pampa (Argen-
ibility primary module. Threshold values for Min and Max fuzzy tina). Biology and Fertility of Soils 42, 556–560.
Glanz, J.T., 1995. Saving our Soil: Solutions for Sustaining Earth’s Vital Resource,
subsets (Table 1) denote the lowest and highest risk for soil erosion, Boulder, CO.
respectively. The lowest soil erodibility scenario (i.e. K factor ¼ 0) is Goossens, L.H.J., Cooke, R.M., Woudenberg, F., Torn, P.v.d., 1998. Expert judge-
represented by a soil profile with values of M ¼ 1700, SOM ¼ 4%, ment and lethal toxicity of inhaled chemicals. Journal of Risk Research 1,
117–133.
S ¼ 1 and P ¼ 1. Max fuzzy subset (i.e. K factor ¼ 0.75) is represented Gregorich, E.G., Carter, M.R., Doran, J.W., Pankhurst, C.E., Dwyer, L.M., 1994. Towards
by a soil profile with values of SOM ¼ 0%, M ¼ 6800, S ¼ 6 and P ¼ 4 a minimum data set to assess soil organic matter in agricultural soils. Canadian
(Renard et al., 1997). Journal of Soil Science 74, 367–386.
Guerif, J., Richard, G., Dürr, C., Machet, J.M., Recous, S., Roger-Estrade, J., 2001. A
review of tillage effects on crop residue management, seedbed conditions and
seedling establishment. Soil & Tillage Research 61, 13–32.
References Haberl, H., Geissler, S., 2000. Cascade utilization of biomass: strategies for a more
efficient use of a scarce resource. Ecological Engineering 16, 111–121.
Alvarez, R., Alvarez, C.R., Lorenzo, G., 2001. Carbon dioxide fluxes following tillage Hall, A., Rebella, C., Ghersa, C., Culot, P., 1992. Filed-crop systems of the Pampas. In:
from a mollisol in the Argentine rolling pampa. European Journal of Soil Biology Pearson, C.J. (Ed.), Ecosystems of the World. Elsevier, The Netherlands, pp.
37, 161–166. 413–449.
Andrews, S.S., Karlen, D.L., Cambardella, C.A., 2004. The soil management assess- Iliadis, L.S., 2005. A decision support system applying an integrated fuzzy model for
ment framework: a quantitative soil quality evaluation method. Soil Science long-term forest fire risk estimation. Environmental Modelling & Software 20,
Society of America Journal 68, 1945–1962. 613.
Andrews, S.S., Karlen, D.L., Mitchell, J.P., 2002. A comparison of soil quality indexing Janssen, B.H., 1998. Efficient use of nutrients: an art of balancing. Field Crops
methods for vegetable production systems in Northern California. Agriculture Research 56, 197–201.
Ecosystems & Environment 90, 25–45. Lara, P., Stancu-Minasian, I., 1999. Fractional programming: a tool for the assess-
Aon, M.A., Sarena, D.E., Burgos, J.L., Cortassa, S., 2001. (Micro)biological, chemical ment of sustainability. Agricultural Systems 62, 131–141.
and physical properties of soils subjected to conventional or no-till Lenhart, T., Eckhardt, K., Fohrer, N., Frede, H.G., 2002. Comparison of two different
management: an assessment of their quality status. Soil & Tillage Research 60, approaches of sensitivity analysis. Physics and Chemistry of the Earth, Parts A/
173–186. B/C 27, 645.
Arshad, M.A., 1999. Tillage and soil quality: tillage practices for sustainable agri- Levitan, L., Merwin, I., Kovach, J., 1995. Assessing the relative environmental
culture and environmental quality in different agroecosystems. Soil & Tillage impacts of agricultural pesticides: the quest for a holistic method. Agriculture
Research 53, 1–2. Ecosystems & Environment 55, 153–168.
Arshad, M.A., Martin, S., 2002. Identifying critical limits for soil quality indicators in Lieth, H., 1975. Modelling the primary productivity of the world. In: Whittaker, R.H.
agro-ecosystems. Agriculture Ecosystems & Environment 88, 153–160. (Ed.), Primary Productivity of the Biosphere. Springer-Verlag, New York, pp.
Barrios, E., 2007. Soil biota, ecosystem services and land productivity. Ecological 237–263.
Economics 64, 269–285. Ma, L., Peterson, G., Ahuja, L., Sherrod, L., Shaffer, M., Rojas, K., 1999. Decomposition
Beck, M.B., Chen, J., 2000. Assuring the quality of models designed for predictive of surface crop residues in long-term studies of dryland agroecosystems.
tasks. In: Saltelli, A., Chan, K., Scott, E.M. (Eds.), Sensitivity Analysis. Wiley, Agronomy Journal 91, 401–409.
Chichester, pp. 402–420. Maddonni, G., Urricariet, S., Ghersa, C., Lavado, R., 1999. Assessing soil quality in the
Bellamy, J.A., Lowes, D., 1999. Modelling change in state of complex ecological rolling Pampa, using soil properties and maize characteristics. Agronomy
systems in space and time: an application to sustainable grazing management. Journal 91, 280–287.
Environment International 25, 701–712. Marcot, B.G., Holthausen, R.S., Raphael, M.G., Rowland, M.M., Wisdom, M.J., 2001.
Clemen, R.T., Winkler, R.L., 1999. Combining probability distributions from experts Using Bayesian belief networks to evaluate fish and wildlife population viability
in risk analysis. Risk Analysis 19, 187–203. under land management alternatives from an environmental impact statement.
Cornelissen, A.M.G., van der Berg, J., Koops, W.J., Grossman, M., Udo, H.M.J., 2001. Forest Ecology and Management 153, 29–42.
Assessment of the contribution of sustainability indicators to sustainable Martin, T.G., Kuhnert, P.M., Mengersen, K., Possingham, H.P., 2005. The power of
development: a novel approach using fuzzy set theory. Agriculture Ecosystems expert opinion in ecological models using Bayesian methods: impact of grazing
& Environment 86, 173–185. on birds. Ecological Applications 15, 266–280.
Davis, J.R., 1995. Expert systems and environmental modelling. In: Jakeman, A.J., Martı́nez-Ghersa, M.A., Ghersa, C.M., Satorre, E.H., 2000. Coevolution of agricultural
Beck, M.B., McAleer, M.J. (Eds.), Modelling Change in Environmental Systems. systems and their weed companions: implications for research. Field Crops
Wiley, Chichester, pp. 505–517. Research 67, 181–190.
Doran, J., Sarrantonio, M., Liebig, M., 1996. Soil Health and Sustainability. In: Matson, P.A., Parton, W.J., Power, A.G., Swift, M.J., 1997. Agricultural intensification
Sparks, D.L. (Ed.), Advances in Agronomy. Academic Press Inc, San Diego, CA, pp. and ecosystem properties. Science, 277.
1–54. Maud, J., Edwards-Jones, G., Quin, F., 2001. Comparative evaluation of pesticide risk
Doran, J.W., 2002. Soil health and global sustainability: translating science into indices for policy development and assessment in the United Kingdom. Agri-
practice. Agriculture. Ecosystems & Environment 88, 119–127. culture Ecosystems & Environment 86, 59–73.
Douglas, C., Rickman, R., 1992. Estimating crop residue decomposition from air Mebratu, D., 1998. Sustainability and sustainable development: historical
temperature, initial nitrogen content, and residue placement. Soil Science and conceptual review. Environmental Impact Assessment Review 18,
Society of America Journal 56, 272–278. 493–520.
Ducey, M., Larson, B., 1999. A fuzzy set approach to the problem of sustainability. Metternicht, G., Gonzalez, S., 2005. FUERO: foundations of a fuzzy exploratory
Forest Ecology and Management 115, 29–40. model for soil erosion hazard prediction. Environmental Modelling & Software
de la Fuente, E.B., Suarez, S.A., Ghersa, C.M., 2006. Soybean weed community 20, 715.
composition and richness between 1995 and 2003 in the rolling Pampas Moscatelli, G., Salazar Lea Plaza, J.C., Godagnogne, R., Gringberg, H., Sánchez, J.,
(Argentina). Agriculture Ecosystems & Environment 115, 229–236. Ferrao, R., Cuenca, M., 1980. Mapa de suelos de la provincia de Buenos Aires
Fabrizzi, K.P., Garcia, F.O., Costa, J.L., Picone, L.I., 2005. Soil water dynamics, physical 1:500000 (Soil map of Buenos Aires province 1:500000). In: Actas de la IX
properties and corn and wheat responses to minimum and no-tillage systems Reunión Argentina de la Ciencia del Suelo. Asociación Argentina de la Ciencia
in the southern Pampas of Argentina. Soil & Tillage Research 81, 57–69. del Suelo, pp. 1079–1089.
Farshad, A., Zinck, J.A., 1993. Seeking agricultural sustainability. Agriculture Mowrer, H.T., 2000. Uncertainty in natural resource decision support systems:
Ecosystems & Environment 47, 1–12. sources, interpretation, and importance. Computers and Electronics in Agri-
Ferraro, D.O., Ghersa, C.M., Sznaider, G.A., 2003. Evaluation of environmental impact culture 27, 139–154.
indicators using fuzzy logic to assess the mixed cropping systems of the Inland Neeteson, J.J., 1990. Development of nitrogen fertilizer recommendations for arable
Pampa, Argentina. Agriculture Ecosystems & Environment 96, 1–18. crops in the Netherlands in relation to nitrate leaching. Nutrient Cycling in
Ferreras, L.A., Costa, J.L., Garcia, F.O., Pecorari, C., 2000. Effect of no-tillage on some Agroecosystems 26, 291–298.
soil physical properties of a structural degraded Petrocalcic Paleudoll of the Nguyen, T.G., de Kok, J.L., 2007. Systematic testing of an integrated systems model
southern ‘‘Pampa’’ of Argentina. Soil & Tillage Research 54, 31–39. for coastal zone management using sensitivity and uncertainty analyses.
Fleming, G., Merwe, M.v.d., McFerren, G., 2007. Fuzzy expert systems and GIS for Environmental Modelling & Software 22, 1572–1587.
cholera health risk prediction in southern Africa. Environmental Modelling & Parton, W.J., Schimel, D.S., Cole, C.V., Ojima, D.S., 1987. Analysis of factors controlling
Software 22, 442. soil organic matter levels in great plains grassland. Soil Science Society of
Giarratano, J.C., Riley, G., 1989. Expert Systems: Principles and Programming. PWS- America Journal 51, 1173–1179.
Kent, Boston, 632 p. Paruelo, J.M., Lauenroth, W.K., Burke, I.C., Sala, O.E., 1999. Grassland precipitation-
Gijsman, A.J., Hoogenboom, G., Parton, W.J., Kerridge, P.C., 2002. Modifying DSSAT use efficiency varies across a resource gradient. Ecosystems 2, 64.
crop models for low-input agricultural systems using a soil organic matter- Paterson, B., Stuart-Hill, G., Underhill, L.G., Dunne, T.T., Schinzel, B., Brown, C.,
residue module from CENTURY. Agronomy Journal 94, 462–474. Beytell, B., Demas, F., Lindeque, P., Tagg, J., Weaver, C., 2008. A fuzzy decision
370 D.O. Ferraro / Environmental Modelling & Software 24 (2009) 359–370

support tool for wildlife translocations into communal conservancies in Coupland, R.T. (Ed.), Ecosystems of the World 8A. Natural grasslands. Intro-
Namibia. Environmental Modelling & Software 23, 521–534. duction and Western Hemisphere. Elsevier, New York, pp. 367–407.
Pearson, C.J., Ison, R.L., 1997. Agronomy of Grassland System. Cambridge University Steinbach, H.S., Alvarez, R., 2006. Changes in soil organic carbon contents and
Press. nitrous oxide emissions after introduction of no-till in Pampean agro-
Quemada, M., 2005. Predicting crop residue decomposition using moisture adjusted ecosystems. Journal of Environmental Quality 35, 3–13.
time scales. Nutrient Cycling in Agroecosystems 70, 283. Steiner, J.L., 1994. Crop residue effects on water conservation. In: Unger, P. (Ed.),
Quinn, M.A., Hall, M.H., 1996. Compensatory growth response of the legume, Managing Agricultural Residues. Lewis, Boca Raton, FL, pp. 41–76.
Medicago sativa, to defoliation and denodulation. Entomologia Experimentalis Steiner, J.L., Schomberg, H.H., Unger, P.W., Cresap, J.,1999. Crop residue decomposition
et Applicata 78, 243–252. in no-tillage small-grain fields. Soil Science Society of America Journal 63, 1817.
Raun, W., Johnson, G., 1999. Improving nitrogen use efficiency for cereal production. Stroo, H.F., Bristow, K.L., Elliott, L.F., Papendick, R.I., Campbell, G.S., 1989. Predicting
Agronomy Journal 91, 357–363. rates of wheat residue decomposition. Soil Science Society of America Journal
Reed, M.S., Fraser, E.D.G., Dougill, A.J., 2006. An adaptive learning process for 53, 91.
developing and applying sustainability indicators with local communities. Takagi, T., Sugeno, M., 1985. Fuzzy identification of systems and its applications to
Ecological Economics 59, 406. modeling and control. IEEE Transactions on Systems, Man and Cybernetics 15,
Renard, K.G., Foster, G.R., Weesies, G.A., McCool, D.A., Yoder, D.C., 1997. Predicting 116.
Soil Erosion by Water: A Guide to Conservation Planning with the Revised Tan, R.R., 2005. Application of symmetric fuzzy linear programming in life cycle
Universal Soil Loss Equation (RUSLE). US Government Print Office, Washington, assessment. Environmental Modelling & Software 20, 1343.
DC. Thorburn, P.J., Probert, M.E., Robertson, F.A., 2001. Modelling decomposition of sugar
Rey, J.M., 1999. Modelling potential evapotranspiration of potential vegetation. cane surface residues with APSIM-Residue. Field Crops Research 70, 223–232.
Ecological Modelling 123, 141–159. Van Der Fels-Klerx, I.H.J., Goossens, L.H.J., Saatkamp, H.W., Horst, S.H.S., 2002.
Reynolds, K., 1999. NetWeaver for EMDS Version 2.0 User Guide: a Knowledge Base Elicitation of quantitative data from a heterogeneous expert panel: formal
Development System. General Technical Report No. PNW-GTR 471. US process and application in animal health. Risk Analysis 22, 67–81.
Department of Agriculture, Forest Service, Pacific Northwest Research Station, Van Keulen, H., 1995. Sustainability and long-term dynamics of soil organic matter
Portland, OR. and nutrients under alternative management strategies. In: Bouma, J.,
Sahinidis, N.V., 2004. Optimization under uncertainty: state-of-the-art and oppor- Kuyvenhoven, A., Bouman, B.A.M., Luyten, J.C., Zandstra, H.G. (Eds.), Eco-
tunities. Computers & Chemical Engineering 28, 971–983. regional Approaches for Sustainable Land Use and Food Production. Kluwer
Sala, O.E., Parton, W.J., Joyce, L.A., Lauenroth, W., 1988. Primary production of the Acad. Publ., Dordrecht, pp. 353–375.
central grassland region of the United States. Ecology 69, 40–45. Viglizzo, E.F., Pordomingo, A.J., Castro, M.G., Lertora, F.A., 2003. Environmental
Saunders, M.C., Sullivan, T.J., Nash, B.L., Tonnessen, K.A., Miller, B.J., 2005. A assessment of agriculture at a regional scale in the Pampas of Argentina.
knowledge-based approach for classifying lake water chemistry. Knowledge- Environmental Monitoring and Assessment 87, 169–195.
Based Systems 18, 47–54. van der Werf, H., Zimmer, C., 1998. An indicator of pesticide environmental impact
Savabi, M.R., Stott, D.E., 1995. Plant residue impact on rainfall interception. Trans. based on a fuzzy expert system. Chemosphere 36, 2225–2249.
ASAE 37, 1093–1098. von Wirén-Lehr, S., 2001. Sustainability in agriculturedan evaluation of principal
Schomberg, H.H., Steiner, J.L., Evett, S.R., Moulin, A.P., 1996. Climatic influence on goal-oriented concepts to close the gap between theory and practice. Agricul-
residue decomposition prediction in the wind erosion prediction system. ture Ecosystems & Environment 84, 115.
Theoretical and Applied Climatology 54, 5. WCED, 1987. Our Common Future. Oxford University Press, Oxford.
Schreiber, J.D., 1985. Leaching of nitrogen, phosphorus, and organic carbon from Winkler, R.L., Clemen, R.T., 2004. Multiple experts vs. multiple methods: combining
wheat straw residues: II. Loading rate. Journal of Environmental Quality 14, correlation assessments. Decision Analysis 1, 167–176.
256–260. Wischmeier, W.H., Smith, D.D., 1978. Predicting Rainfall Erosion Losses––A Guide to
Sheldrick, W.F., Syers, J.K., Lingard, J., 2003. Soil nutrient audits for China to estimate Conservation Planning. US Government Print Office, Washington, DC.
nutrient balances and output/input relationships. Agriculture Ecosystems & Yamada, K., Elith, J., McCarthy, M., Zerger, A., 2003. Eliciting and integrating expert
Environment 94, 341. knowledge for wildlife habitat modelling. Ecological Modelling 165, 251–264.
Soil-Survey-Staff, 1999. Soil Taxonomy: a Basic System of Soil Classification for Yaman, B., 1996. Formal aspects of model validity and validation in system
Making and Interpreting Soil Surveys. US Department of Agriculture Soil dynamics. System Dynamics Review 12, 183–210.
Conservation Service, Washington, DC. Yli-Viikari, A., Hietala-Koivu, R., Huusela-Veistola, E., Hyvonen, T., Perala, P., Turtola, E.,
Soll, J.B., 1999. Intuitive theories of information: beliefs about the value of redun- 2007. Evaluating agri-environmental indicators (AEIs)–use and limitations of
dancy. Cognitive Psychology 38, 317–346. international indicators at national level. Ecological Indicators 7, 150–163.
Soriano, A., León, R.J.C., Sala, O.E., Lavado, R.S., Deregibus, V.A., Cahuépé, M.A., Zadeh, L.A., 1965. Fuzzy sets. Information and Control 8, 338.
Scaglia, O.A., Velázquez, C.A., Lemcoff, J.H., 1991. Rı́o de la Plata grasslands. In: Zimmermann, H.J., 1996. Fuzzy Set Theory and its Applications, Boston.

You might also like