EnviroInfo 2013: Using a web-based SDSS for siting...
Transcript of EnviroInfo 2013: Using a web-based SDSS for siting...
02.07.2013, WandererHerle.doc
1
Using a web-based SDSS for siting solar power plants
Thomas Wanderer, Stefan Herle1
Abstract
In the global energy turnaround towards renewable clean systems the optimal exploitation of natural re-
sources becomes more important. Several spatial decision problems arise in this context, which are usually
faced by Spatial Decision Support Systems (SDSS) and Geographic Information Systems (GIS). This
holds especially for the deployment of solar power stations which depends on a variety of different spatial
factors like the solar radiation. The paper demonstrates the application of a GIS-based spatial Multi-
Criteria Decision Analysis (MCDA) approach to assess sites’ suitability for utility-scale solar power plants
by implementing and evaluating a Site Ranking tool. The tool integrates accepted spatial MCDA methods
to facilitate well-founded siting decisions. Since it is mounted in a service-oriented architecture (SOA) us-
ing common Web standards, the features are accessible to a broad range of users via a WebGIS client. In
addition, the evaluation of the tool is performed by two conducted case studies trying to match and explain
the current spatial distribution of existing solar power plants in the study areas. It becomes evident that
spatial MCDA is an adequate approach to solve site selection problems for solar power plants even if the
spatial distribution of solar installations cannot be confirmed by the resulting suitability maps in the case
studies. Different sources of error and uncertainty can be identified which should be investigated and tack-
led in the future. Nevertheless, the prototype of the tool provides a sound basis for integrating further ca-
pabilities.
1. Background
The procedure from theoretically available renewable energy potentials to finding optimal power plant lo-
cations is influenced by a multitude of different physical, economic or social criteria with many of them
being of spatial nature. Multi-Criteria Decision Analysis (MCDA) therefore is an adequate approach to
answer questions arising from the challenge to develop renewable energies at the right locations and in an
efficient but sustainable and accepted manner. The developed MCDA approach manages heterogeneous
and unrelated criteria for assessing relevant and optimal locations of solar plants. MCDA is a well-known
methodology for this kind of problems and in connection with spatial data, GIS multi-criteria modelling
techniques have already been used for a variety of similar decision problems in transportation, forestry,
resources planning or other fields. For example, in 1991 Carver used the MCDA technique of Weighted
Linear Combination (WLC) that assigns weights for each criterion and sums these to an overall score for
searching suitable locations for nuclear waste disposal in Great Britain (Carver 1991). Malczewski (2006)
and lately Ferretti (2011) offer a profound overview about literature dealing with GIS-based Multi-Criteria
Decision Analysis.
Recently published studies prove the applicability of MCDA methodology for finding suitable sites and
evaluating the potential of locations for solar power plants. Janke (2010) uses the WLC approach to ag-
gregate various data like solar potentials, landcover, population density and the distance to transmission
lines. In the resulting suitability map of Colorado, an overall score is assigned to each pixel representing
1 Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR), Institute for Technical Thermodynamics, Pfaffenwaldring 38-40,
70569 Stuttgart
EnviroInfo 2013: Environmental Informatics and Renewable EnergiesCopyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5
02.07.2013, WandererHerle.doc
2
exactly one alternative of the decision problem. To reduce uncertainties in the weighting of the criteria,
Kengpol (2012) utilises a fuzzy AHP (Analytic Hierarchy Process) approach to find the most suitable sites
for solar plants in Thailand. Other studies use Ordered Weighted Averaging (e.g. Charabi/Gastli (2011))
or Data-Envelope-Analysis with Principle Component Analysis (e.g. Azadeh et al (2008)).
Certain MCDA techniques have already been incorporated into GIS environments implementing mod-
ules and scripts like in IDRISI or ESRI’s ArcGIS (Lidouh et al 2011). Web-based SDSS have also been
developed in the last decade. One example is the CommonGIS application, a rich-client WebGIS which
also integrates decision support by incorporating a MCDA module (Andrienko et al 2003). However, spa-
tial MCDA techniques implemented as Web services using common standards cannot be found in the lit-
erature.
2. GIS-based MCDA
Spatial decision making (SDM) usually involves multiple criteria with geographic location information of
the available alternatives. As a consequence, many MCDA methods have also been adapted for these
kinds of problems. Spatial MCDA requires capabilities of both MCDA tools and Geographic Information
Systems (GIS). Consequently, the framework for solving spatial decision problems can be described as
GIS-based MCDA (Malczewski 1999). This framework describes a general procedure with different con-
secutive steps including identifying relevant criteria, standardising criteria values, expressing preferences
and specifying the decision rule to aggregate the criteria value to a one-dimensional score.
2.1 Standardisation of criteria values
Since the criteria are measured on different scales, the standardisation of the values is a necessary step in
order to aggregate to a single suitability value for each alternative. Further, all criteria should be trans-
formed in a positively correlated way in respect to the suitability. Several standardisation techniques have
been developed like the linear scale transformation or probabilistic approaches (Malczewski 1999).
The tool supports the standardisation via fuzzy set membership functions because they are easy to un-
derstand and provide the user a variety of possible ways to model the suitability of the corresponding cri-
terion. The user is able to alter the transition between unsuitable values and most suitable values for each
criterion (Eastman 2006). Four different transitions are available - namely linear, sigmoidal, jshaped and
user defined curves. While linear, sigmoidal and the j-shaped curve can take an increasing, decreasing or
symmetric shape, the user defined transition is controlled by a freely chosen finite number of control
points. The choice which of the functions should be applied depends on the relationship between the crite-
rion and the decision set as well as on the information and knowledge available to the decision maker.
Linear or sigmoidal functions are sufficient in the most cases according to Eastman (2006).
2.2 Criterion weighting
The criteria vary in their importance for the SDM process according to the decision makers’ preferences.
Information about the relevance of each criterion is an essential input in order to make recommendations
(Drobne/Lisec 2009). Usually, the preferences are expressed by assigning a weight to each of the criteria.
In this way, the weight vector for which holds describes the pre-
ferred structure for n criteria. A variety of techniques for the derivation of that vector exists. In the imple-
mented tool the user can chose out of the following three methods for specifying the personal preferences:
Ranking method: The user is asked to sort the criteria in descending order of importance. Subse-
quently, the weight vector can be computed by different mathematical techniques.
Copyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5
02.07.2013, WandererHerle.doc
3
Rating method: The user assigns a numerical value to each criterion. A high value indicates
thereby a high important criterion. The normalisation of the weights generates the weight vector.
Pairwise comparison: The user compares each two criteria and indicates which of them is more
important on a predefined nine point continuous scale. The principle eigenvector of the resulting
pairwise comparison matrix represents the weight vector.
2.3 Decision rules
The aggregation procedure in the GIS-based MCDA framework to produce an evaluation score is called
decision rule. In this step, the evaluated criteria scores and the criteria weights are combined to assess al-
ternatives’ suitability. The tool provides three different decision rules: WLC, OWA and AHP.
The WLC is the most used and simplest MCDA technique which is basically a weighted summation of
the individual criterion scores. The aggregation method multiplies each normalised criterion map with the
corresponding specified criterion weight and finally sums the results to a Suitability Index (SI) for each
alternative (Drobne/Lisec 2009). But, it is regarded as limited since it allows full substitutability between
the criteria. A low score on one criterion can be compensated by a higher score on another (Jiang/Eastman
2000). The OWA approach by Yager (1988) addresses this limitation and supports the control of risk and
trade-off in the WLC approach. In addition to the criterion weights, OWA introduces order weights which
are associated with the values of the criteria on a location-by-location basis. In this way, it provides a tool
for defining decision strategies and generating different solution maps concerning predictive scenarios.
The order weight vector can be specified directly or a strategy can be chosen by linguistic quantifiers.
For complex problems the aggregation methods can be extended by the analytic hierarchy process
(AHP), a divide and conquer approach invented by Thomas L. Saaty (1980). The user splits the problem
into several subproblems each consisting of subproblems by itself and/or criteria. This hierarchy can be
described as a tree with the overall problem as the root, the subproblems as inner nodes and the criteria as
leafs. Each problem can be solved by an aggregation function like WLC or OWA.
3. Implementation of the WebSDSS
The SDSS has been developed as a web-accessible tool with the intention to enable users setting the anal-
ysis parameters according to their own preferences. In the context of transforming energy systems, availa-
bility of information, public participation and transparent decision making become a key factor for the fast
adaptation of renewable energies.
While traditional GIS-based SDSS are complex systems that require sophisticated hardware and infra-
structure (Manson 2000), recent Web and GIS technologies allow incorporating SDSS tools into main-
stream IT decision support solutions (Sugumaran/Sugumaran 2007). This development has been fostered
by the standardisation efforts of the Open Geospatial Consortium (OGC) and an active development scene
around web-based GIS technologies driven by commercial companies, research institutions and the open
source community.
The WebSDSS was implemented as a server side process accessible via the OGC Web Processing Ser-
vice (WPS) and producing suitability maps that can be pulled from the server via the widely known OGC
Web Map Service (WMS). Next to the benefit of easy accessibility by using such a service architecture, a
greater flexibility in targeting different user groups is gained by the independence of client user interfaces
from the actual SDSS processing. Accessing the siting support SDSS thus becomes possible by using
solely WPS and WMS commands, or by the developed (JavaScript based) graphical user interfaces (GUI).
The last named ones have been implemented in a simplier and a more sophisticated variant to address the
Copyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5
02.07.2013, WandererHerle.doc
4
different levels of expert knowledge. After sending the request and an estimated processing time of 2-5
minutes, the WebGIS displays the calculated location suitability maps in its mapviewer.
The service's GUI is incorporated in the WebGIS as a module following a wizard-like structure to help
the user through the series of well-defined steps in spatial MCDA (Figure 1-4). Each of the four dialogs
supports exactly the specification of one of the four input parameters for the tool starting with the criteria
selection dialog (Figure 1). The user is able to select from the set of predefined criteria in the left list these
ones which influence the decision problem according to personal considerations. In the next dialog (Figure
2), the user can specify a list of exclusions through conventional relational operators for each criterion.
The third dialog (Figure 3) provides the characterisation of each criterion with the help of fuzzy set mem-
bership functions. The type of the transition and the shape of the curve can be indicated by the radio but-
tons at the top. The red inflection points in the graph facilitate the adjustment of the curve by cursor
movements. The last dialog (Figure 4) provides the definition of the MCDA decision rule. In the left tree,
the user successively arranges the criteria in a hierarchical structure by adding criteria groups and criteria
through context menus. The decision strategy for each criteria group (subproblem) can afterwards be de-
fined on the right. Depending on the weighting method in the WLC settings, a different input form ap-
pears, in which the user is asked to specify weights by ranking, rating or pairwise comparison. In addition,
the user can optionally define an OWA strategy by indicating direct weights or using linguistic quantifier.
Figure 1: Criteria selection Figure 2: Constraints definition
Figure 3: Criteria evaluation Figure 4: MCDA decision rule definition
Copyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5
02.07.2013, WandererHerle.doc
5
4. Case Studies
Two case studies are performed by the help of this WebSDSS, each with different assumptions and inputs
based on the desired technology and area. The generated suitability maps are compared with the real situa-
tions in the selected areas in order to examine the capability of the tool to chart and explain the current
spatial distribution of solar power plants. Therefore the choices of the study areas are based on the exist-
ence of deployed technology as well as on the availability of spatial data. Furthermore, the two solar pow-
er technologies, CSP and PV, are taken into consideration. The autonomous community of Andalusia in
Spain is chosen as reference for the suitability mapping of CSP plants since 21 plants already exist in this
region. For the evaluation of the PV suitability results, the utility-scale PV plants in Bavaria, Germany are
used due to the availability of spatial distribution data and the existing amount of installations.
4.1 Evaluating relevant criteria
Site suitability for large PV and CSP power plants is affected by several factors. Geographical criteria in-
fluencing the spatial decision problem of siting solar power plants are identified from the literature and/or
derived from the authors’ considerations. The data used in the case studies are following: (1) Direct Nor-
mal Irradiation (DHI) or Global Horizontal Irradiation (GHI), (2) Slope, (3) Distance to the street network,
(4) Distance to the electric grid, (5) Population Density and (6) Landcover.
First, performance thresholds are defined for each relevant criterion. Alternatives must meet these
thresholds for any of the evaluation criteria in order to be classified as feasible and to remain in the solu-
tion space. Subsequently, for generating commensurate criterion maps with a common scale a fuzzy mem-
bership function is applied to each criterion (see Figure 5-6 for example computations). The curves, their
shapes and the inflection points are chosen according to a literature review.
Figure 5: J-shaped fuzzy membership function (bottom) &
resulting suitability map (top) for DNI
Figure 6: Linear fuzzy membership function (bottom)
& resulting suitability map (top) for "Distance to the
electric grid"
Copyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5
02.07.2013, WandererHerle.doc
6
4.2 CSP plants in Andalusia, Spain
The autonomous community of Andalusia in Spain is chosen as a reference for the site suitability mapping
of solar thermal plants. In March 2013, 21 large-scale CSP plants with a total collector surface of about
796,500 m² and an installed capacity of 947.41 MW were operating in Andalusia. This amount represents
the major portion of Spain’s CSP plants (Agencia Andaluza de la Energía 2013). Since Andalusia is the
region with the highest amount of solar thermal facilities in Europe, it is likely to serve as the best study
area for a comparison of suitable and current solar thermal plant locations with the data provided in the
tool.
An AHP approach is used to subdivide the goal of siting CSP plants in Andalusia according to the al-
ternative’s suitability into two subobjectives. The first objective describes the technical part of deploying
solar thermal facilities. The DNI, the slope and the population density criterion are related to this objec-
tive. The second objective covers the distance criteria (distance to the street network and distance to the
electric grid) and, hereby, represents the economical component of the problem. Figure 7 illustrates the
structure of the decision strategy as well as the chosen weights for aggregating via WLC.
Figure 7: AHP tree for CSP Site Ranking in Andalusia
Figure 8: Suitability map for CSP plants in Andalusia
Copyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5
02.07.2013, WandererHerle.doc
7
Figure 8 shows the final suitability map of Andalusia for concentrated solar power. The existing CSP
plants were added to compare the suitability of the different locations with the actual sites. The most suit-
able sites have the highest score and are coloured reddish orange in the map. These are mainly located in
the corridor stretching from northeast to southwest which constitutes the Guadalquivir Valley. Further-
more, the current CSP plants are added to the map with the majority of them being located there as well.
In detail, each of the current solar thermal plants is located in a feasible area that is not excluded during
the screening process. The average SI of these locations is 215.23 while the average of all feasible sites is
189.79. A right-tailed t-Test reveals that the mean SI of the CSP locations is significantly greater (
).
4.3 Utility-scale PV plants in Bavaria, Germany
The second case study investigates the capabilities of the service for evaluating locations’ suitability for
utility-scale PV plants. Utility-scale farms are defined in this study as PV plants with an installed capacity
greater than or equal to 1 MW. Further, the assumptions in this model are made for ground-mounted in-
stallations. The Free State of Bavaria, Germany serves as the study area since 514 utility-scale PV plants
with a total installed capacity of 1316.259 MW are already operating in Bavaria (Bayerisches Staatsminis-
terium für Umwelt und Gesundheit 2013).
The AHP hierarchy is similar to the one established in the Andalusia case study. Instead of the DNI cri-
terion, PV modules also convert diffuse irradiation and, therefore, the GHI criterion is taken into account.
The weighting methods are replaced by the pairwise comparison approach resulting in slightly different
weights. The computed suitability map is shown in Figure 9.
Figure 9: Suitability map for utility-scale PV plants in Bavaria
Copyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5
02.07.2013, WandererHerle.doc
8
282 utility-scale PV installations are located in feasible locations with an average SI value of 168.70
while the average SI of all feasible sites has a value of 151.39. The t-Test verifies the significant upward
deviation ( ). However, 233 sites of PV plants are already excluded during the screening
process which is obviously an unsatisfying result for the model fit. A manually conducted analysis reveals
that uncertainties in modelling and the quite broad resolution of the underlying data result in those areas
being excluded in the model. For instance, the criterion maps’ resolution of 1 x 1 km assesses PV plant
sites next to streets as unfeasible alternatives.
4.4 Discussion
While the case study of CSP plants in Andalusia yields adequate matching with the current spatial distri-
bution, the case study of utility-scale PV plants in Bavaria reveals the tool’s limitations. Different sources
of error and uncertainty can be identified in each step in the framework but also in the underlying data.
First of all, the model itself can be misspecified by omitting influencing or including irrelevant criteria.
While the definition of criteria constraints is quite strict and, thus, provides convenient handling, the
standardisation process of criteria through fuzzy set membership functions implies unavoidable uncertain-
ties. Since the specification of inflection points and transition curve characteristics is often a non-trivial
task and requires certain knowledge of the data as well as the influence of the criteria, slight variations can
involve unexpected effects on the results. Also criteria weights assigned according to user’s judgements
are still highly subjective and probably differ from user to user. This comprises even more uncertainties
that cannot be eliminated. Another source of error is the underlying data itself like the case study of utili-
ty-scale PV installations in Bavaria shows. Already in the survey of the data sources, inaccuracies and er-
rors occur which continue and propagate in the course of processing. The quite low resolution of the raster
datasets used in the case study of 1 km x 1 km is even more critical. Undeniably, it can also be possible
that installed PV installations in the second case study are placed in unsuitable areas. In any way, it can be
assumed that this is not the case for the majority of power stations.
5. Conclusion
The implemented Solar Site Ranking tool offers spatial decision support for deploying solar power plants
through different incorporated spatial Multi-Criteria Decision Analysis techniques. The mounting and in-
tegration of the tool as a service in a WebGIS environment using common web standards provides in-
teroperability and access for a wide range of potential users. However, the current implementation of the
Solar Site Ranking service can probably support siting decisions in a first approximation like the results of
the case studies reveal. Since spatial MCDA provides a methodological framework for combining unrelat-
ed data on a common scale, it is quite natural that uncertainties emerge which can barely result in a perfect
model fit. Generally, it is a debatable point whether deploying solar power plants relies on too many fac-
tors which cannot be modelled completely.
Even if the charting of the current spatial distribution of solar power plants failed, the tool provides a
sound skeleton for further investigations with additional criteria and other geographical regions. The mod-
ularisation and the open architecture support a flexible future development and the substitution or adding
of criteria map layers.
Copyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5
02.07.2013, WandererHerle.doc
9
Bibliography
Agencia Andaluza de la Energía (2013): Informe de infraestructuras energéticas andalucía, available at http://www.agenciaandaluzadelaenergia.es/sites/default/files/informe_autonomico_0.pdf (Accessed: May 2013).
Andrienko, G., Andrienko, N., Voss, H. (2003): GIS for everyone: the CommonGIS project and beyond, in: Peterson, M. P. (ed.): Maps and the Internet, Oxford, pp. 131–146.
Azadeh, A., Ghaderi, S., Maghsoudi, A. (2008): Location optimization of solarplants by an integrated hi-erarchical DEA PCA approach, in: Energy Policy, vol. 36, no. 10, pp. 3993-4004.
Bayerisches Staatsministerium für Umwelt und Gesundheit (2013), Energie-Atlas 2.0, available at
http://geoportal.bayern.de/energieatlas-karten/ (Accessed: May 2013).
Carver, S.J. (1991): Integrating multi-criteria evaluation with geographical information systems, in: Jour-nal of Geographical Information System, vol. 5, no. 3, pp. 37-41.
Charabi, Y, Gastli, A. (2011): PV suitability analysis using GIS-based spatial fuzzy multi-criteria evalua-tion, in: Renewable Energy, vol. 36, no. 9, pp. 2554-2561.
Drobne, S., Lisec, A. (2009): Multi-attribute decision analysis in GIS: weighted linear combination and ordered weighted averaging, in: Informatica, vol. 33, pp. 459–474.
Eastman, J. R. (2006): IDRISI Andes: Guide to GIS and Image Processing, Worcester.
Ferretti, V. (2011): Integrating Multicriteria Analysis and Geographic Information Systems: a survey and classification of the literature, in: 74th meeting of the European Working Group on Multiple Crite-ria Decision Aiding, 6 - 8 October 2011, Yverdon Les Bains.
Jiang, H., Eastman, J. (2000): Application of fuzzy measures in multi-criteria evaluation in GIS, in: Inter-national Journal of Geographical Information Science, vol. 14, pp. 173–184.
Kengpol, A. (2012): Design of a decision support system for site selection using fuzzy AHP: A case study of solar power plant in north eastern parts of Thailand, in: Proceedings of PICMET’12: Technology Management for Emerging Technologies, pp. 724-743.
Lidouh, K., Smet, Y. D. E., Zimányi, E. (2011): On the Integration of MCDA in GIS: Current Trends and Needed Advances, in: 74th meeting of the European Working Group on Multiple Criteria Decision Aiding, 6 - 8 October 2011, Yverdon Les Bains.
Malczewski, J. (1999): GIS and multicriteria decision analysis, New York.
Malczewski, J. (2006): GIS-based multicriteria decision analysis: a survey of the literature, in: Interna-tional Journal of Geographical Information Science, vol. 20, no. 7, pp. 703-726.
Manson, S. M. (2000): Agent-based dynamic spatial simulation of land-use/cover change in the Yucatán peninsula, Mexico, in: Proceedings of the Fourth International Conference on Integrating GIS and Environmental Modeling (GIS/EM4), Banff, Canada.
Saaty, T. L. (1980): The Analytic Hierarchy Process: Planning, Priority Setting, Resource Allocation. New York.
Sugumaran, V., Sugumaran, R. (2007): Web-based Spatial Decision Support Systems (WebSDSS): evolu-tion, architecture, examples and challenges, in: Communications of the Association for Information Systems, vol. 19, no. 40, pp. 844-875.
Yager, R. (1988): On ordered weighted averaging aggregation operators in multicriteria decisionmaking, in: IEEE Transactions on Systems, Man and Cybernetics, vol. 18, pp. 183–190.
Copyright 2013 Shaker Verlag, Aachen, ISBN: 978-3-8440-1676-5