To perform this analysis, the open source LE model GLAES [21
] is utilized, along with a set of datasets called Priors
which were constructed along with the release of GLAES [24
]. Discussion of these tools is broken into three parts. First, the logic behind choosing the 36 constraints evaluated in this analysis is shown. Following this, an overview of the GLAES model and associated Prior datasets is provided. Finally, the method to determine the constraint evaluation measures is detailed.
3.1. Criteria Identification
Identification of the most common constraints to consider is accomplished via a review of the available LE literature. In total, 43 [15
] publications representing 53 independent LE analyses were considered, and the constraints included in each case were tabulated. These studies cover many different technologies, although most investigate either onshore wind or CSP plants (although PV and biomass are also present). To find a consensus between the sources, generalizations had to be made regarding the constraint used by the authors taking into account their expressed intentions. Furthermore, constraints which were only used by a single author were removed from this stage of the investigation. A full description of how this procedure was carried out, including a deeper explanation of each of the identified constraints, can be found in the methodological report associated with this work [24
displays the result of this literature review. In the first column the name of each identified constraint is given. Many studies used a general description of some constraints while others made distinctions within these groups, therefore these sub-constraints are listed as well. Additionally, the constraints are divided into four distinct motivational groups to differentiate between the underlying reasons for why the authors included these constraints in their analysis. The Social and Political
group refers to constraints which were included due to social preferences or political mandates of local citizens and other stakeholders. The Physical
group refers to constraints derived from limitations imposed by physical characteristics of the land, such as the soil type, or presence of a forest. The Conservation
group corresponds to constraints related to conservation efforts by local, national, and international organizations. Finally, the Technical Economic
group refers to constraints which are fundamentally included for economic reasons, such as excluding distances too far from power lines beyond which connection costs become exorbitant. The term technical
is used in this case to indicate that constraints in this group are generally not a result of detailed economic evaluations, but are instead assumed thresholds. Next to this column, the frequency that each constraint was used in the study’s analysis (but not necessarily in the LE analysis) in comparison to all of the reviewed studies is shown. In the following two columns, the typical expression of each constraint is provided in regards to the threshold value chosen by the authors who used these constraints; for example, most studies which considered settlements excluded distances less than 800 m from any settlement. In actuality, the authors used a wide array of values for each constraint which, in some cases, would fluctuate greatly. Along side these typical value columns, the associated methodological report for the GLAES model [24
] also provides low and high constraint expressions to show this spread. Two exceptions to this are aspect and irradiance, which have been altered from their typical literature expression. An agreement within the literature regarding how aspect should be measured was not apparent, therefore the measure of degrees in the northward direction
was chosen with a typical threshold of 3° N. As for irradiance, the LE studies which include this constraint generally suggested a threshold value around 5 kWh/m
day; however, as this is a far too strict value for a European context, a value of 3 kWh/m
day is used instead. Finally, although one of the Prior datasets is used to represent each constraint, the fundamental data source which was used is given in the last column. Further background to the Prior datasets is provided in the following section.
Due to the European scope of this analysis, a suitable data source could not be found for all of the identified constraints. Therefore, these constraints are not displayed in Table 2
. This includes distance to radio towers, gas lines, power plants, earth quake zones, land slide zones, flood plains, specific types of natural vegetation and private land. Furthermore, distance from historically significant sites is also not listed despite its prevalence in the sources seeing as how a consensus could not be found in the literature regarding what constitutes historical significance. Fortunately, all of these constraints appeared in less than 10% of all of the reviewed studies with the exception of historical significant sites (25%) and private land (13%).
3.2. GLAES and Prior Overview
The GLAES model [21
] is an open source project developed for the purpose of standardizing the implementation of LE analyses. This project was initialized in part to address the methodological inconsistencies currently present in the LE literature. GLAES is designed to be adaptable to common geospatial data formats, to be scalable to large geographical areas, to minimize expected errors resulting from geospatial operations, and to be methodologically transparent [24
]. The model has been implemented in the Python 3 programming language, with primary dependencies on the Geospatial Data Abstraction Library (GDAL) [72
] for geospatial operations and on the SciPy [73
] ecosystem for general numerical and matrix computations, both of which are also open source projects.
To conduct an LE analysis with GLAES, the following steps must be taken. First, a study region must be defined in the form of a vector file and used to initialize a GLAES analysis. Values for resolution and spatial reference system can also be provided. Following this, multiple exclusion constraints can be applied one at a time by providing GLAES with a data source to exclude from and instructions on how to indicate the areas which should be excluded. The manner by which GLAES accomplishes this depends on the data source. If given a raster source, GLAES will expect a minimal and maximal value defining the pixels which should be excluded. If given a vector source, GLAES can accept a Structured Query Language (SQL)filter string to identify the specific features which should be excluded. Furthermore, these data sources do not need to be expressed in the same projection system as the one with which the analysis was initialized, as GLAES is capable of translating between projection systems as needed. Finally, regardless of the type of source which was provided, GLAES can also be given a buffer value by which the indicated exclusion areas can be grown. Once all of the desired exclusions constraints have been applied, GLAES can generate a raster file of the resulting available areas. For the purposes of this work, all computations in GLAES were performed in the EPSG3035 projection system with a spatial resolution of 100 m.
Along with the development of GLAES, an effort was also made to produce a set of general datasets to represent common geospatial criteria and the outcome of this effort is the so-called Prior datasets. Geospatial criteria are useful for many application besides simply as exclusion constraints in LE analyses, such as in multi criterion decision management analyses. Unfortunately, one of the inconsistency issues in GIS-related modeling heavily discussed by Resch [14
] is the inconsistent use of data sources. To address this issue in the European scope, the Prior datasets were created and, like the GLAES model, are also openly available [21
Each Prior dataset represents a single criterion and is simply a processed version of other open datasets. Furthermore, they are all expressed as byte-valued rasters defined over the European context. As discussed in the underlying methodological paper [24
], the same literature review used to generate Table 2
was used to suggest several criteria which should be represented along with the range of values over which each criterion is relevant. Having identified the criteria and their range of relevance, an open source dataset was chosen to express the criteria at several different criteria values, called edges
. For example, as one of the identified criteria, the distance from lakes is commonly used in literature with relevant values ranging between 100 m and 5 km from the closest lake. Therefore, the edges 100, 200, 500, 1000, 2000, 3500, and 5000 m from a lake could be chosen as the edges to be processed. By consulting the chosen dataset, the Prior dataset is then constructed in such a way that a pixel value of 0 refers to all pixels which would have been indicated by the first edge (100), while a pixel value of 1 refers to indications by the second edge (200), and so forth. In this way, the Prior datasets do not exactly recreate the information of their underlying sources, meaning that some information is lost in their use; although the edges are chosen to have reasonable fidelity in the range of high interest. Nevertheless the Prior datasets can be used directly in the GLAES model to allow for rapid criteria evaluation in the European context and thereby make large scale LE and other GIS analyses easier to manage.
Although the production of the Prior datasets are not discussed in detail here, the fundamental databases used are briefly described. The Corine Land Cover (CLC) [23
] is the most frequent fundamental source for the Priors used in this study. This is a raster dataset which describes the land cover at each 100 m patch of land across Europe. Many different land cover classes are found in this dataset, including settlement areas, mining sites, and open water bodies. The OpenStreetMap (OSM) [66
] dataset was extracted and is developed via volunteered geospatial information. Taking the form of a vector dataset, features such as roadways, power-lines, touristic and leisure areas can be easily identified. The Digital Elevation Model Over Europe (EU-DEM) [63
] dataset from the European Environment Agency is a digital elevation raster dataset providing elevation values over Europe and possesses a pixel resolution approximating 30 m. This dataset was used to determine the elevation, slope, and aspect at all locations. The World Database on Protected Areas (WDPA) [64
] is the result of a multinational effort to monitor protected areas and includes designations of protected areas as described by the International Union for Conservation of Nature [74
]. This dataset includes designations for bird areas and habitats identified by the European Union’s bird’s directive [75
] and habitat directive [76
]. Indications of designated protected areas of all types can be found in this vector database, which was filtered differently for each of the conservation constraints. An important note in regards to the WDPA is that features are not mutually exclusive, and thus a single location could be defined as protected according to multiple designations. The World Wildlife Foundations’ HydroLAKES [68
] database is a vector source which was used to identify lakes and other stagnant water bodies. The Global Wind Atlas (GWA) [70
] is the result of a collaboration between the Technical University of Denmark and The World Bank to simulate typical wind speeds at each 1 km by 1 km location across the globe; values at altitudes of 50, 100, and 200 m are provided, but only those at 100 m values are used in this analysis. Similarly, the Global Solar Atlas (GSA) [71
] is the result of The World Bank’s effort to estimate average daily irradiances at most 1 km by 1 km location in the world, excluding latitudes above 60° and below −45.5°. The GSA provides average values for the global horizontal irradiance, direct normal irradiance, and other parameters related to solar energy, although only global horizontal irradiance values are used in this analysis. Finally, three datasets available on EuroStat were used. The first of these differentiated large airports, those with more than 150,000 annual passengers, from smaller airfields, those with fewer than 150,000 annual passengers [67
]. This dataset simply provides location and usage data for airports within Europe, which were then matched with footprints found using CLC. The second EuroStat dataset is a vector source tracing probable routes of running water [69
]. This source was used to identify rivers and stream too small to be found in CLC. The third EuroStat source provides vector representations of urban settlements [65
] and was used to differentiate urban from general settlements as seen in CLC.
3.3. Evaluating Constraint Measures
As the first stage of the evaluation, a visual representation of where these constraints exert influence is sought. To accomplish this, each of the constraints shown in Table 2
were independently evaluated across the entire European continent using the GLAES model and Prior datasets. This operation produces 36 maps of Europe, each of which displaying the influence of one constraint. Instead of plotting each of these results, however, these independent exclusion results were aggregated according to motivational groups. Irradiance and wind speed were not included at this stage as distributions of these values are well known in the European context. As a result, the technical economic motivational group only consisted of exclusion from the access distance and connection distance constraints. This resulted in four maps of Europe in which locations excluded by at least one constraint in the associated motivational group would also be excluded in the aggregated result. Finally, these four maps were overlaid with one another, with a value assigned to each pixel according to the combination of the four motivation groups contributing to the exclusion of that pixel. The resulting map expresses 16 possible combinations at each point, ranging from indications that no motivation groups excluded the considered pixel to indicating that all motivation groups excluded the considered pixel.
Following this, the relative evaluation of the 36 constraints is performed according to the three measures previously introduced: independence, exclusivity, and overlap. To determine independence, the independently evaluated constraint maps generated from the previous step were queried to extract the percent excluded for Europe as well as for all countries in the study area. The constraints were then ordered in a descending manner according to their average exclusion percent across all nations. This order was used to compare relative impact value of each constraint, meaning that earlier constraints are more valuable in the sense that, when their consideration is warranted for the LE analysis in mind, they tend to exclude the most land.
The exclusivity and overlap were investigated by querying two independent constraint maps at a time. By starting with the first, given, constraint and comparing it against the second, overlapping, constraint, the number of pixels excluded by the given constraint as well as the number of pixels excluded by both constraints were recorded. An intermediate value was given by the ratio of these two values (given pixels over shared pixel). This ratio was then found for every pairwise combination of constraints. The final exclusivity value for a single constraint was then the average of all intermediate values where the constraint in question acts as the given constraint. Likewise, the final overlap value for a single constraint was the average of all intermediate values where the constraint in question acts as the overlapping constraint. As before, the constraints were then ordered according to their exclusivity and overlap values. The exclusivity values were put in an ascending order such that constraints near the beginning of this list tend to have little overlap with the other constraints and therefore have a higher measure of exclusivity. These low-exclusivity-scoring constraints can be considered important to properly exclude since the areas indicated by these constraints are less likely to also be excluded by other constraints. On the other hand, the overlap values were put in descending order. In this way, constraints towards the beginning have a higher measure of overlap and are particularly useful as these have a tendency to exclude areas that a researcher may want to exclude for other reasons (and potentially may not have accurate data for).