See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/220649470
Carver, S.J.: Integrating Multi-Criteria Evaluation with Geographical
Information Systems. International Journal of Geographic Information
System 5(3), 321-339
Article  in  Geographical Information Systems · July 1991
DOI: 10.1080/02693799108927858 · Source: DBLP
CITATIONS
933
READS
1,176
1 author:
Some of the authors of this publication are also working on these related projects:
Evaluating management effectiveness: Indicators for Marine Protected Areas in British Columbia View project
Pastoralists, Livestock and Wildlife Interactions View project
Steve Carver
University of Leeds
116 PUBLICATIONS   3,840 CITATIONS   
SEE PROFILE
All content following this page was uploaded by Steve Carver on 26 October 2015.
The user has requested enhancement of the downloaded file.
This article was downloaded by: [117.169.1.144]
On: 02 July 2015, At: 08:49
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place,
London, SW1P 1WG
International Journal of Geographical Information
Systems
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tgis19
Integrating multi-criteria evaluation with geographical
information systems
STEPHEN J. CARVER
a
a
NorthEast Regional Research Laboratory, Centre for Urban and Regional Development
Studies , University of Newcastle upon Tyne , Newcastle, upon Tyne, NE1 7RU, U.K.
Published online: 06 Apr 2007.
To cite this article: STEPHEN J. CARVER (1991) Integrating multi-criteria evaluation with geographical information systems,
International Journal of Geographical Information Systems, 5:3, 321-339, DOI: 10.1080/02693799108927858
To link to this article: http://dx.doi.org/10.1080/02693799108927858
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
INT. 1. GEOGRAPHICAL INFORMATION srsrssrs, 1991, VOL. 5, No.3, 321-339
Integrating multi-criteria evaluation with
geographical information systems
STEPHEN J. CARVER
NorthEast RegionalResearch Laboratory,
Centre for Urban and Regional Development Studies,
University of Newcastle upon Tyne,
Newcastle upon Tyne NEI 7RU, U.K.
AbstracL Geographical information systems (GIS) provide the decision-maker
witha powerful setoftoolsforthe manipulationand analysisofspatial information.
The functionality ofGISis,however,limitedto certaindeterministicanalysesin key
applicationareassuchasspatial search.The integrationofmulti-criteriaevaluation
(MCE) techniques with GIS is forwarded as providingthe user with the means to
evaluate various alternatives on the basis of multiple and conllictingcriteria and
objectives. An exampleapplication based on the search for suitable sites for the
disposal of radioactive waste in the UK using the Arc/Info GIS is included.The
potential use of a combined GI8-MCE approach in the development of spatial
decisionsupport systems is considered.
1. Introduction
Geographical information systems (GIS) provide the decision-maker with a
powerful set of tools for the manipulation and analysis of spatial information. The idea
of GIS as a box of tools for handling geographical data is useful. Items from the toolbox
of GIS can, in various combinations, be used to solve a multitude of problems
involving spatial data. Like most tool-boxes, however, the list oftools provided by GIS,
although impressive, is not complete. For example, in most GIS packages spatial
analytical functionality lies mainly in the ability to perform deterministic overlay and
buffer operations. Such abilities, whilst ideal for performing spatial searches based on
nominally mapped criteria, are oflimited use when multiple and conflicting criteria and
objectives are concerned. The integration of analytical techniques designed to cope
with multi-criteria problems in GIS can provide the user with a valuable addition to the
functionality of the GIS tool-box.
1.1. Site-search procedures using standard geographical information system
map overlays
The application of digital map overlays for the purpose of area screening is a classic
example of the applied use of one subset of tools from the GIS tool-box. One such
example given by Openshaw et al. (1989) is the use of Arc/Info overlay routines to
search a specified region of interest for areas suitable for the disposal of radioactive
waste. Four 'siting factors are considered: geology, population distribution, accessibility
and conservation. Numerical and qualitative criteria are applied to these siting
factors in an exploratory investigation of the search area. However, such overlay
procedures can do little more than identify areas which simultaneously satisfy all the
specified criteria. There is nothing in an analysis ofthis kind that informs the user which
individual site(s) offer the most promising characteristics for development. Additional
0269-3798/91 ssoo © 1991 Taylor & Francis Ltd.
Downloadedby[117.169.1.144]at08:4902July2015
322 S. J. Carver
procedures, based here on multi-criteria evaluation (MCE) techniques. are required to
evaluate the suitability of sites falling within the feasible areas identified using standard
GIS overlay procedures.
This paper describes one way in which MCE techniques may be integrated within a
GIS framework, with special reference to their joint use as a largely self-contained
methodology for site selection and decision support. An introduction to the basic
principles of MCE is followed by a discussion on how the two techniques may be
combined. A worked example employing an integrated GIS--MCE approach to
locating sites for the disposal of radioactive waste in the UK is described. This follows
the standard GIS map overlay procedures before applying MCE analyses to identify
individual sites. In addition, some thoughts are offered on how GIS--MCE systems may
be used in the development of effective spatial decision support systems.
2. An introduction to multi-criteria evaluation
2.1. Origins ofmulti-criteria evaluation
MCE techniques (often referred to as multi-criteria analysis or MCA) began to
emerge during the early I970sfrom a critique oftraditional neoclassical environmental
economics. A number of workers, particularly in the regional economic planning and
decision-making research fields (seeVoogd 1983,Nijkamp 1980,Cochrane and Zeleny
1973, Keeney and Raiffa 1976),have identified certain weaknesses in the neoclassical
view of decision-making and site location. A number of amendments and alternatives
to the neoclassical approach have been suggested in response to a realization that these
conventional methods cannot cope adequately with external negative spillover effects
from environmental and economic developments (e.g.pollution, health risks, planning
blight) and neeed to include more rigorous procedural elements in the planning
process. A significant proportion of these focuses on the paradigm of multi-
dimensionality.
Multi-dimensional decision and evaluation models (of which MCE is a part)
provide tools for analysing the complex trade-offs between choice alternatives (e.g.
sites, plans) with different environmental and socio-economic impacts. The formal
mathematical framework used to describe multi-diinensional decision-making is based
on multi-objective optimization theory in which both conflicting and complementary
objectives are described as a decision problem with multiple objectives. In multidimensional
models appropriate units of measurement are applied to each factor in the
problem rather than trying to impose artificial 'shadow' prices, as in many neoclassicial·
models (e.g. cost-benefit analysis).
2.2. Basic principles ofmulti-criteria evaluation
The basic aim of MCE analysis techniques is 'to investigate a number of choice
possibilities in the light of multiple criteria and conflicting objectives' (Voogd 1983,
p. 21). In doing so it is possible to generate compromise alternatives and rankings of
alternatives according to their attractiveness (Janssen and Rietveld 1990). Given the
current emphasis on site location via a process of map overlay, the problem facing
decision-makers concerns the identification of best compromise sites on the basis of an
evaluation of a finite number of choice alternatives by a finite number of attributes,
while taking into account conflicting viewsand objectives. The term 'choice alternative'
refers to any available option in the .choice set, here defined as an individual site.
The basic starting point of any MCE analysis is the construction of an evaluation
matrix, the elements of which reflect the characteristics of the given set of choice
Downloadedby[117.169.1.144]at08:4902July2015
Multi-criteria evaluation and GIS 323
alternatives on the basis of a specificset of criteria. Assuming a matrix S consisting of
I, ... I alternatives and 1,... J criteria, then element Sij describes the score of alternative
i according to criterion j. Using this notation the evaluation matrix S can be
summarized as follows:
Sll.. ..SIl
S=:
SIJ....SIJ
In many analyses, especially those utilizing quantitative and mixed sources of data,
some form of standardization or criterion scores is necessary to enable meaningful
comparisons to be made on the basis ofcriteria measured on different scales. A number
of common standardization techniques can be used to effect a normalization of
criterion scores. These are discussed in more detail later. To evaluate a matrix purely on
the basis ofstandardized criterion scores would not be very realistic as different criteria
usually have different levels of importance. A further requirement of most MCE
techniques, therefore, is the inclusion of criterion weights, WI, . ..WJ. Understandably,
in many real-world applications of MCE methods, there is often uncertainty
surrounding the validity ofthe chosen weighting schemes (and indeed, of other aspects
of the MCE problem). This willbemade clear later in the paper, but it issufficient to say
here that some form ofsensitivity analysis is required to account for uncertainties in the
definition of criterion weights.
The MCE techniques used for evaluating choice alternatives are many and varied.
A whole suite of methods has been developed for solving a range of separate multicriteria
problems. A number ofthese methods,especially those taken from the regional
economic planning field, exhibit characteristics that are particularly relevant to the
solution of complex spatial problems. However, as a general rule, most MCE
techniques have been developed for evaluating small numbers ofchoice alternatives on
the basis of only a limited number of criteria, ideally in the order of eight alternatives
and eight criteria (Voogd 1983). Many problems ofspatial search willinevitably involve
a much larger range of alternatives and criteria, necessitating the modification and
automation of MCE methods. Three MCE methods [ideal point analysis (IPA),
hierarchical optimization (HO) and concordance-discordance analysis (CDA)) are
modified here to evaluate the thousands of potential sites for the disposal of radioactive
waste. These methods are described in the following sections.
2.3. Multi-criteria evaluation techniques
Some consideration of the MCE techniques adapted here for use in a GIS
framework is required to explain how they work. The information in this section is
derived from published sources and so represents nothing new, but it is thought
desirable to provide an outline of MCE techniques before describing how they can be
implemented within a GIS framework.
Both IPA and CDA models are based on the direct comparison of alternatives by
those factors included in the evaluation. Since the various factors taken into account in
describing each of the choice alternatives are likelyto be measured in different units, the
evaluation matrix S needs to be transformed onto a normalized scale (i.e. 1= max,
O=min). As indicated above, several common standardization techniques are available.
These are described by Voogd (1983) and include methods with an additivity
constraint (i.e. the standardized scores are based on raw scores divided by the sum of
the raw scores), ratio scale properties and interval scale properties. The last method is
Downloadedby[117.169.1.144]at08:4902July2015
324 S. J. Carver
used here since this is thought to be the most appropriate for use with MCE techniques
involving pair-wise comparisons of criterion scores (e.g.CDA) and may also be used
with techniques which utilize the magnitude of individual scores (e.g. IPA). This kind of
standardization can be expressed as follows:
raw score-minimum raw score
Standardized score = . . .
maximum raw score-minimum raw score
Using this method of standardization means that the worst criterion score will always
have a standardized score ofO,and the best a score of 1.This equation is applicable only
to 'benefit' criteria (i.e.those where a higher score implies a better score). When the jth
factor is a 'cost' criterion (i.e. when a higher score implies a worse score), then the
standardization procedure becomes:
m;uSij-Slj
elj
m!lxSI}- min 5i}
I I
Weights or criterion priorities allow the decision-maker to specify the perceived
importance .of individual factors relative to the others included in the evaluation. A
number of weighting methodologies have been described (e.g. complete ranking, partial
ranking, paired comparison matrices). These are reviewed by Voogd (1983). A sevenpoint
scale, first advocated by Osgood et al. (1957), is used in the example presented
here because this is thought to be sufficient to allow the user to express adequately a
level of preference. To make the scale more meaningful a semantic differential is added
with the inclusion of descriptive labels, from very important (7) to unimportant (I).
Ideal point models are based on the deviation between a set of ideal solutions and a
set of efficient solutions (Zeleny 1976). Although the ideal solution willalmost certainly
not exist, it does serve as an important frame of reference. The best compromise
solution is defined as that which is the minimum distance from the theoretical ideal.
Distances from the ideal with respect to individual factors are weighted by the decisionmaker.
Thus increasing distance from the ideal for factors placed higher on the scale of
importance incurs a greater penalty than increasing distance from the ideal for those
placed lower. The IPA model may be written as follows:
,
mind= I !-Jj{I-ei)
j= 1
where: rnind e minimum distance from ideal solution; W;=jth criterion weight;
and ei}=standardized score.
Concordance-discordance analysis is a widely used method of MCE based on the
pair-wise comparison of choice alternatives. The degree to which choice alternatives
and factor weightings confirm or contradict the 'outranking' relationship between
alternatives is measured. Both the differences in factor weights and criterion scores are
analysed separately via concordance and discordance procedures.
Downloadedby[117.169.1.144]at08:4902July2015
Multi-eriteria evaluation and GIS 325
Aubert (1986) defines the outranking relationship for two alternatives a and a' as:
(I) a scores equal or better than a' on a sufficient number of criteria (concordance
index);and (2)the differencesin the factor scores where a'is better than a isnot too high
(discordance index).
The concordance index of alternative a with respect to a' for all decision factors
(i.e. where a scores equal to, or better than, a') is defined as:
sum of weights of the criterion for which a~a'
C...
sum of weights for all criteria
LW;c ,=a~G'
.. LW;
where: C... =concordance index for a compared with a'
The discordance index of alternative a with respect to a' for all decision factors is
defined as:
D...
D...
maximum difference between weighted scores when a< a'
maximum difference between weighted scores for the criterion yielding
the maximum difference between the weighted scores when a-c.a for
all alternatives
maxW...
..'
max(W9.·)-min(~e')
1 1
where: D... = discordance index for a compared with a'; B=argmax(W...); and I = all
alternatives.
Using a user-defined minimum concordance index and maximum discordance
index, a dominance matrix can be calculated from the standardized evaluation matrix
showing the outranking relationship of each alternative over all others. A score of I is
accumulated each time an alternative outranks one of the others. Bydirect summation
of each alternative's dominance score a total dominance index can be derived and the
alternatives ranked accordingly. The best compromise solutions are characterized by
increasingly higher total dominance indices.
Hierarchical optimization models attempt to rank all factors according to their
relative priorities. Optimization is carried out in a step-wise fashion so that higher
ranking factors are maximized before the lower ranking factors. After the relative order
of priority assigned to individual factors has been established by complete ranking
methods, then the evaluation matrix can itself be ranked, first by the factor with the
highest priority, then the second, and so on. Aftereach ranking, the evaluation matrix is
truncated according to goals or achievement levelsspecified by the decision-maker for
the factor in question. In this way choice alternatives not meeting attainment levelsare
removed from consideration. No standarization of criterion scores is required since
there is no direct comparison of disparate criteria. This technique is inherently
interactive, requiring that the user goes through a learning process of multiple model
Downloadedby[117.169.1.144]at08:4902July2015
326 S. J. Carver
runs before a satisfactory solution is found. A hierarchical model can be formalized as
follows:
Step 1... maxSI(I)
loS
Step 2... maxS2(1)
loS
SI(i)~fJSI(i?)
Step 3... maxS3(i)
loS
SI(i)~fJISI(i°)
Sii)~fJ2S2@
etc ...
where fJI = tolerance parameter indicating the maximum deviation from the optimum
SJ(IJ) allowable by the decision-maker.
Voogd (1983) summarizes some of the advantages and disadvantages of MCE
methods. Among the advantages cited are that they provide a more surveyable
classification of the problem and a better insight into the various objectives, and
increase the accountability, structure, control and considered nature of the decisionmaking
process. Among the disadvantages associated with the MCE methods are their
complex nature and the risk of their being used as a 'scientific sauce' over decisions
already made. Although it is generally accepted that the advantages gained outweigh
the disadvantages, the complexity of use and the lack of a suitable framework of
application represent significant barriers to the successful and widespread implement"
ation ofMCE methods in spatial search and general evaluation tasks. It is this problem
that this paper addresses.
3. Limitations of oyerlay analyses
GIS are often used to identify suitable areas for land development, be it for
afforestation, a new hospital or, as in the example given here, a radioactive waste
disposal facility. As was suggested earlier, however, the functionality of GIS in this
context is essentially limited to overlaying deterministic digital map information to
define areas simultaneously satisfying two or more siting criteria Few, if any. GIS
provide the user with a comprehensive set of tools for the evaluation of problems
involving multiple criteria and conflicting policy objectives. Overlays are ideal for area
screening using detenninisticaUy defined siting criteria but have certain limitations
when dealing with information ofa non-deterministic nature. These are summarized by
Janssen and Rietveld (1990) as follows:
(I) Digital map overlays are difficult to comprehend when more than four or five
factors are involved.
(2) Most overlay procedures in GIS do not allow for the fact that variables may
not be equally important .
(3) When mapping variables for overlay analyses the problem arises of how the
threshold values used therein are defined (the outcome of area screening
depends strongly on the choice of threshold values).
(4) The use of threshold values to map continuous variables, such as population
density, on a nominal basis will inevitably lead to substantial losses of
information.
Downloadedby[117.169.1.144]at08:4902July2015
Multi-eriteria evaluation and GIS 327
Ways offurther reducing the number ofchoice alternatives from a series of feasible
areas to a restricted short-list ofpotentially suitable sites and, finally, the site chosen are
therefore limited. It is not thought here to be practicable to proceed past the area
screening stage with overlay routines or other standard GIS tools. Add-on procedures
are required if site-specific evaluation analyses are to be undertaken successfully within
a GIS framework. The following description of a methodology for site-search and
evaluation is based on the combination of both GIS and MCE techniques.
4. Adding multi-eriteria evaluation to the geographical information systems tool-box
4.1. A combined GIS-MCE approach to facility location
The combined GIS-MCE approach to facility location described here can be
divided into two stages: (I) survey and (2) preliminary site identification. In the survey
stage standard G IS facilities are used to input, transform, store and manipulate digital
map data relevant to the problem. Area screening techniques are used to identify all the
potentially feasible areas in which to look for sites suitable for development. This is
achieved by overlaying relevant siting factors (e.g. population distribution) to identify
all the areas which simultaneously satisfy the specified numerical and qualitative
criteria (e.g. population density less than 500 persons per km"),The siting criteria used
in this stage ofthe analysis are often very deterministic in nature. For example, the site
is required to be within 3km of a motorway junction (i.e. an inclusion criterion) and
must be outside any conservation areas (i.e. an exclusion criterion). This allows the
decision-maker to progress from a very large number of alternatives to a smaller and
more manageable short-list in a single, well defined set of operations. GIS based
overlay routines are well suited to this kind of analysis and also allow for quick
comparative re-evaluation ('What if?' modelling) to be carried out on selected factors.
Thus, despite the deterministic nature of their application, GIS allow a degree of
flexibility to be maintained, thereby allowing survey stage siting criteria to be changed
as desired to meet particular requirements.
The survey stage can, however, only point to those areas in which all the specified
criteria are met. There is nothing in an analysis of this kind that informs the decisionmaker
which sites within the defined feasible areas offer the best combination of sitespecific
characteristics. It may be possible to combine all conceivably relevant siting
factors in a massive polygon overlay exercise and so identify which areas satisfy all the
specified criteria. It stands to reason, however, that these areas (if indeed any were to
remain!) would be very limited in extent and so would contain an unnecessarily
restrictive set of choice alternatives. One possible exception to this general rule is the
use of weighted overlay functions provided by some GIS systems (e.g. Tydac
Technologies' Spans). Such hybrid overlay techniques allow map layers and the
categories they contain to be weighted and overlayed to identify degrees ofdesirability
in the composite map, using (in the case ofSpans) an average ranking technique. This is
similar to weighted summation, one of the most basic MCE techniques.
The second stage of the proposed methodology, that of preliminary site identification,
is aimed at identifying compromise solutions from the whole range of
alternatives taken from the feasible areas identified by GIS based area screening. This
stage is operationalized using MCE techniques embedded in the GIS framework and
additional site-specific information relevant to secondary siting factors. These factors
are, unlike those used in the overlays of the survey stage, ofa non-deterministic nature
and are weighted according to their perceived level of importance. The evaluation
matrix containing site information is built using the data-handling facilities provided
Downloadedby[117.169.1.144]at08:4902July2015
328 S. J. Carver
by GIS. MCE techniques are used to identify 'best' compromise solutions on the basis
of this site-specific information and associated weights. Once a satisfactory result has
been obtained;short-listed sites can be displayed using the graphical capabilities of the
GIS.
4.2. Practical problems
From this desciption it might seem that the proposed GIS-MCE approach to site
location is simple. However, there are several points which need elaboration: how are
evaluation matrices produced from the feasible areas identified in the overlay based
survey stage; how are MCE techniques embedded in the overall GIS framework; and
how are differences in the results from applying different MCE techniques and
weighting schemes accounted for? These questions are answered in the following
section.
Data input to the MCE techniques is performed using the advanced data-handling
facilities of the GIS. In more traditional site-search procedures, preliminary site
identification is, owing to restrictions on time and resources (including information),
based on a sample ofsites taken from the feasible areas defined in the survey stage (e.g.
UK Nirex Ltd 1989). With a GIS based approach it is possible to include all the sites
contained within the feasible areas in the preliminary stage of site identification. This is
achieved here by dividing the output of feasible areas from the survey stage into sitesized
land parcels on a grid basis. The GIS are then used to compile a database of sitespecific
information using new and existing data sources (i.e. those used in the survey
stage). It is noted that the division of the feasible areas into sites on a grid basis is not
essential and other areal units may be used if available. Under most circumstances,
however, it is unlikely that appropriate parcels of data (with units of near uniform size
approximating the amount ofland required for development) would be available. The
division of feasible areas on a grid basis is therefore deemed to be the most convenient
and acceptable solution.
Site-specific information is attached to individual grid cells using point, line and
area overlays. For example, point-in-polygon routines are used to determine the
attributes of grid cells by overlaying their centroids on the relevant digital maps (e.g.
land use, soil type, relief, conservation status, or administrative boundaries). In
addition, nearest neighbour routines are used to find the distances of grid cells from
point, line and area features (e.g.distance of grid cell centroids from the nearest rail
head may be used as a proxy for accessibility). The site-specific database compiled in
this manner forms the basisof the evaluation matrix used by the add-on MCE routines.
The focus here is on using GIS as a framework for the application of MCE
techniques. The MCE techniques are in themselves only mathematical formulae with
no in-built means ofdata handling (i.e.input; output, display, analysis, reporting). Used
on their own, the successfulapplication ofMCE techniques, relies heavily upon the skill
and experience ofthe userin supplying the appropriate data, using"appropriate models
and interpreting the results correctly. This fact alone remains one of the main barriers
to their widespread use.The MCE algorithms used here are programmed externally to
the GIS in FORTRAN 77. They are run from within the GIS via system or macro
commands to analyse the contents of the site evaluation matrix held in the GIS
database. The three MCE techniques described earlier have been programmed and
implemented within the Arc/Info GIS in this way and are outlined in the next section.
Different MCE techniques, together with different weighting schemes, will inevitably
produce different results when applied to the same evaluation matrix. These
Downloadedby[117.169.1.144]at08:4902July2015
Multi-criteria evaluation and GIS 329
sources of variation may be referred to as technique and priority bias, respectively.
Technique bias results from the different ways in which MCE techniques rank
alternatives and calculate trade-offs between conflicting characteristics and objectives.
Priority bias refers simply' to how different weighting schemes specified by various
parties (possibly reflecting divergent views)affect where and how trade-offs are made
between choice alternatives. Assumingthree MCE techniques and three weighting
schemes are used in conjunction with a single evaluation matrix, then nine separate
short-lists of sites will be produced. Some form of sensitivity analysis is therefore
required to account for the variation introduced by technique and priority bias and so
to provide a single short-list of sites.
One solution is simply to identify those sites which are common to all (or most) of
the original nine short-lists. Adistance factor is also included which allows the number
ofadjacent sites to be counted within a specifiedradius. This attempts to take the effects
of spatial autocorrelation, data uncertainty and methodological differences between
MCE techniques into account. In this way it is possible to count the number of times
that a particular site and also its close neighbours appear in any of the nine short-lists.
This simple tally is used as the basis for a crude indicator of 'robustness' for each site.
This sensitivity analysis is again carried out in a FORTRAN 77 program run from
within the GIS, the final list of compromise sites being displayed using its graphical
output capabilities.
This method assumes that the definition of short-lists is an objective process. As
with most analyses, however, there are several subjective inputs, e.g. choice of MCE
model and definition of weighting schemes.The reliability of these depends very much
on the knowledge and expertise of the decision-maker. This isespecially true in the case
of the HO model, which is inherently interactive(as opposed to the 'grey' box approach
ofthe IPA and CDA models)and requires the decision-maker to go through a step-wise
procedure specifying the desired achievement levels (goals) at each stage of the
optimization process. A second approach towards sensitivity analysis, therefore, is to
allow for a more subjective'role for the decision-maker by finding and agreeing on a
final compromise solution through incremental changes to weighting schemes and
multiple model runs.
4.3. User interface
Most GIS allows the user to write macro programs to link together existing GIS
functions and run external routines (e.g. the Arc Macro Language (AML) in ESRI's
Arc/Info GIS package). The followingexample of searching for nuclear waste disposal
sites using Arc/Info makes use of AML to link external MCE routines into the GIS
framework. Using AML programs, the user interface of GI5-MCE tools can be
designed around a format similar to existing ArcjInfo commands. Alternatively, AML
programs may be used to create menu-driven interfaces between the user, the GIS and
the MCE routines. Whether the standard Arc/Info command format or menu interface
is used, by embedding the MCE routines in AML programs an apparently seamless
join between GIS and MCE can be created.
Most of the AML programs developed in this work are specificto the nuclear waste
example and its associated databases, but it isenvisaged that more general tools could
evolve from these and across a range of other GIS packages. These may, in the case of
Arc/Info, appear similar to existing add-on packages which surround the core Arc
package (e.g. Arcplot, Arcedit, Tin). Creating an evaluation matrix for input and
evaluation using MCE tools is very problem-specific. The contents of the matrix
Downloadedby[117.169.1.144]at08:4902July2015
330 S. J. Carver
depend on how the problem is regarded and what criteria are considered relevant.
However, the evaluation matrix itself is stored as an Arc(lnfo point coverage with the
characteristics of each point or site stored as attributes in the point attribute table
(PA1"). Compiling point coverages and adding attribute information from other
coverages and databases to the coverage PAT is common practice in Arc/Info using
standard tools which should be familiar to all users. No new add-on tools are therefore
considered necessary for the construction of evaluation matrices. Once such an
evaluation matrix has been created, then the MCE tools used can be entirely generic
with inputs from the user stating which matrix is to be evaluated and by which method,
and the criterion priorities to be used.
The times taken for some of the more complex MCE models (e.g. concordancediscordance
analysis) to run are fairly long (depending on the size of the evaluation
matrix), and as such are perhaps more suited to batch rather than interactive
processing at a terminal or work station. Simpler models, such as ideal point analysis,
can easily be run with fairly large evaluation matrices (e.g. 5878 by 16 as used in the
following example) over just a fewminutes, depending on the hardware configuration.
This is, of course, an important consideration to be taken into account when designing
any GIS-MCE interface, but as GIS work stations become ever faster and more
powerful, this may cease to be a significant problem.
S. Searching for nuclear waste sites: an example application
Openshaw et al. (1989)discuss the use ofthe Arc/Info GIS to identify feasible sites
for the disposal of radioactive waste in the U.K. The GIS operations therein constitute
little more than a sequence of map overlays in the form of a Boolean search. This
process is re-examined here with the addition of MCE techniques to the overall GIS
based analysis.
The object of the overlay analysis is to determine which areas of the U.K. are
potentially suitable for the deep geological disposal of low and intermediate level
radioactive wastes using the four siting factors given in the Introduction (geology,
population distribution, accessibility and conservation),
The numerical and qualitative criteria applied to these four siting factors are:
(I) The site must be within an area characterized by a suitable deep hydrogeological
environment.
(2) The site must be located in an area with a population density less than 490
persons per km",
(3) The site must be within 3km of the rail network.
(4) The site must not be within a designated conservation area.
Using these factors, mapped according to the four criteria, it is possible to define all
the potentially feasible areas using the basic overlay and buffer, functions in Arc/Info.
These are, however, only feasible areas, and actual sites need to be identified before
development can take place. Here, feasible areas resulting from the overlay analyses are
'rasterized into 2 x 2 km grid cells. This is the approximate size of site (i.e., 400 hal
considered by the U.K. nuclear industry to be necessary for development to take place
(UK Nirex Ltd. 1989). Each of these grid cells represents a single site-sized land parcel
for the purposes of further analysis. There are in total 5878 such sites in this example.
The data handling capabilities of Arc/Info are then used to create an evaluation matrix
in the Info database using new and existing data sources on non-deterministic
secondary siting factors. Sixteen separate factors are considered (table I). The
Downloadedby[117.169.1.144]at08:4902July2015
Multi-criteria evaluation and GIS 331
evaluation matrix isoutput from the Info database as an ASCII fileand normalized for
use with the IPA and CDA models. Three different hypothetical weighting schemes are
defined and assumed for the purposes of this work to berepresentative of the views of
the three major interest groups. These emphasize site remoteness, accessibility and
ecological suitability, respectively (table 2). The three MCE techniques are run
interactively to identify the 25 best compromise sites for each of the three weighting
schemes. These are analysed for coincidence and adjacency to identify 'robust' sites.
Site data are then displayed using the map-drawing facilities of Arc/Info.
Figure I shows the feasible areas identified using overlay analyses in the survey
stage. From these areas 5878 sites are derived. Figure 2 shows the distribution of all
nine sets of the top 25 compromise sites identified in the application of three MCE
routines with three different criterion weighting schemes. The rather confusing scatter
of potential sites in this map demonstrates the combined effect of technique and data
uncertainty on the results from the nine MCE model runs and exemplifies the need for
sensitivity analyses to filter out truly robust locations. Attention is drawn to the
apparent clustering of 'best' sites which indicates some degree of homogeneity in the
results. Technique and priority bias are investigated here by keeping one source of bias
constant and altering the other.
Table 1. Evaluation criteria for sites for the disposal of radioactive waste.
Site No. Description
jI Host rock type
j2 Hydrogeological environment
j3 Coastalenvironment
j4 Population density
j5 Population potential/accessibility
j6 Proximity to built-uparea boundary
j7 Strategic accessibility
j8 Proximity to existing nuclearinstallation
j9 Proximity to railway line
jlO Proximity to motorway
jII Availability of on-siterail access
j12 Quality of on-site road access
j13 Proximity to conservation area
jl4 Proximity to conservation site
jl5 Land-use type
jl6 Political marginality
Table 2. Hypothetical criterion weights for the nuclear industry, the general public and
environmental groups.
Criteria
Views jl j2 j3 j4 j5 j6 j7 j8 j9 jlO jll jl2 j13 jl4 jl5 jl6
Nuclear industry 7 6 5 0 0 0 7 5 '5 5 3 3 0 0 0 I
General public 0 0 2 5 7 3 0 0 0 0 0 0 0 0 0 0
Environmentalists 0 0 0 4 5 3 0 6 0 0 '0 0 7 6 3 0
Weights specified on seven-point scale with key to criteria in Table 1.
Downloadedby[117.169.1.144]at08:4902July2015
332 S. J. Carver
Figure I. Feasible areas for the disposal of radioactive waste using GIS overlay analyses.
Downloadedby[117.169.1.144]at08:4902July2015
Multi-criteria evaluation and GIS
< .
~
Figure 2. Distribution of all compromise sites. (6) Compromise site.
333
Downloadedby[117.169.1.144]at08:4902July2015
334 S. J. Carver
Figure 3. Technique bias: nuclear industry preferred site. MCE technique. ~ IPA; ~ HO;
and ~ CDA.
Downloadedby[117.169.1.144]at08:4902July2015
Multi-criteria evaluation and GIS 335
Figure 4. Priority bias: ideal point analysis model. Weighting scheme. I!l Nuclear industry;
Q:I general public; and Ql environmental groups.
Downloadedby[117.169.1.144]at08:4902July2015
336 s. J. Carver
Figure 5. Robust sites: upper quartile of 'robustness' index distribution for all model runs.
Symbols represent robust sites in upper quartile.
Downloadedby[117.169.1.144]at08:4902July2015
Multi-criteria evaluation and GIS 337
Figure 3 illustrates the effectof technique bias on the distribution of results using
the 'nuclear industry' preferred weights from table 2 The symbols used represent
groups or 'clusters' of short-listed sites and are split and shaded according to how each
appearsjn the different model runs. The variation in the distribution ofsites shows how
different evaluation techniques can affect the outcome of the site search. Similar
variations appear when testing for technique bias with the other two weighting schemes
in table 2. Although any particularly robust sites should appear when using most MCE
methods, it is suggested that two or more methods should be applied to dilute the effect
oftechnique bias. Figure 4 shows the effectof priority bias on the distribution of results
using, in this case, the IPA method. The variation in the distribution of sites shows how
different weighting schemes can affect the outcome of the site search. Again, similar
variations are evident when testing for priority bias with the other two MCE
techniques described here.
Although considerable variations are evident between the results, it is also notable
that groups of short-listed sites cluster with recurring regularity in certain areas of the
country [e.g, Anglesey, East Lothian and Caithness). Figure 5 illustrates how this is
reflected in the distribution ofthe most robust sites(i.e.those in the upper quartile of the
'robustness' index frequency distribution). It is noted that some of the additional
'scatter' exhibited by the results in figures 3 and 4 may be attributable to the limited
number ofvery robust sitesin the U.K. The most robust sitesare picked out by all or the
majority of the nine model runs and so occur in most or all of the maps. Sites of lower
robustness are relatively mediocre and so suffer more from the effectsoftechnique and
priority bias. As a result, these sites show a more scattered distribution in figures 3 and
4. It should be noted that figures 3-5 have been drawn in a diagrammatic fashion to
enhance their clarity. Assuch the location ofgroups or 'clusters' of short-listed sites are
approximate, indicating only the general areas in which they are located.
5.1. Spatial decision support systems for facility location
The advantages gained from combining GIS and MCE suggest that this approachrepresents
a major contribution towards the development of effectivespatial decision
support systems (SDSS).SDSS based around GIS-MCE systems may be used to assist
decision-makers in site identification and plan evaluation. Ideally, SDSS provide a
framework for the integration of spatial analysis, database management systems
(DBMS),graphical display, tabular reporting and the expert knowledge of the decisionmaker.
Although it is normal to find certain spatial analysis techniques, DBMS,
graphical and tabular output in GIS, only the addition of advanced spatial analysis
capabilities(e.g.MCE), the expertise of the decision-maker and a suitable user interface
can constitute an SDSS.
Many spatial problems, such as facility location, are complex (often involving
multiple criteria and objectives) and so require the use of spatial analysis models for
"their solution. It is suggested" here that the siting of nuclear installations, such as
facilities for the disposal of radioactive waste, is a typical example of a complex spatial
problem. Site-search procedures could significantly benefit from the use of SDSS, from
initial surveys to the final decision and public inquiry. It isenvisaged that a PC or work
station based GIS-MCE system and an experienced operator in a committee room
could create significant improvements in the way decisions for siting are made. In"
addition to these roles of decision-making and support, SDSS may also have an
important role to play in providing more effective means of public participation and
Downloadedby[117.169.1.144]at08:4902July2015
338 s. J. Carver
consultation throughout the site-search process by allowing for public input through
interaction with customized GI8-MCE systems and feedback to decision-makers
regarding public sentiment.
6. Conclusions
GIS overlay analyses, while ideal for performing spatial searches on nominally
mapped 'criteria,are oflimited use when multiple and conflicting criteria and objectives
are concerned.
MCE techniques offera means of making and supporting complex siting decisions,
involving multiple criteria. In this context of site location, a combined GI8-MCE
approach is advocated for use in the making of controversial decisions on siting. The
advantages of pursuing such an approach (and in conjunction with its use as the basis
for a dedicated SDSS) can be summarised as follows:
(1) GIS is an ideal means of performing deterministic analyses on all types of
geographical data.
(2) GIS provides a suitable framework for the application of spatial analysis
methods, such as MCE, which do not have their own data management
facilitiesfor the capture, storage, retrieval, editing, transformation and display
of spatial data.
(3) MCE procedures provide the GIS with the means of performing complex
trade-offs on multiple and often conflicting objectives while taking multiple
criteria and the expert knowledge of the decision-maker into account.
(4) GIS and MCE based systems have the potential to provide a more rational,
objective and non-biased approach to making decisions on siting than used
hitherto.
Much work has been carried out on the development of both GIS and MCE
techniques. but mostly in isolation. By combining the two approaches in the way
described their full potential as evaluatory and decision-making tools may be realized.
There now exists the opportunity to develop combined GIS-MCE methodologies
further by including MCE routines as add-ons to existing GIS packages. There is
currently a fundamental lack of basic spatial analysis tools in most proprietary GIS
packages over and above that provided by map overlay and bufferoperations. Little or
no means ofcarrying out an evaluation of a siting problem past the area based analyses
of map overlays is provided in the standard GIS tool-box. By the addition of MCE
functionality as suggested here, the user isfullyable to take a site-search from the basic
sieve-mapping stage, through the evaluation of a large number of potential sites, to the
final evaluation and choice between a limited number of alternatives on the basis of
information taken from all available sources (including field surveys) using standard
GIS tools. .
Integrating MCE with GIS for spatial decision-making purposes is both a
worthwhile aim and one which willbe of widespread utility to those involved in what is
(and will contine to be) a mainstay application of GIS technology. This will in turn
create the opportunity for the increased use of GIS based technology as the basis of
decision support systems. It is important "to note, however, that GIS and MCE
techniques are merely tools which provide a means to an end. Without the knowledge
and expertise of the operator and decision-maker, and without appropriate data, such
tools would be useless. Nevertheless, GI8-MCE applications appear to represent
potentially fruitful areas for further research and development.
Downloadedby[117.169.1.144]at08:4902July2015
Multi-criteria evaluation and GIS 339
References
AUBERT, P. J., 1986, The siting of nuclear power plants in the Netherlands: on the possible use of
multicriteria decision methods in energy policy. Energy Science Project, Department of
Inorganic Chemistry, State University of Utrecht, The Netherlands.
CocHRANE, 1. L.,and ZELENY, M. (editors) 1973, Multiple CriteriaDecision-making. (Columbia:
University of South Carolina Press],
JANSSSEN, R., and RIETVELD, P., 1990, Multicriteria analysis and GIS: an application to
agricultural landuse in The Netherlands. In: Geographical lnfonnation Systemsfor Urban
'andRegional Planning, edited by H. J. Scholten and J. C. H. Stillwell (Dordrecht: Kluwer).
KEENEY, R. L., and RA!FFA, H., 1976, Decisions with Multiple Objectives: Preferences and Value
Trade-off». (New York: Wiley).
NIJKAMP, P, 1980, Environmental PolicyAnalysis:Operational Methods and Mode/so (New York:
Wiley).
OPENSHAW, S., CARVER, S. J., and FERNIE, J., 1989, Britain's Nuclear Waste: Safety and Siting.
(London: Belhaven Press).
OSGOOD, C. E., Suci, G. J., and TANNENBAUM, P. H., 1957, The Measurement of Meaning.
(Urbana, IL: University of Illinois Press).
UK NIREX LTD, 1989, Deep repository project: preliminary environmental and radiological
assessment. UK Nirex Ltd., Harwell.
VOOCD, H., 1983, Multicriteria Evaluation for Urban and Regional Planning. (London: Pion).
ZELENY, M., (editor) 1976, Mulliple Criteria Decision-making. (Berlin: Springer).
Downloadedby[117.169.1.144]at08:4902July2015
View publication statsView publication stats