Forest Ecology and Management 256 (2008) 1408–1417

Contents lists available at ScienceDirect

Forest Ecology and Management

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

Definition of priority areas for forest conservation through the ordered

weighted averaging method
Roberta de Oliveira Averna Valente *, Carlos Alberto Vettorazzi
Rural Engineering Department, Luiz de Queiroz School of Agriculture, University of São Paulo (USP), Av. Pádua Dias 11, Caixa Postal 9, 13418-900 Piracicaba, SP, Brazil

A R T I C L E I N F O A B S T R A C T

Article history: The general objective of this study was to evaluate the ordered weighted averaging (OWA) method,
Received 25 October 2007 integrated to a geographic information systems (GIS), in the definition of priority areas for forest
Received in revised form 2 July 2008 conservation in a Brazilian river basin, aiming at to increase the regional biodiversity. We demonstrated
Accepted 5 July 2008
how one could obtain a range of alternatives by applying OWA, including the one obtained by the
weighted linear combination method and, also the use of the analytic hierarchy process (AHP) to
Keywords:
structure the decision problem and to assign the importance to each criterion. The criteria considered
Forest fragmentation
important to this study were: proximity to forest patches; proximity among forest patches with larger
Multicriteria evaluation
Sensitivity analysis
core area; proximity to surface water; distance from roads; distance from urban areas; and vulnerability
to erosion. OWA requires two sets of criteria weights: the weights of relative criterion importance and the
order weights. Thus, Participatory Technique was used to define the criteria set and the criterion
importance (based in AHP). In order to obtain the second set of weights we considered the influence of
each criterion, as well as the importance of each one, on this decision-making process. The sensitivity
analysis indicated coherence among the criterion importance weights, the order weights, and the
solution. According to this analysis, only the proximity to surface water criterion is not important to
identify priority areas for forest conservation. Finally, we can highlight that the OWA method is flexible,
easy to be implemented and, mainly, it facilitates a better understanding of the alternative land-use
suitability patterns.
ß 2008 Elsevier B.V. All rights reserved.

1. Introduction that biodiversity involves all species of plants, animals, micro-

Anthropic forest fragmentation has been one of the main causes
of changes in the structure, as well as in the ecological processes, in
different types of landscapes. Fragmentation can be described by
some of its effects, such as the increase in the isolation of the patches,
decrease in their size, and increase in their exposure to external
disturbance, such as the invasion by alien species or changes in their
physical conditions (Geneletti, 2004a). Those effects can potentially
lead to a decline in the ecosystem biodiversity, as well as in its
stability and ability to recover from disturbances (Baskent, 1999;
Saunders et al., 1991).
The actions of forest conservation must go, therefore, in the
opposite direction to the fragmentation (Geneletti, 2004b). Thus, the
maintenance and/or the restoration of the biodiversity of affected
landscapes will be guaranteed. Mcneely et al. (1990) highlighted
* Corresponding author. Tel.: +55 19 34478533; fax: +55 19 34178571.
E-mail addresses: roavalen@esalq.usp.br (R.O.A. Valente), cavettor@esalq.usp.br
(C.A. Vettorazzi).
organisms, as well as the ecosystems and processes in which it takes
part. An important aspect of conservation planning should be the
spatialization of those actions (Phua and Minowa, 2005). In this
context, the definition of priority areas represents an effective and
economic method. According to Collins et al. (2001), the priority
areas analysis aims at identifying the most appropriate spatial
pattern for future land uses, according to specific requirements,
preferences or predictor factors of some activity or objective.
Multicriteria evaluation (MCE) is one of the decision-making
processes employed in the prioritization of areas, and its integration
with geographic information systems (GIS) has been considered
an important improvement to the conventional map overlay
approaches (Malczewski, 1999; Thill, 1999; Eastman, 2001). Such
integration (what and where to conserve) has facilitated, according
to Kangas et al. (2000) and Vlahos and Herbst (2000), the planning,
optimization, and success of forest conservation actions, driving the
actions according to their degree of risk, suitability, or priority.
The basis to decision making, in MCE, that can be measured and
evaluated, is named criterion. It is the evidence upon which an
individual can be assigned to a decision set. Criteria can be of two

0378-1127/$ – see front matter ß 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.foreco.2008.07.006
 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417
kinds: factors and constraints, and can pertain either to attributes of
an individual or to an entire decision set (Eastman, 2001). A factor is a
criterion that enhances or detracts from the suitability of a specific
activity under consideration. The factors used in the decision-
making process are those that represent the critical characteristics of
a habitat (Randhir et al., 2001). It is, therefore, most commonly
measured on a continuous scale, what is an advantage in relation to
the representation of the landscape characteristics in specific
intervals, what normally leads to the loss of information and error
propagation. Constraints impose limitations on the set of decision
alternatives (Malczewski, 1999). They serve to exclude areas and to
determine space limits to the distribution of the choice possibilities
(Eastman, 2001).
In general, decision makers rank a set of decision alternatives
and choose the best according to their preferences. To be able to
rank, they select the criteria that are relevant to the current
problem and that are of significance in their choice (e.g. Bouyssou
et al., 2001). To rank the alternatives, each one of them should
be evaluated with respect to each criterion. Decision-support
methods can be used to analyze the decision situation to help the
decision maker to make the best, or at least a satisfactory choice
(Kangas and Kangas, 2005).
Malczewski (2006) cited that there are two fundamental classes
of MCE methods in GIS environment: the Boolean overlay
operations (non-compensatory combination rules) and the
weighted linear combination (WLC) methods (compensatory
combination rules). The WLC method is more flexible in terms
of evaluating management alternatives compared to Boolean
operations (Malczewski, 2000). It has been the most often used
approach in different applications. Lathrop Jr. and Bognar (1998),
Geneletti (2004b) and Phua and Minowa (2005) employed WLC to
identify forest conservation priorities. Mendoza and Prabhu (2000)
and Gkaraveli et al. (2004) applied the GIS-based WLC method to
the determination of priority areas for forest restoration.
The Boolean and WLC approaches can be generalized within
the framework of the ordered weighted averaging (OWA)
(Malczewski, 1999; Jiang and Eastman, 2000; Malczewski et al.,
2003; Malczewski and Rinner, 2005). The OWA is a family of
multicriteria aggregation operations (or multicriteria decision
rules) and its concept has been developed in the context of fuzzy
set theory (Yager, 1988). However, the use of these operations is
not limited to fuzzy sets (Malczewski et al., 2003). They involve
two sets of weights: the weights of relative criterion importance
and the order weights. By specifying an appropriate set of order
weights one can generate a wide range of outcome maps (decision
strategy) (Malczewski, 2006). Although OWA is a relatively new
concept, there have been several applications of this approach in
the GIS environment. Malczewski (1999) used the OWA for
analyzing land suitable for a housing project in Mexico. Jiang and
Eastman (2000) demonstrated the utility of the OWA for land-
use suitability problems in Africa. Calijuri et al. (2000) used it
in the identification of areas suitable for landfills, in Brazil.
Malczewski et al. (2003) successfully tested the OWA concept
and implementation as a core of a multicriteria evaluation
spatial decision support system to evaluate areas in a Canadian
watershed for rehabilitation and enhancement projects. Makro-
poulos and Butler (2006) suggested that OWA could be useful as a
decision-making tool for urban water management in London.
Boroushaki and Malczewski (2008) illustrated the application of
GIS-based OWA in the identification of the most suitable lands for
housing development in a Canadian Province.
According to Eastman and Jiang (1996), the OWA offers a
complete spectrum of decision strategies along the primary
dimensions of the tradeoff degrees (among criteria) and risk in
the decision-making process. Malczewski (2004) mentioned that
1409
the method is considered flexible due to its ability in assuming a
variety of solutions, ranging from risk-aversion (a location must
meet every criterion for being included in the decision set) to risk-
taking (a location will be included in the decision set even if only a
single criterion passes the test).
In this context, the general objective of this study was to
evaluate the ordered weighted averaging (OWA) method, inte-
grated to a GIS, in the definition of priority areas for forest
conservation in a Brazilian river basin, aiming at to increase the
regional biodiversity. The specific objective was to evaluate the
importance and influence of the selected factors on the decision-
support process, through a sensitivity analysis.
2. Methodology
2.1. Study area
Corumbataı́ river basin is located in the Central-Eastern region
of the State of São Paulo, Brazil, between the latitudes 228040 4600 S
and 228410 2800 S, and the longitudes 478260 2300 W and 478560 1500 W,
with approximately 170,000 ha. Its original forest cover is highly
fragmented, due to an unplanned process of land-use and
occupation. This basin’s landscape represents well the situation
of a large portion of the Brazilian territory originally covered by
the Atlantic Rainforest (considered one of the world’s hotspots),
where the natural resources and, consequently, the biodiversity,
are seriously affected by anthropic action. This river basin (Fig. 1)
is covered by only 11% of native forest (Semideciduous Seasonal
Forest) and 1% of Savanna. Pastures and crops are the predominant
land-use, with the first covering 44% of the basin and sugarcane
covering around 26%.
The forest remnants play a relevant role in nature conserva-
tion, considering that conservation actions, at the landscape
level, have to guarantee the processes and structure integrity
of the anthropic landscapes. Due to the inherent importance
of Corumbataı́ river basin as a representative of the Atlantic
Rainforest landscape and the Corumbataı́ river as an strategic
source of fresh water for a large region, there is a concentrated
effort from local, state and national government agencies,
experts related to universities and research institutes, and NGOs
to study the components (e.g. flora and fauna) and processes (e.g.
forest connectivity) of that basin and to define actions for its
restoration. From 1999 through 2001, a Director Plan for the
basin was developed, focusing on the relation between forest
cover and water resources. The plan was funded by the Water
Department of the Municipality of Piracicaba, the largest city in
the Corumbatai River Basin. This study is an evolution of the
Director Plan, and can help decision makers (specially govern-
ment agencies) in the forest restoration of the river basin,
including the protection and, overall, the sustainability of quality
and quantity of its components and processes.
2.2. Ordered weighted averaging
The ordered weighted averaging (OWA), in a GIS environment
(IDRISI Kilimanjaro), was employed in the definition of priority
areas for forest conservation. The steps, from the definition of the
criteria set important to decision-making process through the
criteria aggregation, using GIS-based OWA, will be described in the
following items.
2.2.1. Criteria
In the definition of criteria (factors and constraints) and of the
criterion importance weights, the Participatory Technique was
used, which involved opinions of a group of experts in the various
1410 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417
Fig. 1. Corumbataı́ river basin: land-use/land-cover and location.
areas related to the specific subject (Eastman, 2001). For this
project, this group involved ten experts familiarized with the
Corumbataı́ river basin, representing areas related to the study:
forest management, forest hydrology, forest restoration, forest
conservation, and wildlife management. The researchers received,
by e-mail, the preliminary summary of the project, with its
objectives; and a table to be filled out with (i) a criteria set
considered important for forest conservation, (ii) the reason to
include the criteria in this study, and (iii) criteria weights (1–10
scale), according to the criteria importance for the purposes of the
study. The experts are kept apart to each other (some unknown to
each other) to ensure that the responses were truly independent
and not influenced by others in the group. The responses (tables)
were then shared with the other participants, who responded
by revising or giving further arguments supporting their answers.
The following criteria set (Fig. 2) was considered important to the
study, and represents the consensus among the researchers
(experts and authors of this study).
(A) Proximity to surface water: vegetated corridors along streams
and rivers have an ecological importance in the dynamics of
the landscape, allowing the movement of fauna along the
landscape and contributing to the vegetation dispersion, what
will help the process of forest connectivity (Harper et al., 1992;
Forman and Collinge, 1997; Eastman, 2001).
(B) Proximity to forest patches: with the connectivity among
patches reestablished (directly or through other patches) the
subpopulations of a landscape are interconnected into a
demographic functional unit (Farina, 1998).
(C) Proximity among forest patches with the larger core areas:
larger and connected ecosystems are typically better for
conserving biodiversity than the smaller and isolated ones
(Noss et al., 1997). This is because the first group of patches
tends to host a higher number of species, and populations less
subjected to extinction with respect to the second group (Noss
et al., 1997; Farina, 1998).
(D) Distance from urban areas: for the success of conservation
actions it is interesting the prioritization of areas that are
distant from sources of disturbance (e.g. urban areas and
roads). Forest patches close to urban areas suffer from risk of
reduction in area or even complete extinction, as a conse-
quence of unsuitable land-use in these regions (Gutzwiller and
Barrow, 2003), increase in the fire hazard (Chen et al., 2001),
and expansion of the urban perimeter (Saunders et al., 1991),
amongst other factors.
(E) Distance from roads: roads cause both a direct and an indirect
reduction of ecosystems. Direct habitat loss refers to the
reduction of the total area of an ecosystem (i.e. by the
conversion of the original land cover into an artificial surface);
and indirect loss refers to the fragmentation and degradation of
the ecosystems (Geneletti, 2006).
(F) Vulnerability to erosion: through this factor, attention was
given to those soils with the highest values of erodibility and
slope. This is a risky situation to forest conservation since such
areas are more susceptible to soil erosion (Rosa et al., 2000).
(G) Factor maps were generated from the following layers: forest
patches (factors 1 and 2); drainage network (factor 3); urban
areas (factor 4); roads (factor 5); slope and soil erodibility (factor
6). Forest patches and urban areas were extracted from the land-
use/land-cover map of the study area, produced by supervised
digital classification (maximum likelihood algorithm) of LAND-
SAT-7/ETM+ images, with a global classification accuracy of 93%.
Forest patches smaller than 1 ha were not considered in the
analysis. Forest patches with core area were generated by the
elimination of a 20-m edge width. Other layers were obtained
from the Cartographic Database of the Geoprocessing Labora-
tory, University of São Paulo. These layers were projected in UTM
23S (Corrego Alegre Datum) with a 20-m resolution.
In order to generate the factor maps, except the vulnerability
to erosion, maps of distance from the features of interest were
produced. Such maps were standardized to a common scale (0–255
 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417
Fig. 2. Factors maps: (A) proximity to surface water, (B) proximity to forest patches, (C) proximity among forest patches with the larger core areas, (D) distance from urban
areas, (E) distance from roads and (F) vulnerability to erosion.
bytes) keeping the priority relations previously established for
each one. Thus, the linear monotonically decreasing function was
employed in order to minimize the proximity maps, that is, to
guarantee that the highest-priority areas were those located close
to forest patches, mainly to those with the largest core area
and, initially, close to the surface water. Otherwise, the linear
monotonically increasing function was used to maximize the
distance maps and it allowed that the highest-priority areas were
located away from urban areas and roads. Finally, the factor map
vulnerability to erosion was obtained by a combination between
the maps of slope and erodibility, already standardized (using
linear monotonically increasing function). The previous maximi-
zation of the maps lead to the fact that areas with the highest
values of soil erodibility and slope were included as the ones with
the highest priority, in this decision process.
The constraints in this study were the limits of the river basin
(i.e., the final map should consider only the region inside the limits of
the Corumbataı́ river basin); urban areas; and mining areas, i.e.
conservation actions did not take into consideration these areas. The
constraints are limitations imposed by nature or by human beings
that do not permit certain actions to be taken (Keeney, 1996).
2.2.2. Weights definition
2.2.2.1. Criterion importance weights. The purpose of the impor-
tance weights is to express the relative importance of each
1411
criterion in the set of criteria under consideration (Eastman, 2001;
Malczewski, 2004). Thus, at this stage, the decision maker’s
preferences with respect to the evaluation criteria are incorporated
to the decision process. After debate and careful analysis of the
criteria importance for the purposes of the study (firstly in the
scale from 1 to 10—Participatory Technique), the proximity among
forest patches with the larger core areas was considered the most
important criterion. On the other hand, the vulnerability to erosion
was considered the least important. The proximity to surface water
and proximity to forest patches were classified as second in
importance, and these two criteria are, at the same time, equally
important. The remaining two criteria (distance from urban areas
and distance from roads) occupied the third place, with the same
importance. In order to obtain the criterion importance weights we
considered the criteria ‘‘ranking’’ (including the relative impor-
tance between criteria) and the pairwise comparison method,
developed by Saaty (1980) in the context of the analytic hierarchy
process (AHP).
The pairwise comparison method employs an underlying
continuous scale (Fig. 3), with values from 1 to 9, to rate the
relative preferences for two criteria (Malczewski et al., 2003). The
value 1 indicates that two criteria are ‘‘equally’’ important and the
value 9 implies that one criterion is ‘‘extremely’’ more important
than the other. The comparisons are entered into a pairwise
comparison matrix (Table 1). The matrix is symmetrical, so that
only the upper triangular half needs to be filled in. The remaining
1412 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417

Fig. 3. Continuous rating scale used to prepare the pairwise comparison matrix.

Table 1
Pairwise comparison matrix, involving all criteria and criterion importance weight set, employed in the definition of priority areas for forest conservation in the Corumbataı́
river basin, São Paulo State, Brazil

Factors Vuln. Dist. urban Dist. roads Prox. water Prox. forest Prox. core Criterion importance weight

Vuln. 1 0.0725
Dist. urban 2 1 0.1171
Dist. roads 2 1 1 0.1171
Prox. water 2 2 2 1 0.1736
Prox. forest 2 2 2 1 1 0.1736
Prox. core 4 2 2 3 3 1 0.3462

Consistency ratio (CR) = 0.03 Total 1.0000

Vuln.: vulnerability to erosion; Dist. urban: distance from urban areas; Dist. roads: distance from roads; Prox. water: proximity to surface water; Prox. forest: proximity to
forest patches; proximity core: proximity among forest patches with the larger core areas.

cells are the reciprocals of the upper triangular half. The criterion
weighting pairwise comparison routine (GIS–IDRISI) takes as an
input the pairwise comparison matrix and generates the weights of
relative criterion importance as output. The weights are deter-
mined by normalizing the eigenvector associated with the
maximum eigenvalue of the (reciprocal) ratio matrix. In addition,
the GIS-routine allows for checking the consistency of pairwise
comparisons. The consistency ratio (CR) is designed in such a way
that if CR < 0.10, than the ratio indicates a reasonable level of
consistency in the pairwise comparison matrix; if however
CR  0.10, than the ratio values are indicative of inconsistent
judgments; in such cases one should reconsider and revise the
original values in the pairwise comparison matrix (Malczewski et
al., 2003). In this context, to obtain the ideal comparison value we
considered the criteria ranking, according to criterion relative
importance, previously established by the researchers (experts and
authors of this study) and the value of CR. Thus, a few iterations
and adjustments in the values were necessary in the pairwise
comparison matrix to obtain a satisfactory level consistency. Table
1 provides the final results of the pairwise comparison process.
2.2.2.2. Order weights. The order weights control the manner in
which the weighted criteria are aggregate (Yager, 1988). Mal-
czewski (2006) cited that they are associated with the criterion
values on a location-by-location (object-by-object) basis, and they
are assigned to a location’s attribute values in decreasing order
without considering which attribute the value comes from.
According to Jiang and Eastman (2000) while importance weights
are associated to the relative significance of a particular criterion
Fig. 4. Decision strategy space in OWA.
for the decision set, order weights control the position of the
aggregation operator on a continuum between the extremes AND
(risk-averse) and OR (risk-taking) as well as the degree of tradeoff
(Fig. 4).
We used the parameters, RISK (ANDness) and TRADEOFF, to
characterize the process of decision-making, as in the following
equations:
 X
1
RISK ¼ ½ðn  iÞW i  (1)
n1
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X ðWi  1=nÞ2
TRADEOFF ¼ 1  (2)
n1
in which n is the total number of factors, i is the order of factors;
and W is the weight for the factor of the ith order.
With different sets of order weights, one can generate a wide
range of decision strategies (i.e. solutions), in terms of risk and
tradeoff (Malczewski et al., 2003). The way in which the order
weights are assigned on the basis of the triangular decision
strategy space will determine the position in the risk dimension.
The logical OR (i.e. ANDness = 0) and AND (i.e. ANDneess = 1)
operators refer, respectively, the solutions associated with the
most risk-taking and risk-averse in the decision process (Fig. 4).
The ANDness = 0 (i.e. risk = 0) implies no tradeoff among evalua-
tion criteria. The map representing this solution indicates that
there is at least one criterion with the weighted criterion value of
zero at each location. This means that under this strategy no
conservation action should be taken (Eastman, 2001). Increasing
the value of ANDness (i.e. risk) from 0 to 0.5 corresponds to the
increasing tradeoff among evaluation criteria. This implies that
higher probabilities (ordered weights) are assigned to the higher-
ranking criterion values at a given location at expense of assigning
smaller probabilities to the lower ranking criterion values for that
location. As a result, the size of areas that could be recommended
for prioritization becomes gradually larger (Malczewski et al.,
2003). The strategy for ANDness = 0.5 (i.e. risk = 0.5) represents a
neutral attitude. It is also a strategy resulting in a full tradeoff
among criteria (represents the alternative corresponding to the
conventional weighted linear combination). Increasing the value of
ANDness from 0 to 1 represents decreasing the level of tradeoff
among criteria. This implies that greater and greater probabilities
(order weights) are assigned to the higher criterion values. The
solution represented by ANDness = 1 (i.e. risk = 1) implies that all
 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417 1413

river basin should be considered for prioritization (Malczewski et Table 3

Pairwise comparison matrix without the proximity to surface water factor and
al., 2003).
criterion importance weight set, employed in the definition of priority areas for
In this study, the order weights were distributed in the decision forest conservation in the Corumbataı́ river basin, São Paulo State, Brazil
strategy space (a continuum between the extremes AND and OR—
Factors Vuln. Dist. urban Dist. Prox. Prox. Criterion
Fig. 4) in order to obtain solutions with low, medium, and high risk-
roads forest core importance
taking. We considered twice the interval of 10% from the medium weight
risk-taking (0.5 in the continuum of the decision strategy space—
Vuln. 1 0.0867
Fig. 4) to characterize the intervals of risk-taking. According to
Dist. urban 2 1 0.1520
Malczewski et al. (2003), the degree of risk-taking (ANDness) is Dist. roads 2 1 1 0.1520
governed by the relative distribution of order weights among the Prox. forest 2 2 2 1 0.2198
ranked factors. Thus, the first step to obtain solutions with low, Prox. core 4 2 2 3 1 0.3895
medium, and high risk-taking was to define the rank order of Consistency ratio (CR) = 0.03 Total 1.0000
the criteria set. Successively associating the total order weight
Vuln.: vulnerability to erosion; Dist. urban: distance from urban areas; Dist. roads:
(total = 1) to each criterion and applying OWA aggregation opera-
distance from roads; Prox. forest: proximity to forest patches; proximity core:
tor implemented as a GIS routine, the following rank order was proximity among forest patches with the larger core areas.
obtained: distance from roads (lower ranking); distance from
urban areas; vulnerability to erosion; proximity among forest The sample size (52 points) was statistically defined, considering
patches with the larger core areas; proximity to forest patches; and the size of the study area (see Eastman, 2001). By sampling the
proximity to surface water (higher ranking). The rank order of the maps of criteria and priorities, the evaluation of the priority
criteria set reflects the influence of each criterion on the decision- associated to each point was done and, for each location, it was
making process (Eastman, 2001). determined the contribution of each criterion to the priority level.
Several methods for determining the order weights have been This procedure was also adopted: (i) to the selection of the weight
suggested in the literature. In the IDRISI implementation of OWA, sets presented in Table 2, in this case to choose between order
it is assumed that the decision maker can ‘‘intuitively’’ identify weight sets for each risk-taking level; (ii) in the sensitivity
the order weights based on the degree of ANDness and tradeoff analysis; and (iii) to choose the best alternative for the definition of
among criteria (Malczewski, 2006). In this study we considered priority areas for forest conservation in the Corumbataı́ river basin.
the influence of each criterion (i.e. rank order), as well as the The best alternative should be that one representing a correct
importance of each one (i.e. criterion importance), on the interaction between criteria, and one that best suit the needs and
decision-making process. Two order weight sets for each risk- expectations of conservation actions. According to Jiang and
taking level (e.g. two sets for low risk-taking) were proposed and Eastman (2000) the selection process (i.e. choose among alter-
evaluated (see Section 2.2.3), except for medium risk-taking. The natives) is wrought with subjectivity and it is intrinsically related
appropriate order weight sets are shown in Table 2. with the knowledge and sensibility of the analyst.

2.2.3. Criteria aggregation 2.3. Sensitivity analysis

The results of criteria aggregation, using OWA as a GIS routine,
were three maps for forest conservation in the Corumbataı́ river Sensitivity analysis should be performed to determine robust-
basin, i.e. alternatives associated with low, medium, and high risk- ness to the decision-making process (Malczewski, 1999). It is
taking, according to the order weight sets in Table 2. Those maps defined as a procedure for determining how the recommended
(on a 256-value continuous scale) were reclassified in order to course of action is affected by changes (in this case, through the
show only five levels of priority: very low, low, medium, high, and criteria set and its order weights) in the inputs of the analysis. The
very high. The class interval was based on the histogram and on the exploratory approach, as proposed by Malczewski (1999), was
limits of all maps. The three alternatives were carefully evaluated employed in the analysis: the matrix (Table 1) was reorganized
in order to choose the ‘‘best alternative’’ among them. A 52-point with the elimination of each criterion, and the relative importance
stratified random grid was generated to perform the evaluation. between the other criteria was maintained, to obtain the new
importance weights sets. This procedure took into considera-
tion the CR and the criteria importance. Consequently, in some
Table 2
Order weight sets employed to produce priority maps for forest conservation, in the
Corumbataı́ river basin, São Paulo State, Brazil, associated to low, medium and high Table 4
risk-taking Order weight sets employed to produce priority maps for forest conservation, in the
Corumbataı́ river basin, São Paulo State, Brazil, without the proximity to surface
Factors Order weights water factor and associated to low, medium and high risk-taking
Low risk-taking Medium risk-taking High risk-taking Factors Order weights
Vuln. 0.1200 0.1667 0.0800 Low risk-taking Medium risk-taking High risk-taking
Dist. urban 0.2200 0.1667 0.1800
Dist. roads 0.2200 0.1667 0.1800 Vuln. 0.0800 0.1667 0.0800
Prox. water 0.0500 0.1667 0.0800 Dist. urban 0.2300 0.1667 0.1800
Prox. forest 0.0500 0.1667 0.0800 Dist. roads 0.2300 0.1667 0.1800
Prox. core 0.3400 0.1667 0.4000 Prox. forest 0.1300 0.1667 0.1600
Prox. core 0.3300 0.1667 0.4000
Total 1.0000 1.000 1.0000
Total 1.0000 1.000 1.0000
Risk-taking 0.6200 0.5000 0.4000
Tradeoff 0.7000 1.0000 0.7800 Risk-taking 0.5400 0.5000 0.4600
Tradeoff 0.7800 1.0000 0.7300
Vuln.: vulnerability to erosion; Dist. urban: distance from urban areas; Dist. roads:
distance from roads; Prox. water: proximity to surface water; Prox. forest: Vuln.: vulnerability to erosion; Dist. urban: distance from urban areas; Dist. roads:
proximity to forest patches; proximity core: proximity among forest patches with distance from roads; Prox. forest: proximity to forest patches; proximity core:
the larger core areas. proximity among forest patches with the larger core areas.
1414 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417
situations the matrix had to be re-evaluated. In the definition of the
order weights, the weight assigned to the excluded criterion was
distributed among the other criteria, independently of the risk-
taking in the process. The alternatives proposed by the elimination
of criteria were analyzed to determine the importance and
influence of each criterion in the process (Table 3). The sensitivity
analysis indicated a new scenario (solution) to define priority areas
for forest conservation in the Corumbataı́ river basin. In order to
evaluate this scenario, maps were produced with low, medium,
and high risk-taking (Table 4).
3. Results and discussion
3.1. Priority areas for forest conservation by aggregation of all criteria
The evaluation of alternatives associated with different risk-
taking indicated that the best solution for priority areas was the
one associated to high risk-taking (R = 0.40) and tradeoff = 78%
among criteria (Fig. 5). According to Malczewski (1999), the
best solution is not always the one with the smaller risk. One
should evaluate the characteristics of the landscape and the
proposed solutions in order to determine the best alternative. In
that map (Fig. 5), one can notice an improvement in the shape and
connectivity among the forest patches with the largest core areas.
Restoration of these regions will result in an increase in the forest
connectivity of the whole landscape (Geneletti, 2004b). The
importance weight (0.3462) and the order weight (0.4000)
assigned to the factor proximity among forest patches with the
larger core area (Table 1) were responsible for this connectivity
Fig. 5. Priority areas for forest conservation: aggregation every criteria, in the Corumbataı́ river basin, São Paulo State, Brazil.
among the patches. On the other hand, this alternative leads
to some high-priority areas mainly in function of the factor
proximity to surface water, regardless the presence of forest
patches (Fig. 5, zoomed region). One cannot disconsider this
influence of proximity to surface water, however, if we have in
mind the objective and the importance of the criteria for the
decision-making process, this alternative promotes a better
solution than those proposed by maps associated to medium
and low risk-taking. In this alternative (best solution), the weights
associated to the criteria proximity to forest patches, distance
from roads, distance from urban areas, and vulnerability to
erosion, also promoted the identification of priority areas (except
to areas suffering the influence of proximity to surface water)
according to the perspective that was adopted to this process.
The map associated to medium risk-taking (i.e. order weights
assuming equal values for all criteria) represents the alternative
corresponding to the conventional weighted linear combination
(multicriteria aggregation operation). This alternative is charac-
terized by risk = 0.5 and a full tradeoff among criteria. Under this
alternative, most of the basin area should be considered of high-
priority for forest conservation. This condition reflects the
excessive influence of the criteria: proximity to surface water
and proximity to forest patches (the first two in the criteria
ranking). This influence is related to the distribution of criteria
features along the basin.
The alternative that represents low risk-taking (risk = 0.62 and
tradeoff = 70% among criteria) did not promote the definition of
priority areas in accordance with the objective of this study. This
map showed the excessive influence of the criteria distance from
 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417
roads and distance from urban areas (the lowest in the criteria
ranking), as a consequence of high tradeoff (low importance
weights and high order weights) associated to these criteria.
Comparing this result with the ‘‘best solution’’ there was an
increase of areas assigned as low-priority.
3.2. Sensitivity analysis
The sensitivity analysis showed that the criteria distance from
roads; distance from urban areas; vulnerability to erosion;
proximity among forest patches with the larger core areas; and
proximity to forest patches are important to the definition of
priority areas in the Corumbataı́ river basin, and that the weights
associated to them are in accordance with the objective of this
decision-making process. On the other hand, this analysis leads to
the elimination of the criterion proximity to surface water as a new
scenario in the definition of priority areas for forest conservation
(see criteria weights in Tables 3 and 4). Under this perspective, the
best alternative was the map (Fig. 6) associated to medium–high
risk-taking (R = 0.46; tradeoff = 73% among criteria—Table 4).
The surface water is used by several animal species. Thus,
vegetated corridors along streams and rivers have an ecological
importance in the dynamics of the landscape, allowing the
movement of fauna along the landscape, what contributes to
plant dispersion, helping the process of forest connectivity (Harper
et al., 1992; Forman and Collinge, 1997). In the Corumbataı́ river
basin, the main problem related to the prioritization of areas as a
function primarily of the proximity to surface water, resides in the
fact that in some regions of the basin, rivers and streams are not
Fig. 6. Priority areas for forest conservation: aggregation of criteria without proximity to surface water, in the Corumbataı́ river basin, São Paulo State, Brazil.
1415
accompanied by their respective riparian forests. Another situation
comprises those areas with a few water bodies and several small
forest patches.
The best alternative (without proximity to surface water)
resulted in criteria and weights in accordance to the objective of
this decision-making process. In this map the highest-priority
classes are located close to forest patches, mainly to those with the
largest core area (Fig. 6, zoomed region); and in regions more
sensitive of the basin (high values of slope and erodibility) and,
when it is possible, distant from sources of disturbance (e.g. roads
and urban areas). Besides allowing the connectivity among forest
fragments, the factor proximity to forest patches, together with the
factor proximity among forest patches with larger core areas,
collaborated to the definition of most of the areas with the higher
priority in those regions of the Corumbataı́ river basin concentrat-
ing the larger areas of native forest. Such regions (sub-basins of
Passa-Cinco river and Alto Corumbataı́), according to Valente and
Vettorazzi (2002), also present the best forest fragments (larger,
close to each other, with the best shape and larger core area). The
forest patches larger than 60 ha, in the Passa-Cinco sub-basin, are
the best conserved and preserved in the whole river basin, based
on their physiognomic characteristics (Azevedo et al., 2003).
Malczewski et al. (2003), in the context of land-use suitability
analysis for identifying the priority areas for rehabilitation in one
watershed in Canada, mentions that the restoration activities
should start in those areas close to native vegetation.
According to the final map (best solutions—Fig. 6), for those
cases where it is not possible, or at least difficult, to have the
largest patches connected, as in some regions of this study area, the
1416 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417
connectivity can be established through small patches (stepping
stones). Thus, the importance of the factor proximity to forest
patches to the final map of priorities can also be highlighted.
Proximity to forest patches has been commonly used as one of the
criteria to define areas for forest restoration (Trani and Giles, 1999;
Kindvall and Petersson, 2000; Geneletti, 2006).
Another characteristic of the final map (Fig. 6) is the continuity
of its classes. The highest-priority class (class 5) is connected to the
previous one (class 4) and so on. This is related to the continuous
way how the criteria are represented in the multicriteria approach
and, over all, to the weights associated to the factors.
Regarding other alternatives of this scenario, associated to
medium and low-medium risk-taking, their evaluation demon-
strated that they did not lead to the definition of priority areas in
accordance with the objective of this study. The map associated to
medium risk-taking (risk = 0.5 and a full tradeoff among criteria)
provided the definition of priority areas mainly in function of the
proximity to forest patches. This is related to the criterion
importance (i.e. importance weight; Table 3) for the process; the
full tradeoff among criteria; and the criteria ranking (i.e. influence of
criteria on the process). The proximity to forest patches was the first
criterion in the criteria ranking (scenario without proximity to
surface water) followed by proximity among forest patches with
the larger core areas; vulnerability to erosion; distance from
urban areas; and distance from roads. Under this perspective, the
alternative with medium risk-taking was not in accordance with the
criteria importance to this study, which highlighted that larger and
connected ecosystems are typically better for conserving biodiver-
sity than the smaller and isolated ones (Noss et al., 1997). This is
because the first group of patches tends to host a higher number of
species, and populations less subject to extinction with respect
to the second group (Noss et al., 1997; Farina, 1998). The map
associated to medium risk (i.e. order weights assuming equals
values to all criteria; full tradeoff among criteria) represents the
alternative falling midway in between the extremes AND/OR (Fig. 2),
and consequently it is not a risk-aversion or risk-taking solution
(Malczewski, 2000).
In order to obtain the map with medium-low risk-taking
(R = 0.54; tradeoff = 78%), it was necessary the high tradeoff (low
importance weights and high order weights) of the criteria
distance from urban areas and distance from roads. The
consequences were (i) the excessive influence of these criteria
on the decision process; and (ii) the increase of areas assigned as
low-priority, when compared to the best alternative. Thus, regions
of the Corumbataı́ river basin concentrating the larger areas of
native forest were assigned as medium or high-priority class; and
this alternative showed some highest-priority areas without the
presence of forest patches and forest patches with the larger core
areas. Actions of forest conservation must go, therefore, in the
opposite direction to fragmentation (Geneletti, 2004b). Thus, the
maintenance and/or the restoration of the biodiversity of the
affected landscape will be guaranteed. The forest remnants and/or
the proximity to them, contribute to the reestablishment of natural
integrity at the landscape level.
4. Conclusions
This paper presents an application of the multicriteria evalua-
tion, through the ordered weighted averaging (OWA) method, for
the definition of priority areas for forest conservation. The method
is flexible, easy to be implemented and, mainly, it allows to take
into consideration decision makers/expert’s opinions and char-
acteristics of the landscape, in the decision-making process.
Consequently, the method facilitates a better understanding of
the alternative land-use suitability patterns. The selection of the
multicriteria evaluation method is an important problem because
the alternative identified as the best one depends on the decision
rule used in the decision-making procedure. In other words, even if
the decision problem has been perfectly structured, the use of an
unsuitable method can lead to decisions that cannot be justified.
This paper demonstrates how one could obtain a range of alterna-
tives (solutions) by applying OWA, including the one obtained
by the weighted linear combination method. In some decision
situations the multicriteria methods could be combined. The paper
is an example of the use of the analytic hierarchy process (AHP) to
structure the decision problem and to assign the importance to
each criterion.
The sensitivity analysis validates the decision-making process.
The coherence among the criterion importance weights, the
order weights, and the solution, adds robustness to the process.
According to this analysis, among the proposed criteria set only the
proximity to surface water is not important to identify priority
areas for forest conservation, in the case of the Corumbataı́ river
basin. Without this criterion, the highest-priority areas for forest
conservation are located close to forest patches, mainly to those
with the largest core area, in regions more sensitive of the basin,
and, when it is possible, distant from sources of disturbance.
Finally, we can highlight that the prioritization of land for
conservation leads to the optimization of resources and actions. In
this context, it is possible to identify the most appropriate land-use
spatial pattern according to specific requirements, preferences, or
predictors of some objective or activity. This is because the process
considers the intrinsic characteristics (importance and influence)
of the landscapes.
Acknowledgements
We thank the Fundação de Amparo à Pesquisa do Estado de São
Paulo/FAPESP for the scholarship (proc. 02/03026-6) granted to the
first author. The authors are grateful for helpful comments from
two anonymous reviewers.
References
Azevedo, T.S., Manzatto, A.G., Ferreira, M.C., 2003. As perturbações ambientais
sofridas pelos fragmentos de matas ciliares no setor da alta bacia do rio Passa
Cinco, Ipeúna-SP: uma abordagem baseada em ecologia da paisagem e car-
acterização fisionômica da vegetação. In: Claudino-Sales, V., Tonini, I.M.,
Dantas, E.W.C. (Eds.), Proceedings of the 6th Congresso de Ecologia do
Brasil, vol. 3, Brazil, pp. 87–88.
Baskent, E.Z., 1999. Controlling spatial structure of forested landscapes: a case study
towards landscape management. Landsc. Ecol. 14, 83–87.
Boroushaki, S., Malczewski, J., 2008. Implementing an extension of the analytical
hierarchy process using ordered weighted averaging operator with fuzzy.
Comput. Geosci. 34 (4), 399–410.
Bouyssou, D., Jacquet-Lagrez, E., Perny, P., Slowinski, R., Vanderpooten, D., Vincke,
Ph., 2001. Aiding Decisions with Multiple Criteria. Kluwer Academic Publishers,
Netherlands.
Calijuri, M.L., Melo, A.L.O., Lorentz, J.F., 2000. Identificação de áreas para implan-
tação de aterros sanitários com uso de análise estratégica de decisão. Inform.
Pública 4 (2), 231–250.
Chen, K.P., Blong, R., Jacobson, C., 2001. MCE-RISK: integrating multicriteria evalua-
tion and GIS for risk decision-making in natural hazards. Environ. Model. Softw.
16 (4), 387–397.
Collins, M.G., Steiner, F.R., Rushman, M.J., 2001. Land-use suitability analysis in the
United States: historical development and promising technological achieve-
ments. Environ. Manage. 28 (5), 611–621.
Eastman, J.R., 2001. IDRISI 32: guide to GIS and image processing. Clark Labs, Clark
University, Worcester.
Eastman, J.R., Jiang, H., 1996. Fuzzy measures in multi-criteria evaluation. In:
Proceedings of the 2nd International Symposium on Spatial Accuracy Assess-
ment in Natural Resources and Environmental Studies, Colorado, pp. 527–534.
Farina, A., 1998. Principles and Methods in Landscape Ecology. Chapman & Hall,
London.
Forman, R.T.T., Collinge, S.K., 1997. Nature conserved in changing landscapes with
and without spatial planning. Landsc. Urban Plan. 37, 129–135.
 R.O.A. Valente, C.A. Vettorazzi / Forest Ecology and Management 256 (2008) 1408–1417
Geneletti, D., 2004a. Using spatial indicators and value functions to assess ecosys-
tem fragmentation caused by linear infrastructures. Int. J. Appl. Earth Obs.
Geoinf. 5, 1–15.
Geneletti, D., 2004b. A some common shortcomings in the treatment of impacts of
linear infrastructures on natural habitat. Land Use Policy 21, 149–160.
Geneletti, D., 2006. A GIS-based decision support system to identify nature
conservation priorities in an alpine valley. Environ. Impact Assess. Rev. 26,
257–267.
Gkaraveli, A., Good, J.E.G., Williams, J.H., 2004. Determining priority areas for native
woodland expansion and restoration in Snowdonia National Parks, Wales. Biol.
Conserv. 115, 395–402.
Gutzwiller, K.J., Barrow Jr., W.C., 2003. Influences of roads and development on bird
communities in protected Chihuahuan Desert landscapes. Biol. Conserv. 113,
225–237.
Harper, K.T., Sanderson, S.C., McArthur, E.D., 1992. Riparian ecology in Zion National
Park, Utah. Int. Gen. Tech. Rep. 298, 32–42.
Jiang, H., Eastman, J.R., 2000. Application of fuzzy measures in multi-criteria
evaluation in GIS. Int. J. Geogr. Inf. Sci. 14 (2), 173–184.
Kangas, J., Kangas, A., 2005. Multiple criteria decision support in forest manage-
ment—the approach, methods applied, and experiences gained. For. Ecol.
Manage. 207, 133–143.
Kangas, J., Store, R., Leskinen, P., Mehtätalo, L., 2000. Improving the quality of
landscape ecological forest planning by utilizing advanced decision-support
tools. For. Ecol. Manage. 132, 157–171.
Keeney, R., 1996. Value-focused thinking: identifying decision opportunities and
creating alternatives. Eur. J. Oper. Res. 92, 537–549.
Kindvall, O., Petersson, A., 2000. Consequences of modelling interpatch migration as
a function of patch geometry when predicting metapopulation extinction risk.
Ecol. Model. 129, 101–109.
Lathrop Jr., R.G., Bognar, J.A., 1998. Applying GIS and landscape ecological principles
to evaluate land conservation alternatives. Landsc. Urban Plan. 41, 27–41.
Makropoulos, C.K., Butler, D., 2006. Spatial ordered weighted averaging: incorpor-
ating spatially variable attitude towards risk in spatial multi-criteria decision-
making. Environ. Model. Softw. 21, 69–84.
Malczewski, J., 1999. GIS and Multicriteria Decision Analysis. John Wiley, New York.
Malczewski, J., 2000. On the use of weighted linear combination method in GIS:
common and best practice approaches. Trans. GIS 4 (1), 5–22.
Malczewski, J., 2004. GIS-based land-use suitability analysis: a critical overview.
Prog. Plan. 62, 3–65.
1417
Malczewski, J., 2006. Ordered weighted averaging with fuzzy quantifiers: GIS-based
multicriteria evaluation for land-use suitability analysis. Int. J. Appl. Earth Obs.
Geoinf. 8, 270–277.
Malczewski, J., Rinner, C., 2005. Exploring multicriteria decision strategies in GIS
with linguistic quantifiers: a case study of residential quality evaluation. J.
Geogr. Syst. 7 (2), 49–268.
Malczewski, J., Chapman, T., Flegel, C., Walters, D., Shrubsole, D., Healy, M.A., 2003.
GIS-multicriteria evaluation with ordered weighted averaging (OWA): case
study of developing management strategies. Environ. Plan. 35 (10), 1769–1784.
Mcneely, J.A., Miller, K.R., Reid, W.W., Mittermeier, R.A., Werner, T.B., 1990. Con-
servating the World’s Biological Diversity. The World Bank, New York.
Mendoza, G.A., Prabhu, R., 2000. Multiple criteria decision making approaches to
assessing forest sustainability using criteria and indicators: a case study. For.
Ecol. Manage. 131, 107–126.
Noss, R.F., O’Connell, M.A., Murphy, D.D., 1997. The Science of Conservation Plan-
ning: Habitat-Based Conservation under the Endangered Species. Act. Island
Press, Washington, DC.
Phua, M., Minowa, M., 2005. A GIS-based multi-criteria decision making approach
to forest conservation planning at a landscape scale: a case study in the
Kinabalu area, Sabah, Malaysia. Landsc. Urban Plan. 71, 207–222.
Randhir, T.O., Connor, R.O., Penner, P.R., Goodwin, D.W., 2001. A watershed-based land
prioritization model for water supply protection. For. Ecol. Manage. 143, 47–56.
Rosa, D., Moreno, J.A., Mayol, F., Bonsón, T., 2000. Assessment of soil erosion vulner-
ability in western Europe and potential impact on crop productivity due to loss of
soil depth using the ImpelERO model. Agric. Ecosyst. Environ. 81, 179–190.
Saaty, T., 1980. The Analytic Hierarchy Process. McGraw-Hill, New York.
Saunders, D.A., Hobbs, R.J., Margules, C.R., 1991. Biological consequences of eco-
system fragmentation: a review. Conserv. Biol. 5, 18–32.
Thill, J.C., 1999. Multicriteria Decision-Making and Analysis: A Geographic Infor-
mation Sciences approach. Ashgate, New York.
Trani, M.K., Giles, R.H., 1999. An analysis of deforestation: metrics used to describe
pattern change. For. Ecol. Manage. 114, 459–470.
Valente, R.O.A., Vettorazzi, C.A., 2002. Análise da estrutura da paisagem na bacia do
Rio Corumbataı́, SP. Sci. Forest 62, 114–129.
Vlahos, N., Herbst, J., 2000. Prioritizing watershed land for conservation: a case
study in the Hammonasset watershed, Connecticut. J. N. Engl. Water Works
Assoc. 144, 10–25.
Yager, R.R., 1988. On ordered weighted averaging aggregation operators in multi-
criteria decision making. IEEE Trans. Syst. Man Cybern. 18 (1), 183–190.