Multicriteria analysis and GIS application in
the selection of sustainable motorway corridor
by
Kamila Małgorzata Belka
2005-06-10
Master’s Thesis
Multicriteria analysis and GIS application in
the selection of sustainable motorway corridor
by
Kamila Małgorzata Belka
2005-06-10
ISRN-LIU-IDA-D20--05/019--SE
Supervisor: Vivian Vimarlund
Effects of functioning transportation infrastructure are receiving more and more environmental and social concern nowadays. Nevertheless, preliminary corridor plans are usually developed on the basis of technical and economic criteria exclusively. By the time of environmental impact assessment (EIA), which succeeds, relocation is practically impossible and only preventative measures can be applied.
This paper proposes a GIS-based method of delimiting motorway corridor and integrating social, environmental and economic factors into the early stages of planning. Multiple criteria decision making (MCDM) techniques are used to assess all possible alternatives. GIS-held weighted shortest path algorithm enables to locate the corridor. The evaluation criteria are exemplary. They include nature conservation, buildings, forests and agricultural resources, and soils. Resulting evaluation surface is divided into a grid of cells, which are assigned suitability scores derived from all evaluation criteria. Subsequently, a set of adjacent cells connecting two pre-specified points is traced by the least-cost path algorithm. The best alternative has a lowest total value of suitability scores.
As a result, the proposed motorway corridor is routed from origin to destination. It is afterwards compared with an alternative derived by traditional planning procedures. Concluding remarks are that the location criteria need to be adjusted to meet construction requirements as well as analysis process to be automated. Nevertheless, the geographic information system and the embedded shortest path algorithm proved to be well suited for preliminary corridor location analysis. Future research directions are sketched.
Key words: geographical information system (GIS), motorway routing, multiple criteria decision analysis (MCDA), road corridor location, sustainability.
This thesis would not be possible without help and support of many people.
Especially, I would like to acknowledge Vivian Vimarlund, deputy head of the Division of Information Systems and Management (ISM) for her fruitful supervision and words of encouragement to start, continue and finish the work. I realized that it is me who is fully responsible for the project from the beginning till the end. Furthermore, I would like to thank Jalal Maleki, director of studies at the Division of Human-Centred Systems for his constructive solutions during the difficult moments of the studies. I would also like to mention Aleksander Gumoś, my opponent, whose valuable comments let me improve my work and to thank him for all the creative and inspiring discussions that took place in the GIS lab while working on our own theses.
The thesis would not be possible without technical and merit support of the following people: Aleksander Bielicki, Chief Land Surveyor of Łódź province, Department of Geodesy and Cartography of marshal’s office in Łódź who provided spatial data used in the thesis, Dariusz Dzionek, director of Regional Centre for Geodetic and Cartographic Documentation in Łódź who was the first person to contact about spatial data accuracy and nuances of cartography, and a support in obtaining the data, all people from gleba@denali.geo.uj.edu.pl e-mail discussion list who replied to my questions, particularly Stanisław Gruszczyński from the University of Science and Technology (AGH), Kraków who assisted me in defining soil classification of the project data, as well as Cezary Kraszewski, Chief of the Division for Geotechnology and Basement Construction of the Research Institute for Roads and Bridges, Anna Clark who corrected the language, technical stuff at the Department of Computer and Information Science (IDA) without whom lab work would not be possible, and others not mentioned here by name, who provided me their helpful advice.
My great thanks go particularly to my parents and my sister for their always being with me, their patience during my long absence at home and their valuable financial support during that time. Last but not least, I would like to express my great thanks to Aleksander Gumoś, Azza Salah Eldin Ali, Mohamed Soghayroon and Vimalkumar Vaghani, my best friends, who contributed to this thesis as well as in great extent to my personal development by their generous moral support and advice. Without you, my friends, the stay in Linköping would be much less enjoyable and rewarding.
Finally, I appreciate all other people who by their thoughtful comments, kind reply, and good wish or in any other way contributed to the finishing of this work.
This work has been supported by Regional Centre for Geodetic and Cartographic Documentation in Łódź.
Wojewódzki Ośrodek Dokumentacji Geodezyjnej i Kartograficznej w Łodzi
1. INTRODUCTION... 1
1.1. Background... 1
1.2. Study area description... 1
1.3. Political declarations and strategies... 3
1.4. Research objectives ... 4 1.5. Restrictions ... 4 1.6. Terminology ... 5 2. LITERATURE REVIEW... 6 3. THEORY... 9 2.1. What is GIS?... 9 2.2. Functions of GISs ... 11 2.3. Decision making ... 16
2.4. Multiple criteria decision making – an overview ... 18
2.4. Multi-criteria decision making and GIS ... 20
4. METHODOLOGY... 32
4.1. Data collection ... 32
4.2. Geographic data and its quality ... 32
4.3. Software, data model and technical support ... 35
4.4. Assumptions for motorway corridor criteria ... 36
4.5. Analysis framework... 36
5. RESULTS... 44
5.1. Criterion maps ... 44
5.2. Cost surface raster and least-cost path... 47
5.3. Comparison of least cost path and A2 corridor ... 50
6. DISCUSSION ... 54 7. CONCLUSIONS ... 57 8. REVISION OF RESULTS... 58 9. FURTHER RESEARCH... 59 REFERENCES... 60 APPENDICES... 64
Table 1.1. Area of different land use types at the study area ... 2
Table 3.1. Comparison of raster and vector data types in different tasks ... 10
Table 3.2. Example of strait rank weighting procedure ... 27
Table 3.3. Assessing weights by ratio estimation procedure ... 27
Table 3.4. Illustration for pairwise comparison method ... 28
Table 4.1. Datasets used... 34
Table 4.2. Soil quality and land use scoring and its standardization results ... 40
Table 4.3. Geotechnical soil sort and its suitability for road construction... 41
Figure 1.1. Study area location and administrative borders ... 2
Figure 1.2. Land cover and A2 right-of-way plans, section near Łódź... 3
Figure 2.1. The three-stage and sustainability approaches to the infrastructure planning (adapted from Rapaport and Snickars 1998)... 7
Figure 3.1. Node/link representation of raster... 14
Figure 3.2. Procedure to calculate least-cost path in raster-based GIS ... 15
Figure 3.3. Conceptual visualisation of gateway function ... 16
Figure 3.4. A general model of MCDM (after Jankowski 1995) ... 19
Figure 3.5. Spatial multicriteria evaluation: a) binary overlaying; b) multiple values overlaying; c) multiple values weighted overlay ... 21
Figure 3.6. Spatial multicriteria analysis in GIS after Malczewski (1999), modified ... 22
Figure 3.7. Score range procedure in GIS ... 25
Figure 3.8. Simple additive weighting method performed in GIS on raster data ... 29
Figure 4.1. Soil map’s attributes ... 33
Figure 4.2. Positional accuracy of RSIP and SOILS datasets, buildings and built-up areas ... 34
Figure 4.3. Hierarchical structure of evaluation criteria ... 37
Figure 4.4. High quality of soils data layer (vector). In the picture the roads are not mapped, giving ‘No Data’ value while converted to raster... 39
Figure 4.5. Fraction distribution of agricultural mechanic soil groups shown on the Feret’s triangle and construction suitability classes assigned to them ... 42
Figure 5.1. Land use and soil productivity factor ... 44
Figure 5.2. Soils construction suitability classes and its scores ... 45
Figure 5.3. Distance to buildings ... 45
Figure 5.4. Distance to protected areas ... 46
Figure 5.5. Feasible alternatives... 46
Figure 5.6. The cost surface raster with the least-cost corridor ... 48
Figure 5.7. The cost surface with relation to forests ... 48
Figure 5.8. Cost surface and buildings ... 49
Figure 5.9. Planned A2 motorway and the least-cost path on the cost surface ... 50
Figure 5.10. Graphical comparison of the two corridors in respect to distance to buildings... 51
Figure 5.11. Comparison of the two corridors in respect to distance to protected areas... 52
Figure 5.12. Graphical comparison of the soils intake... 52
Appendix 1. Example calculation of standardized scores under maximum score and score
range procedures ... 64
Appendix 2. List of persons interviewed during the beginning phase in Poland... 64
Appendix 3. Findings of the interviews ... 65
Appendix 4. Data sources available prior to the analysis (full list) ... 68
Appendix 5. Specification of UWPP 1965 coordinate system for zone 1 ... 68
Appendix 6. Attributes of soils datasets, its English translations and Polish equivalents ... 69
Appendix 7. Geotechnical soil groups (English translation of their Polish equivalents) ... 71
Appendix 8. Data preparation in ArcMap and ArcCatalog, part 1: vector ... 72
Appendix 9. Data preparation in ArcMap, part 2: raster... 74
Appendix 10. Analysis, part 1: assigning scales ... 76
Appendix 11. Analysis, part 2 and part 3: weighting and least-cost path ... 77
Appendix 12. Coding of some attribute categories of soils in the reclassification process ... 78
Appendix 13. Agricultuaral and geotechnical soil classifications ... 79
Appendix 14. Assignment of suitability classes... 80
1. INTRODUCTION
1.1. Background
The following thesis is a concluding part of the Master’s Program in Geoinformatics taken at the Department of Computer and Information Science at Linköpings University, Sweden. Capabilities of geographic information systems (GISs) to be applied to a motorway corridor location study were demonstrated on the case of two communes in central Poland.
In recent years, transportation operations have become an environmental issue, mainly in the highly developed countries. Air and noise pollution, biodiversity loss, and human health deterioration are some of the urgent problems brought about by increased transportation needs. Polish course towards economic development relies to great extent on motorway construction as a main driving force. A need for transportation system improvement is recognized by the national road investment program, which stipulates among others construction of new motorways and ring roads around major cities (MI 2003, 2005a and 2005b). Parts of the national road network will be included into trans-European transport network (TEN) (EC 1996). Due to its geopolitical location Poland has the potential to serve as a transit country towards East European markets. All these herald traffic increase in the near future.
Environment of Poland is comparatively little transformed by human. It inhabits large diversity of animals and plants and appeals with unique landscapes. Almost one third of the country is covered by forests, from which more than 30% is protected. Together with small scale family farms they form functioning ecosystems. This fragile equilibrium is easy to be broken under the pressure of extensive urbanisation, spatial infrastructure development and industrial growth. Construction of motorways seems to be inevitable nowadays but economic benefits should not overwhelm environmental cost. Wise and careful planning can reduce harmful consequences. One of the strategies is to relocate the investment into less susceptible places with respect to effective natural resource management. This is accompanied by latest technologies and good environmental practices.
1.2. Study area description
The study area is located in the central Poland and covers area of two communes (gmina), namely Stryków and Zgierz, both belonging to the district of Zgierz (powiat zgierski). The total area equals above 360 km2. North-south extension is about 23 km and east-west extension about 33 km. From the south the two communes are neighbouring Łódź agglomeration of around 900 thousands inhabitants. There are two small towns in the region, Zgierz and Stryków. Both of them are the local governments’ seat. The former serves also as a regional government office. Apart from that there are smaller villages and settlements distributed all over the region. Built-up area takes about 18 km2 which is 5% of total area.
Figure 1.1. Study area location and administrative borders
Southern part of the land is dominated by fertile soils. Consequently, it is mainly devoted to agriculture. The total agricultural ground is around 272 km2, which is more than 75%. Figure 1.2 shows that the agricultural landscape dominates over both communes. It consists of small scale, low productive family farms which form a diverse mosaic of fields with green belts dividing them.
Middle and western part of the study area is covered with forests, which take up to 65 km2 constituting 18% of the total land. There are few bigger state-owned and a number of smaller privately owned forests dispersed all over the region. Apart from that, there are few nature reserves and a part of the Lodz Upland Landscape Park (Park Krajobrazowy Wzniesień Łódzkich) whose majority belongs to the neighbouring commune. Nevertheless, the mosaic of small agricultural fields contributes to the biodiversity of the flora and fauna.
Table 1.1. Area of different land use types at the study area
Land use type Area [km2] Area [%]
Agricultural land 272.16 75.29 Forests 65.50 18.12 Built-up area 18.45 5.10 Other land 3.55 0.98 Water 1.84 0.51 Sum 361.51 100.00
Figure 1.2. Land cover and A2 right-of-way plans, section near Łódź
Hydrology is limited to several small lakes and ponds and a network of little brooks and canals previously transformed by humans. The water reservoirs do not exceed 2 ha of the area. Topographically, the area is rather flat with small elevation differences, which makes roads building relatively easy and cheap.
The study area exhibits a typical Polish lowland settlement’s type. It is covered with a number of small spatially dispersed villages. Partly, the houses are newly constructed villas usually seen in the suburbs. Dispersion of settlements and population density makes it difficult to avoid close vicinity to houses and to reach consensus without thorough field survey and public involvement. Land ownership is very complex nowadays in Poland. Even if efforts are being made to regulate ownership situation, it is still a hindrance to many development processes. As a result, it is sometimes a case that a small parcel of land is owned by many persons, or the legal rights are not properly documented (Bielicki 2004). For these reasons the construction of roads is exposed to additional costs connected to the buy-out of the land and costly legal procedures.
1.3. Political declarations and strategies
A draft of the National Transportation Policy for the years 2005 – 2025 recognizes the need of the quality improvement of the transportation system in Poland in accordance with sustainable development. The meaning of sustainability is understood there, not like in this thesis though (see chapter 1.6), as a balance between social, economic, spatial and
environmental aspects of the market economy development. The development of the network of motorways and expressways is one of the priorities of this transportation policy. The document points out which section of motor- and expressways as well as local ring roads are to be constructed in the near future.
The Strategy of Development of Transport Infrastructure for the years 2004-2006 and beyond prepared by the Ministry of Infrastructure (MI 2003) stipulates building the A2 motorway between Warsaw and western Polish border. The section from Emilia to Stryków, which passes through the case study area, should be finished by the year 2006. The project of the
Strategy of Transport Development for the years 2007-2013 (MI 2005) also envisages further
development of the motorway’s and expressway’s network in Poland. According to the common practice, political and strategic statements result in a follow up at regional level. Currently, the A2 corridor in the district of Zgierz has been designed and particular plans for the A2 motorway are presented in Figure 1.2. The motorway is assumed to pass close to Stryków from its southern part and will encompass with larger radius Zgierz. Furthermore, it avoids all major forested areas but there were some social and environmental disagreements regarding the landscape park.
1.4. Research objectives
The aim of the study is to demonstrate the possibility of using a geographic information system (GIS) as a decision support tool in the strategic infrastructure planning, specifically in a phase of proposing a road corridor location. Particularly the tasks are scheduled to:
− propose a set of evaluation criteria for a preliminary motorway corridor location that meets sustainability objectives,
− combine the selected evaluation criteria in order to derive general preference for every location in the study area with help of multicriteria evaluation method, and − propose a least-cost path connecting two predefined locations, based on the
proposed criteria
Lastly, as far as plans for a motorway passing the study area exist, it will be interesting to compare the proposed corridor with current projects in order to test how well the developed criteria meet the objectives used in the thesis.
The possible target group to use the GIS application for road corridor location is the national road administration agencies as well as regional governmental authorities which are in charge of preparing land management plans and facing the problem of wise resource management.
1.5. Restrictions
The corridor proposal will be restricted to the strategic level of planning. Therefore engineering design requirements should be specified at more advanced level of planning. In the study, I will not evaluate socio-economic nor environmental effects that are likely to occur after construction of a motorway. I will in contrary try to limit the inevitable harmful effects of roads to the most vulnerable locations (human settlements and protected
environmental sites), try to efficiently manage natural resources (soil, forests and other land use types) and reduce construction cost (sort of soils).
In addition, I am far away from doing the research on selecting the best alternative from the set of proposed alignments, which is widely practised by e.g. Environmental Impact Assessment (EIA) of projects. In contrary, the approach used here aims at finding the best alignment among undefined number of alternatives under specified criteria.
In the study it is assumed that the origin and destination of the segment of the motorway is known (in the demonstration of the GIS application those points are taken arbitrarily)
Lastly the approach does not consider communication problems such as congestion (excessive traffic and limited road capacity), travel demand management, accessibility, and so on. Measures, including alternative ways of communication, effective use of existing infrastructure and alternative modes of transport, are to be considered by transportation policies which are beyond the scope of this thesis.
1.6. Terminology
The concept of sustainability is understood here in the context of the triple bottom line (or tree pillars) approach which requires the integrated consideration of environmental, social and economic issues for a sound development. This approach assumes that the development meets the criteria of the three mentioned aspects simultaneously (Gibson 2001, Pope 2003).
Multicriteria evaluation terminology introduces the term evaluation criteria, which contain both objectives and attributes. Objectives describe the purpose of change and attributes, the way the change will be measured. However the objectives are further formulated by a set of
criteria (or factors). If attributes can have numerical values they can be referred as scores.
Therefore objectives and criteria are treated in the thesis interchangeably, as well as do attributes and scores. In technical terms, the term ‘attribute’ is reserved to non-spatial information related to spatial objects in GIS.
The terms least-cost path, least-cost corridor, shortest weighted path and shortest path are used interchangeably in relation to the routing problem.
2. LITERATURE REVIEW
It is often reported that early transportation planning practice does not take environmental considerations into account. Originally, transportation plans were generated exclusively on economic, physical and technical constraints. It is sometimes even questionable if a corridor selected in this way is really the optimal one (Guimarães Pereira 1996, Jankowski and Richard 1994). Technical advancements introduced optimization techniques and over time they became more and more computationally powerful. However, most of these techniques still incorporate the traditional factors in order to minimize construction, users, maintenance and other costs (Trietsch 1987, Gipps 2001, Jha and Schonfeld 2004). All above models require that the input variables should be quantifiable. Some authors (Jha 2003, Jankowski and Richard 1994) note that it is very difficult to incorporate environmental objectives into traditional non GIS models because they are difficult to quantify. Nevertheless, environmental consciousness and knowledge made it compulsory to include environmental and social considerations into planning. The problem was challenged by many authors, among others Siskos and Assimakopoulos (1989), Rapaport and Snickars (1998), Grossardt et al. (2001), Bejleri et al. (2002), Bailey (2003) and few different approaches to the problem was developed simultaneously.
For example, the environmentalist’s model of Siskos and Assimakopoulos (1989) included such indicators as human activity, environmental protection, land production, and landscape. It uses a regression optimization model in order to assure the selection of the best environmental routings. However, the result does not consider construction or economic factors which would make the proposal unacceptable for road planners. In conclusion, the above work did not solve the problem of incorporating of environmental criteria into an economic and technical model.
Another approach was necessary in order to identify possible environmentally sensitive issues at the beginning of alternatives’ consideration stage. Such an approach, called environmental screening, was motivated by a need to avoid unnecessary delays and duplication of work during transportation planning. However, they are still relying on the technically or economically optimized alternatives (e.g. Bejleri et al. 2002).
Following the economic optimization approach (which originated from traditional road planning practice), some authors tried to incorporate environmental factors into their optimization models (Jha 2001, Jha and Schonfeld 2004, Gipps et al. 2001). Such models could incorporate costs of property acquisition, split and usability degradation assessment, as well as road intersections, river crossings and environmental penalties for violating design requirements. On the users’ side, they could consider travel-time, vehicle operations and accidents costs. Some attempts were made to incorporate criteria for environmental impacts such as floodplains and wetland crossings (Jha 2003, Jha and Schonfeld 2004). Because the optimization process requires all the variables to be brought down into a common measure (most preferably cost), it quickly turned out that many of the decision criteria (aesthetic, environmental, political, etc.) could not be effectively quantified. The best option seemed to be to generate few low cost or cost-effective optimized alternatives and leave consideration to the decision-makers (Guimarães Pereira 1996, Jha 2003). The most common approach to generating and evaluating possible alignments is genetic algorithms (GA). Following the option of few low cost alternatives the problem of its evaluation remains.
The common thought concerning the above approaches is that they rely on the environmental assessment of previously proposed alternatives. Rapaport and Snickars (1998) contrasted this well-established practice with a sustainability ideal. Nowadays, economic and socially agreed alternatives are proposed and then usually evaluated during the environmental impact assessment (EIA) iterative procedures (Figure 2.1a). The sustainability approach, proposed by the authors, imposes that the infrastructure project handles economic, social and environmental issues simultaneously (Figure 2.1b).
Figure 2.1. The three-stage and sustainability approaches to the infrastructure planning (adapted from Rapaport and Snickars 1998).
The present practice is an obvious discrimination of “soft”, non monetary factors, which are treated as secondary. It is common that, the best fitting of all the sustainability criteria might not be reached. Recent regulations of the Swedish Road Administration, being in accordance with EU law, “take into consideration environmental concerns while developing the road alternatives, not only after pathways have been derived” (Rapaport and Snickars 1998). The authors propose being more proactive in the reduction of risks of environmental impacts. They construct a cost surface from all location criteria and define the best suitable path. Multiple criteria techniques are being used for location suitability analysis.
The GIS-based shortest path approach to locating a motorway corridor is well recognized from the theoretical point of view (Huber and Church 1985). It is primarily based on the three following steps. At first, geographic data is represented in the framework of a square grid. Secondly, each cell in the grid is assigned a ‘suitability score’ which represent how well the cell is suited for a given purpose. Lastly, the shortest path (e.g. least impact or cost path) algorithm is executed from origin to destination and a string of cells that have the least sum of suitability scores is derived. The procedure has some weaknesses and further modifications. Geometric distortions, common for grid approach, are widely discussed in Huber and Church (1985). However, the most subjective moment in the above procedure is an assignment of suitability scores. The authors recognize the Delphi method and the nominal group technique, but multicriteria evaluation methods seem to work best because they are capable of handling different, usually incomparable, criteria (Jankowski and Richard 1994, Malczewski 1999 and 2004, Jha 2003).
There are some examples of using the least-cost path algorithm for location studies applied in GIS and different approaches to the problem of suitability scores assignment. Previously mentioned work of Rapaport and Snickars proposes a set of three conflicting objectives: to
Economic and social issues
Alternatives
Environmental issue (EIA)
a) Present practice b) Desirable practice
Environmental quality Social well-being Economic prosperity Sustainability
minimize risk of mischievous environmental impact, minimize building costs and minimize travelling time. The first objective was composed of seven different environmental risks, which increase as distance decreases. Linear and logarithmic functions were used to show the level of noise dispersion. After re-scaling, all of the criteria were combined by simple additive overlay technique. Second objective’s estimates are derived from the road authority’s calculations on the basis of slope and geological factors, land use type and level of disturbance of existing traffic during the construction. As long as the speed of travel between two points is constant the travel time is proportional to the distance, therefore the best path would equal to the straight line connecting them.
Bailey and his colleagues (in Grossardt et al. 2001; Bailey 2003 and Bailey et al. 2005), unlike previous authors who apply objective distance and cost indicators, use stakeholder consultations to identify important evaluation criteria and derive values for the friction surface. The authors see a corridor alignment issue as a decision problem and propose combination of decision evaluation methods with a similar GIS-based least-cost path algorithm through the surface. The analytic hierarchy process (AHP) was used to evaluate the relative importance of each of the 69 factors grouped in five affinity groupings. It took 6 months to finish the consultation process but public consensus was reached over the location of the road.
The two GIS applications of the shortest path corridor location problem aim at incorporating all sustainability factors into alternatives generating stage of planning. However, the shortest or least-cost path method for facilities location is not the only approach. This can be also seen as a site selection problem, investigated for example, by Cova and Church in 2000 (after Husdal 2000). Environmental impacts were also investigated from the functional point of view by Swedish road planning authorities (Seiler and Eriksson 1995, Eriksson 2004). These approaches will not be followed in this paper.
* * *
The short review of the corridor location approaches presented above structures the ways of seeing the corridor location problem. In order to assure sustainability, proper evaluation criteria must be identified already in alternatives generation stage. It seems most appropriate to address this problem by multicriteria location suitability and pathway analyses. Additionally, the thesis makes use of the least-cost path algorithm available under ArcView desktop license.
The next chapter will introduce the reader to the concept of GIS, multicriteria evaluation methods and least-cost path issues.
3. THEORY
2.1. What is GIS?
Spatial information has been stored in the form of maps from as early as ancient Egypt times. Recently, invention of the microprocessor in the 20th century boosted the development of data collection and data processing technologies. It was only after this that manual cartographic computations were, to large extent, replaced by automated and computerized ways of gathering data. For example, precise surveying instruments, global positioning system (GPS), aerial photographs and scanning add up to the field. Growing processing capabilities of desktop computers allowed development of relational database system, processing of satellite images and air photos and manipulation of large quantities of data, which all constitute a modern GI system. For the first time, GIS was adopted into geodesy and natural sciences, but soon it turned out that it can be applied into other research disciplines, like business or transportation. Further, the environmental modelling applications evolved into demographic and business activities analyses and the results could be used for decision making (Bernhardsen 2002).
A geographic information system (GIS) can be intuitively defined as a computer system to derive information from digital maps. Scientifically, the GIS is an information system that have capability to input, store, retrieve, manipulate, analyze, and present spatial data. Furthermore, it is equally made up from hardware, software, as well as data input and output devices, and a communication system between them (Bernhardsen 2002, Malczewski 2004). It is not exclusively mapping device. The false believe comes probably from association of GIS with computer-assisted design (CAD) systems and other geodetic mapping software. The latter are much more powerful in terms of cartographic accuracy and map production. However, the novelty of GIS lies in the possibility of manipulation of spatial features together with its database attributes. In comparison to the traditional analogue overlay, GIS wins with the easiness of retrieval of spatial data from the geodatabase and displaying it on the screen. Furthermore, the technology supports unlimited number of ways of combining different thematic layers and presenting them to an audience. Ability to integrate data from different sources is one of the most valuable functions of GIS.
As mentioned earlier, the GIS’ graphical interface is interconnected with a database. The most common relational database allows displaying data as a set of thematic layers. Depending on the data model stored in the geodatabase, one can distinguish vector and raster approaches.
In the vector data model every geographic object is graphically displayed as a point, line or a combination of these – a polygon. On the database side instead, geometric shapes are coded with its geographic coordinates. The vector GIS is characterized by smooth, scalable shapes. Frequently precision of display is much higher than actual precision of data acquisition. Database features displayed in the form of map are characterized by topological relationships like connectivity, adjacency, containment, etc. Topology allows more advanced spatial analysis. Analytical functionality of GISs will be discussed later in the chapter. Apart from relational database, object-oriented database approach is gaining in popularity among researchers.
The raster data model is constructed of regularly distributed spatially referenced grids of cells. Square-sized cells are called pixels. Accuracy of raster data is determined by a pixel size used for data collection. Below that limit any further approximation of data is not possible because a cell characteristic is assumed to be identical over its whole extent. From the file code side,
the image is generated on the basis of a binary code. A numerical value, being that integer or float, is assigned to every cell to describe its attributes. In raster, unlike in vector, spatial feature’s integrity is not maintained. This is because one object can be composed of many adjacent cells. For this reason, as well as due to more ‘blurred’ object borders, raster representation is more suitable for continuous phenomena (like temperature or precipitation). Neither raster nor vector data model is perfect, but each suits different purposes. Overview of pros and cons is presented in Table 3.1. It is worth mentioning that both raster and vector data types can be combined in order to boost its presentation capabilities. For example, raster elevation data is often overlaid with vectorized topographic elements. The result gives a better visual perception of physical surface.
Raster and vector data models work originally in the two-dimensional environment. The third dimension could be elevation, time, and movement. For example, physical surface can be represented in raster as a digital terrain model (DTM) where every grid cell is assigned elevation value. In vector, triangulated irregular network (TIN) is derived from irregularly distributed height points and consists of a tessellation of triangles connecting them. Vector-based three-dimensionality is usually obtained by adding third variable (z) to the x and y coordinates.
Table 3.1. Comparison of raster and vector data types in different tasks
Raster Vector In data input
Efficient in data acquisition (scanned images) Data sources include satellite images and aerial photographs
Time-consuming and work intensive in data acquisition (on-the-screen, on-the-desk digitizing) Data needs to be digitised either from paper or from digital images. GPS can be used to collect some data
In storage
Requires large space capacity (data intensive), but compressing techniques exist
Geographic objects are stored as collections of adjacent grid cells
Less storage space consuming, however, space requirements grow exponentially together with data complexity
Geographic features are stored as ordered pairs of coordinates in a relational geodatabase Preserves non-spatial information in the form of attribute tables
In analysis
Raster is analysis oriented.
Efficient and fast in performing the analysis Occupies relatively less operational memory Uses simple but powerful map algebra operations like add, subtract, multiply, divide
Vector is data management oriented. Operations are more complicated and mathematically advanced
Requires more computation capacity and occupies more operation memory
Uses geometric operations like union, intersect and identity
In presentation
Suited for presenting continuous phenomena Gives rough borders of objects due to pixel-like construction of image
One phenomenon can be presented in a layer
Well suited to present discreet geographic features
Gives smooth lines and nice looking objects Easy to overlay and present many spatial features
2.2. Functions of GISs
The GIS data manipulation functionality tools include acquisition and verification, compilation, storage, updating and changing of spatial data and theirs attributes as well as management, exchange, retrieval and presentation of results. Important part of every GIS is capability to combine spatial data and analyze spatial patterns (Bernhardsen 2002).
What distinguishes all GISs from other mapping software is the capability to read input data, manage them in a geodatabase and generate output in the graphical form of maps. Actually, handling of spatial data and storing attribute information as spatial objects is fundamental to any GIS. Visual presentation of data on maps enables understanding of spatial relations between objects which might not be possible in any other way.
Analytical functions of GIS should not be underestimated. The GIS operations can be divided into basic and advanced. Upon this distinction, basic functions are those that can be applied to a wide variety of applications and therefore are included into most of the commercial GISs (like ArcGIS/ESRI, MapInfo/MapInfo Corporation, Idrisi/Clark University Graduate School of Geography, GRASS/US Army Corps of Engineers, GeoMedia/Intergraph, SPANS/Tydac Research Inc. and TransCAD/Caliper Corporation). On the other hand, advanced analytical functionalities of GIS are usually dedicated for one purpose and are customised towards specified user. As such, they are built on basic components to form new capabilities (Malczewski, 2004).
The basic GIS functions include among others, measurements of length, area and perimeter, queries to database, reclassification of raster images, buffering and neighbourhood functions, overlaying, spatial analysis, surface analysis, and spatial interpolation. Most of them are described in Heywood et al (2002). Advanced operations can be e.g. based on theoretical models, which were recently implemented into GIS environment. Shortest path algorithm, least-cost path algorithm and cartographic modelling are some of the examples of such applications.
Measurements
The length measurements are usually performed to obtain the distance between two points. This can be done both in raster and vector, but different mathematical formulas are applied. One can consider several types of distances, for example Euclidean distance, Manhattan distance and distance along the network. The first is the distance along a straight line, seen as a shortest path, calculated using Pythagorean geometry. Manhattan metric resembles city distances. In raster, Manhattan distance can be used for simplification of operations. It is less commonly used with vector where it is substituted with distance along the network. The network distance is a sum of lengths of all segments forming a link between origin and destination. The distance measurement in GIS is a foundation for more advanced functions like buffering and other proximity analysis. Length measurements are also used by spatial statistical functions like data clustering and interpolation methods.
The area calculation in vector is approximated from the sum of areas of simple geometric figures included in a polygon. Perimeter is a sum of straight line segments forming the boundary of an object. In raster GIS, the area is computed as a sum of cells multiplied by the cell’s area.
Database queries
There are two types of query to a geodatabase: spatial and non-spatial. The former is about location of the object and the latter involves attributes of features. The queries can be used to search for errors in a database as well as to get information about spatial features and location. As a highly interactive function it can be used in both directions, from database to location and from location to information. Database queries usually apply Boolean operators (AND, NOT, OR, XOR). Queries can be also combined with questions about distances, perimeters and areas. Queries to the attributes are also useful when managing the data and editing. Database queries are applied mostly to vector data types because they apply to a database management system. Nevertheless, interactive information tools can be used to derive pixel values from floating point raster on the screen. Since attributes can provide vital information on the characteristics of the features they are a base for further analysis.
Overlaying in vector
The idea of overlaying is to put one or more thematic layers on top of each other to see their spatial relationship. Additionally to visual comparison, GIS offers computational possibilities to produce new layers based on the two input layers.
Vector overlaying is a relatively complex computational task. The function uses geometric transformations like union, intersect and identity, which are referred as geoprocessing tools. They are basically available in point-in-polygon, line-in-polygon, polygon-on-polygon overlaying. It is important that data are topologically correct. The computational overlaying process will create new data layers but it is still possible to visually assess the spatial relationship between features. Vector is well suited for this kind of tasks. Additionally, topology can be used in spatial queries to select features from one layer based on features in another layer. In ArcView spatial queries’ output needs to be saved in the database in order to preserve new layers.
Overlaying in raster (Map algebra)
Overlay analysis in raster is conceptually easy and fast. Raster overlying is known as scalar overlaying because it uses such mathematical operations as addition, subtraction, multiplication and division. It is described as map algebra. The process requires reclassification of the original attribute cell values in order to receive meaningful results. The overlay serves as the analytical comparison tool which can be used to analyse differences in two input layers. The output is always a raster layer which may be temporary or permanent. Visual interpretation of two overlaid layers is also possible. It is most useful for elevation data and aspect data by manipulating transparency of one of the layers.
Overlaying can be used together with buffering functions or spatial queries to answer more advanced spatial problems. It is also used with attribute queries to receive required results. The layers are combined using one of the available techniques, like set and arithmetic operators or map algebra.
Vectorization and rasterization
Most GISs offer a function of converting raster to vector and vector to raster. In a rasterization process polygons, lines or points are recognized at a vector layer and overlaid with a defined cell size grid beginning at the bottom-left corner of the vector. Then, each cell is assigned a code from vector feature to which it belongs. For polygons, the centre of the cell must be
contained within the polygon in order for it to take polygons attribute. During the same process for lines, the cells that touch the line are given the same values, so that the rasterized line consists of linearly adjacent cells. The same process is performed for points, which are in turn represented by single pixels (Bernhardsen 2002; Heywood 2002).
Vectorization is a more complicated process and different programs can give different results for the same raster. Generally, having the raster with coded values, regions of same values are recognized and boundaries set between them. Then, the coordinates of the end points of the straight line segments are stored in a vector layer. Vectorization requires some after treatment, which may include line smoothing, topology creation and allocation of IDs (Bernhardsen 2002; Heywood 2002). In both transformations some information can be lost, which makes the data less accurate.
Reclassification
Reclassification is helpful for executing the analysis in raster as well as for raster data querying. In the latter, a result is a new image. One can perform excluding reclassification or weighting reclassification. The first aims at giving the value of ‘1’ to the items that are of interest and the value of ‘0’ to those that should be excluded. Weighted reclassification assigns new values for cells based for example on the importance of its representations.
Neighbourhood
Neighbourhood functions are “operations in raster GIS where values of individual cells are altered by adjacent cells” (Heywood et al. 2002). This can be applied to filter remotely sensed imagery in order to a get clearer picture or to eliminate excess variation in data. In this process, new values are assigned to cells on the basis of the values of cells contained in a specified neighbourhood. Different algorithms, like minimum and maximum filter, most frequently occurring class, number of different classes or mean filter, are used to recalculate the value of target cell.
Buffering
The buffer function finds distance from each point and creates zones of given distance around specified features. Buffering is difficult task from the computation point of view. Buffer around points is the simplest operation. Computations of buffer zones around lines and polygons are more mathematically advanced. Buffering is successfully combined with other data layers and has multiple applications. Together with proximity functions it is often used in optimal site selection problems, when simple linear distance to features is a constraint. It can also be used to create zones of equal distance to a destination.
Least-cost path
The least-cost path algorithm is commonly included in commercial GISs. It determines the shortest weighted distance from origin to a destination through a cost surface. The cost is defined by its neighbouring cells’ relations and can be represented in monetary or other non-monetary units. Therefore the weighted distance is not measured in geographic units but in cost units (Miller and Shaw 2001).
The least-cost path performed in raster GIS is based on the regular square grid (RSG) which is conceptualized to vectorized representation (Figure 3.1). The algorithm perceives each cell as
a point placed in the centre of it. The centric point of each raster cell is connected with adjacent cells through links. Therefore this representation of raster is called node/link representation. If the interactions are in horizontal, vertical and diagonal directions the neighbours are based on the so-called ‘queen’s case’. If it is only in the direction parallel to the axes it is referred as ‘rook’s case’ (Miller and Shaw 2001, ESRI Desktop Help).
Figure 3.1. Node/link representation of raster
In GIS, a process of generating the least-cost path is performed in two steps. All calculations are based on a cost surface raster and a source raster (Figure 3.2a). Using a node/link representation, every node is assigned a value from the cell and every link is assigned a resistance value representing cost of travel between the two neighbouring cells. When movement is in horizontal or vertical direction the resistance values are calculated using the formula below.
a1 = (cost1 + cost2)/2 (1)
Another formula applies for diagonal movement:
a2 = 1.4142 (cost1 + cost2)/2 (2)
The accumulative cost is a sum of all costs from the source to the destination cell.
Calculation of the least accumulative costs for each cell is an iterative process which leads to the creation of a weighted-distance raster (Figure 3.2b). Firstly, the source cells are located and assigned zero value because there is no cost to travel to them. Next, all of the immediate neighbours are assigned cost of travel from them to the source and the least cost cell is selected and assigned to the output raster. Simultaneously, the direction of movement from the selected cell is coded as a back link raster. Again the accumulative cost of travel is recalculated for all the immediate neighbours of the selected cell. If the cost is higher than that previously calculated, it is ignored. However, if the cost value is lower, the old accumulative cost is replaced. The cells that are already assigned to the output raster are not recalculated and treated as ‘permanent’. The process is repeated until the least accumulative cost is assigned to all cells and the back link raster is completed. Generation of the weighted-distance raster is based on graph theory and the shortest path is based on the Dijkstra algorithm (ESRI Desktop Help).
Node
Link
On such defined accumulative cost surface, it is possible to perform a least-cost path to any location from the origin point (source cell) (Figure 3.2c). If the origin is to change, a new weighted-distance raster needs to be calculated. Note that ‘No Data’ cells are excluded from the possibility of travel. The procedure described “ensures that the lowest accumulative cost is guaranteed for each cell”. Furthermore, the cost calculation formula guaranties that the cost-distance is adjusted to the cost-distance to travel because every link value is multiplied by the cell size. Therefore, cost assigned to every link reflects a resistance value of moving over one geographic distance unit (ESRI Desktop Help).
Figure 3.2. Procedure to calculate least-cost path in raster-based GIS
Some modifications to the least-cost path allow performing the least-effort path through a physical surface. This function is called ‘Path Cost’ in ArcGIS. It compensates the actual surface distance resulting from elevation and slope as well as other horizontal and vertical Source raster
Back link raster Weighted-distance raster Accumulative cost: Acc_cost = a1 + a2 5.9 4.4 9 7.5 11.5 5.5 3 4.5 0 14 13 10.4 10.8 5.8 1 1 1 0 1 0 ‘No Data’ Source cell Least-cost path a2 = 1 + 1 2 = 1.4 * 1.4142 2 1 5 8 9 4 1 4 5 5 8 7 5 1 Cost surface raster
a1 = 5 + 1 2
= 3
Example of the least cost path
1 a)
b)
c)
factors (wind direction, friction, etc.) that may influence the travel. ‘Path Cost’ is used in hydrological modelling, movement and dispersion processes modelling (ESRI Desktop Help). In addition to the basic path algorithm, it is possible to modify it to meet the requirements of more demanding applications. One of the possibilities is to extend the neighbourhood to the second immediate in order to increase the number of possible movement directions. Such a procedure would smooth the path line (Miller and Shaw 2001).
Figure 3.3. Conceptual visualisation of gateway function
In planning, it is often a requirement to have more than one proposal. To ensure their spatial dispersion, a midpoint location can be used (Figure 3.3). In theory such a solution is called a gateway shortest path problem (GSP). The GSP problem is solved by performing twice the Dijkstra shortest path algorithm, first from the origin to destination and then the other way round. The resulting accumulative distance (or cost-weighted) surfaces are summed and accompanied by a direction raster, which shows the shortest path to either of the origin or destination. In this way any point in the space between the origin and destination can serve as a gateway (Husdal 2000 after Lombard and Church 1993, Miller and Shaw 2001).
Cartographic modelling
Cartographic modelling is an example of an advanced method of GIS analysis. It is a “methodology for structuring a GIS analysis scheme” (Heywood 2002). It consists of data pre-processing, graphical representation of the process of modelling as a flowchart, and execution of the model using GIS operations. Cartographic modelling applies map algebra tools together with other basic analysis operations in GIS. It has graphical environment in most of commercial GIS systems. The ESRI’s Model Builder is prepared for ArcView/Spatial Analyst. This technique can be used to automate the analysis consisting of many similar steps.
2.3. Decision making
Decision making problem arises when there is a difference between the present state and the desirable state, the need for change is identified and there are at least two courses of action possible. Preferably, the best alternative should be selected. Solving such a problem can be managed by dividing it into smaller, easier to handle parts, analyze each part separately and integrate all the analysis results into one meaningful proposal. A way of integrating the parts is vital for understanding and justification of the final output and is a core of the art of decision making.
Gateway
Destination Origin
Decision making is a key element of business operations which is in turn an interest of management science. As a broader perspective it is any situation when more than one course of action is possible, and can be used in many more variety of fields (Malczewski 1999). The decision itself depends on “the nature of the problem, the policy of the decision maker and the overall objective of the decision”. The possible types of action include: choice of an alternative solution, ranking of the alternatives from the best to the worst ones or the assignment of the considered alternatives into predefined classes (Doumpos 2002).
One can classify decision making problems as discrete problems and continuous problems. The former recognize a discrete set of alternatives. The attributes of each alternative can serve as evaluation criteria. On the other hand, the continuous problems consist of infinite number of alternatives. The decision space is rather represented as a region of feasible solutions where any point belonging to this region corresponds to a specific alternative. An example of this form of problems is resource allocation (Doumpos 2002).
There are many frameworks to analyze decision-making. One most widely accepted is a three step approach, which distinguish following phases:
1. intelligence 2. design 3. choice
In the first phase, the problem and the opportunities of change are recognized. Secondly, feasible alternatives are identified and lastly the choice has to be made. It must be pointed out that making is an iterative process which means that at any stage of the decision-making process it may be necessary to loop back to an earlier phase (Malczewski 1999 after Simon 1960).
Another, more precise decision making procedure comprises four stages specified as follows: 1. Problem definition
2. Search for alternatives and selection of criteria 3. Evaluation of alternatives
4. Selection of alternatives and final recommendation
In the first stage, a need for changing the present state in the specified direction is identified. Secondly, the feasible alternatives are selected and decision criteria are proposed. In the third step, the selected alternatives must be evaluated in terms of what impact each of them has on every evaluation criterion. This involves strategies for limiting the number of possible actions. For example, reduced processing involves iterative tightening of the threshold values until only one alternative is left. Another, much more elaborative are multiple criteria evaluation techniques, which consider all the possible solutions and develop trade-offs between different decision’s criteria. The last step produces final recommendation in the form of one, several top-listed alternatives or ranking of all the alternatives from best to worst (Jankowski 1995 after McKenna 1980).
2.4. Multiple criteria decision making – an overview
Multicriteria decision making (MCDM) is a term including multiple attribute decision making (MADM) and multiple objective decision making (MODM). MADM is applied when a choice out of a set of discrete actions is to be made. In MODM, it is assumed that the best solution can be found anywhere in the feasible alternatives space, and therefore is perceived as continuous decision problem. MADM is often referred as multicriteria analysis (MCA) or multicriteria evaluation (MCE). Instead, MODM is more close to Pareto optimum searching with use of mathematical programming techniques (Jankowski 1995, Malczewski 1999). In this paper, the term multicriteria decision making is used in reference to multiple attribute decision-making and the other expressions are used as equivalents.
The main objective of MCDM is “to assist the decision-maker in selecting the ‘best’ alternative from the number of feasible choice-alternatives under the presence of multiple [decision] criteria and diverse criterion priorities”. Every MCDM technique has common procedure steps, which are called a general model (after Jankowski 1995). This procedure includes following actions (Figure 3.4):
1. Deriving a set of alternatives 2. Deriving a set of criteria
3. Estimating impact of each alternative on every criterion to get criterion scores
4. Formulating the decision table with use of the discrete alternatives, criteria and criterion scores
5. Specifying decision-maker’s (DM) preferences in the form of criterion weights
6. Aggregating the data from the decision table in order to rank the alternatives (simple and multiple aggregation functions)
7. Performing sensitivity analysis in order to deal with imprecision, uncertainty, and inaccuracy of the results
8. Making the final recommendation in the form of either one alternative, reduced number of several ‘good alternatives’, or a ranking of alternatives from best to worst All the MCDM techniques are based on the above presented general model. However, division can be made for compensatory and non-compensatory methods. The compensatory methods can be further subdivided into additive and ideal point techniques, where the first includes e.g. weighted summation, concordance analysis and Analytical Hierarchy Process and the latter, Technique for Order Preference by Similarity to Ideal Point (TOPSIS), Aspiration-level Interactive Method (AIM) and Multi-Dimensional Scaling (MDS). Non-compensatory techniques are for example dominance, conjunctive, disjunctive and lexicographic techniques. There is no place in this paper to describe specifically all of them. Only two, most popular are discussed. Good summary of the MCDM techniques and its choice strategy is given by Jankowski (1995); Voogd (1983) provides a comprehensive theoretical background.
Figure 3.4. A general model of MCDM (after Jankowski 1995)
All additive methods, being compensatory techniques, are based on the standardized criterion scores, which can be then compared and added. Standardization allows comparison of criterion scores within one alternative, to come into some kind of trade-off when poor performance of the alternative under one criterion can be compensated by a high performance under another criterion. Total score for each alternative is achieved by multiplying criterion score with its appropriate weight and adding all weighted scores. Weighted summation technique, being a basic form of additive methods, can be written down in the matrix algebra as follows:
The weighted summation allows for evaluation and ordering of all alternatives based on the criteria preferences by decision-makers. However, there are techniques which allow setting preferences to both criteria and criterion scores. Second technique, Analytical Hierarchy Process (AHP) “uses a hierarchical structure of criteria and both additive transformation function and pairwise comparison of criteria to establish criterion weights” (Jankowski 1995).
Set of alternatives (A)
Final recommendation Sensitivity analysis DM’s preferences Aggregation function Decision table Criterion scores Set of criteria (C) S1 . . Si C11 ... Cj1 . . . . C1i ... Cji W1 . . Wj = x Where:
Si is a total score for alternative i,
Cji is a criterion score for alternative i and criterion j Wj is criterion weight.
2.4. Multi-criteria decision making and GIS
GIS has good capabilities of handling spatial problems, and as such can be used to support spatial decision-making. Solving a complex multiple criteria problem without spatial analytical and visualization tools would be computationally difficult, if not impossible (Jones 1997).
Multicriteria decision making techniques, as stand alone tools, have been computerized and nowadays there is much software to use. However, it is not common that such software is capable to handle spatial problem in the form of maps. There exist two strategies: loose and tight, for coupling of GIS with MCDM techniques (Jankowski 1995). The loose coupling relies on a file exchange mechanism which enables communication with the two types of software. Separate tasks are performed in either of software. GIS is used for performing land suitability analysis, selecting a set of criteria and their scores in order to export the decision table into MCDM program. The MCDM module is used for executing multicriteria evaluation and the result is transferred again into the GIS for display. The tight coupling strategy instead, is realized by a common interface and common database for GIS and MCDM. This in fact means that the multicriteria evaluation functions are embedded into the GIS software. The advantage is that all necessary functions are on place and troublesome data exchange is avoided. However, not every proprietary GIS have developed such a facility in its basic version. There is example of IDRISI, which employs pairwise comparison and Analytic Hierarchy Process to evaluate weight scores (Clark Labs). Another software Spans, by Tydac Technologies, has inbuilt weighted overlay functions, which are similar to weighted summation MCE technique (Carver 1991). The ESRI software provides a cartographic modelling tool called ModelBuilder, which is capable to handle similar decision problems, hence requires some initial input of work.
Generally speaking, multicriteria evaluation with use of GIS can be done in two stages, (i) survey and (ii) preliminary site identification. In the first step, the area is screened for feasible alternatives using deterministic decision criteria. Here, all the sites, which meet all the exclusion criteria (constraints) simultaneously, are identified and taken away from the analysis. This stage is sometimes referred as suitability analysis, traditionally performed by manual map overlay, further revolutionized by GIS digital maps. The second stage, called preliminary site identification, is operationalized by MCE techniques. First, secondary siting factors are elaborated and then weighted according to their importance. The second stage allows handling multiple objective problems (Carver 1991, Jankowski 1995).
Multiple criteria overlay was proposed by McHarg (1969) who suggested identifying physical, economic and environmental criteria in order to assure social and economic feasibility of the project. The complexity of the decision problem determines whether binary or multiple values overlay technique is used (Figure 3.5a and b.). In geographic analysis, most commonly used operations are AND and OR (Boolean), which correspond to spatial ‘intersection’ and ‘union’. If the decision factors have different levels of importance, weighted overlay should be used (Figure 3.5c). However, special scores aggregation procedure is required to achieve meaningful results (Jones 1997).
Figure 3.5. Spatial multicriteria evaluation: a) binary overlaying; b) multiple values overlaying; c) multiple values weighted overlay
Malczewski (1999) proposes more elaborated structuring of the spatial decision problem (Figure 3.6). Traditionally, all begins with problem definition, where raw data are obtained, processed and examined for opportunities or problems. In this stage GIS capability of data storage, management, manipulation and analysis are of great advantage. The consecutive steps, described below, emphasize the GIS role in the process.
Evaluation criteria
An evaluation criterion is a term used to encompass both objectives and attributes of multicriteria decision problem (Malczewski 1999). Other authors refer them as decision criteria or factors and scores respectively (Voogd 1983, Carver 1991). The objectives describe the desirable state of a geographical space. They formulate the criteria that need to be fulfilled in order to make the right decision by “minimizing” or “maximizing” some variables. The attributes, on the other hand, contain measures used to assess the level of achievement of the criterion by each alternative. Evaluation criteria are presented in GIS as thematic maps or data layers.
It is required that decision attributes fulfil several requirements. Firstly, they need to be measurable, which implies that it should be easy to assign numerical values that correctly asses the references to or the level of achievement of the objective. Secondly, an attribute should clearly indicate to what degree the objective is achieved, which is unambiguous and understandable for decision maker. This is called comprehensiveness of an attribute. Furthermore a set of attributes should be operational. If the attribute is understandable for the decision maker, he/she can correctly describe relation between the attribute and a level of achievement of the overall objective than it can be used meaningfully in the decision-making process. A set of attributes should also be complete, which means that it covers all aspects of a decision problem. The set of attributes should be minimal, which form the smallest possible
set that completely describes the decision problem. No redundancy means that consequences of valuation of decision influence only one attribute. The test of coefficient of correlation can be used for every pair of attributes to test for no redundancy. Lastly the set of attributes should be decomposable. It is true if evaluation of the attributes in the decision process can be simplified into few smaller decisions. Usually evaluation criteria form a hierarchical structure (Malczewski 1999).
Figure 3.6. Spatial multicriteria analysis in GIS after Malczewski (1999), modified Selecting a proper set of evaluation criteria can be done by means of literature study, analytical studies or survey of opinions. Literature can be found with some authors providing literature review of criteria evaluation to a specific spatial decision problem. Governmental agencies and governmental publications can provide guidelines for selection of evaluation criteria. Another method is to recognize objectives from governmental or other documents and review relevant literature to identify attributes associated with every objective. Analytical studies can be performed for example by system modelling. Opinions’ survey is aimed at people affected by decision or a group of experts, where several formalized techniques exist (Malczewski 1999). Problem definition Final recommendation Evaluation criteria Constraints Alternatives Decision matrix Decision rules Sensitivity analysis Criterion weights Criterion maps Ordered alternatives Constraint/feasible alternatives map DM’s preferences Geographic data accuracy MCE technique DM’s preferences
A set of objectives and attributes used for a specific decision is affected by data availability. It may not be feasible to obtain required information for the ideal set of attributes designed for a specific objective, or data may not exist. The choice of attributes is also limited by cost and time of gathering the data. It must be a trade-off between the accuracy of prediction and cost and time required.
An example is taken from the case study considering location of a water transmission line, where six pipeline corridor alternatives are evaluated. The criteria were, among others: total cost of route, amount of public right-of-way, area of wetlands and length of streams falling inside each corridor. All of the cited criteria have natural measured scale, dollars, acres and meters respectively. The decision table would have rows representing the alternatives, columns representing the criteria and fields for criterion scores. The field values are derived from spatial analysis. Another table was constructed to weight every criterion and then the total score for each alternative calculated (Jankowski and Richard 1994). Another example of criteria could be e.g. geology, land use type, land acquisition cost, buildings, conservation, etc. Certain type of behaviour is assigned to each of them.
Criterion maps
Criterion maps form an output of evaluation criteria identification phase. This follows after input of data into GIS (acquisition, reformatting, georeferencing, compiling and documenting relevant data) stored in graphical and tabular form, manipulated and analyzed to obtain desired information. Usually, with help of various GIS techniques a base map over the study area is created and used to produce several criterion maps.
Each criterion is represented at a map as a layer in GIS environment. Every map represents one criterion and can be called a thematic layer or data layer. They represent in what way the attributes are distributed in space and how they fulfil the achieving of the objective. In other words, a layer represents a set of alternative locations for a decision. The alternatives are divided into several classes or are assigned values to represent the level of preference of the alternative upon given criterion. This is a kind if internal relation within a layer between alternative locations in respect to the attribute. In this way one visualizes more and less desirable alternatives.
The attributes need to be measured in certain scale, which reflects its variability. The scale can be classified as qualitative or quantitative. For example, soil types and vegetation types are expressed in qualitative scale, while precipitation level in a quantitative measure. Scales can be natural or constructed. The natural scale is a scale expressed in objective units, for example in km or in quantity per square km. The constructed scale is a subject of personal judgment e.g. landscape aesthetic, ranked witch numbers or assigned linguistic scale. Another issue is raised for direct and proxy scales. The direct scale measures directly the level of achievement of an objective. If the objective is a cost of building a road, the direct scale would map sites with respect to cost associated with building a road there. The proxy scale is used when the attribute for specific criterion is not obvious and should be measured indirectly. Different techniques are used to generate various types of criterion maps scales.
