Quantifying a realistic, worldwide wind and solar electricity supply

Front matter

Quantifying a realistic, worldwide wind and solar electricity supply

Yvonne Y. Denga, *, Martin Haighb, Willemijn Pouwels c, Lou Ramaekersc, Ruut Brandsmac, Sven Schimschard, Jan Grözinger d, David de Jagerc

a Ecofys UK, 1 Alie Street, London E1 8DE, UK

Abstract

Nearly all long-term energy projections rely heavily on renewable energy sources on the assumption of abundance. Yet, already today, wind and solar projects can encounter local objections and competition with other uses. This paper presents the ranges of realistic potential supply for solar and wind electricity, using a 1 km 2 grid level analysis covering the whole world at country level. In addition, the potential for building-based solar electricity is assessed. We find that long-term combined potentials range between 730 and 3700 EJ/a worldwide, depending crucially on the acceptable share of land—up to 3.5% of total (non-ice covered) land on earth. Realistic potentials account for limitations such as land-use competition and acceptance, together with resource quality and remoteness as proxies for cost. Today’s electricity demand (65 EJ/a) is well covered by the range, but constraints may occur in the long run locally. Amongst large countries, Nigeria and India may need imports to meet electricity demand.

Introduction

Faced with the twin challenge of energy security and the largely unmitigated externalities of conventional energy sources, and buoyed by the economic co-benefits and continuing increase in cost-competitiveness of renewables (IEA RETD, 2012), most governments are formulating policy frameworks which encourage a high penetration of renewable power sources in the medium- to long-term. Renewables face many challenges and uncertainties as they grow and become integrated into the energy system. Yet projections often assume that the resource base provides no limitation, notably for wind or solar energy. To test this assumption, we have conducted a detailed, global analysis to provide credible, practical, realistic and consistent potentials for electricity from solar and wind sources, using the highest resolution datasets that are publicly available. We not only take into account technological development but also attempt to address implementation constraints such as grid connection, competition with other uses or possible local opposition. Most significantly we try to quantify real availability of these surfaces, beyond simple technical or geographic limitations, using an availability factor.

We stress that we do not present a full energy system analysis here as is done typical energy analysis studies (IEA, 2012a; GEA, 2012; Lund and Mathiesen, 2009; Deng et al., 2012; Shell, 2013), but rather a detailed quantification of technical resource potentials which are used as an input to such analyses. For example, the results of this study have been used as a key input for the renewable energy projections in Shell’s (2013) New Lens Scenarios.

The IPCC Special Report on Renewable Energy Sources and Climate Change Mitigation summarised the existing body of work on renewable energy potentials. It highlighted the difficulty of comparing different studies varying in geographic scope, technology scope and approach (Edenhofer et al., 2011). The surveyed studies (Krewitt et al., 2009; REN21, 2008; UNDP et al., 2000; Hofman et al., 2002; Trieb et al., 2009; EEA, 2009; Siegfriedsen et al., 2003; DLR, 2006; Defaix, 2009; Hoogwijk, 2004; de Vries et al., 2007; Zhou et al., 2009; Denholm and Margolis, 2008; Schwartz et al., 2010; Jacobson and Delucci, 2011) are frequently cited, but most of them:

are primarily meta-analyses themselves (Krewitt et al., 2009; REN21, 2008; UNDP et al., 2000), or

assess only a single or a handful of related technologies (Hofman et al., 2002; Trieb et al., 2009; EEA, 2009; Siegfriedsen et al., 2003; DLR, 2006; Defaix, 2009) or

miss a part of the potential (e.g. assessing roofs, but not facades for solar photovoltaic (PV) energy systems) (Defaix, 2009;

Hoogwijk, 2004; de Vries et al., 2007; Zhou et al., 2009; Denholm and Margolis, 2008), or

pertain to a single country or region, not the whole world with differences in approach or assumptions between regions (EEA, 2009; Siegfriedsen et al., 2003; DLR, 2006; Defaix, 2009; Zhou et al., 2009; Denholm and Margolis, 2008; Schwartz et al., 2010), or

cover the whole world, but do not provide sub-global detail (Jacobson and Delucci, 2011), or

focus on theoretical or technical potentials, taking only a limited number of, or no, additional barriers to implementation into account (EEA, 2009; Hoogwijk, 2004; de Vries et al., 2007), or

use low resolution datasets, potentially over- or under-estimating potential significantly locally (Hoogwijk, 2004; de Vries et al., 2007).

They commonly find (very) large potentials: for land-based solar and wind potentials combined, results on the order of a few thousand to tens of thousands of exajoules of final electric energy per annum are reported.

In contrast to these existing studies this paper presents, for the first time, a comprehensive, bottom-up assessment of renewable electricity potentials across the entire globe for all solar and wind electricity technologies: wind, solar photovoltaic (PV) and concentrated solar power (CSP) including land- and sea-based resources as well as building-based PV potential. The focus in this study is not on full technical potential, as is the case for many existing studies, but on an estimate of the realistic, or constrained, technical potential, which accounts for technical as well as non- technical limitations, such as acceptance, cost, competition with other uses or remoteness. For land- and sea-based technologies, we achieve this through a detailed analysis of land use and cover in a geographic information system (GIS) with up to?1 km 2 resolution, the most detailed, global analysis to date. We first find the technically suitable area, by successive exclusion of geographic and technical factors. In addition, we include a final step which aims to assess a realistic future maximum availability share of all land which is technically suited for use.

For building-based PV potentials we use the same technology and resource assumptions as on land, but calculate the available area on both roofs and fac¸ades at country level, starting from a building stock analysis of a representative set of reference countries.

In addition, and for completeness, we present a meta-analysis of existing studies for hydro- and geothermal electricity, with extrapolations to assess realistic long-term potential.

Methods

The overall framework to assess any solar or wind energy potential followed a three step approach by determining:

a. the available area (on land, on sea, on building roofs and building fac¸ades)

b. the amount of resource incident upon this area (wind speed, solar irradiation)

c. the amount of energy a technology could capture of this total resource (i.e. conversion)

The three steps above are then combined according to Eq. (1) to yield the overall potential estimate for three study periods: 2010, 2030 and 2070.

where P = potential in EJ per technology, for each study period; A = area in 2 per technology; I = resource intensity, i.e. potential per area, in EJ/2, which is calculated differently for solar and wind technologies and can vary per study year; p = the study period (2010, 2030, 2070); t = the technology per category, i.e. CSP, wind on land or sea, PV on land or building roofs, or fac¸ades.

The terms and steps above differ by technology (PV, CSP, wind) and category (sea, land, buildings) and are discussed in detail below.

Of the two factors in Eq. (1), the first factor, the available area, has a much larger uncertainty than the second factor, the effective resource per area. Note that the available area can also vary over time, due to land use changes, e.g. from urbanisation and deforestation. However, these changes are expected to have a much smaller bearing than other elements, most notably the ‘availability factor’ and have therefore not been taken into account here.

For hydro- and geothermal electricity, the potential is much more discretely distributed across the world, as it is associated with specific localised features. These two resources were not assessed in depth, but via a meta-analysis of existing studies.

Available area

Land- and sea-based resources were calculated using a geographic information system with datasets with up to 1 km? 1 km resolution. The available area for these resources was calculated by starting with the total world surface area (146? 10 6 km 2 on land and 361? 10 6 km 2 on sea) and succes- sively excluding areas which would not be suitable for use for a given technology (wind, PV, CSP). In a final step we attempted to estimate the percentage of suitable area which would be realistically available for renewable electricity production. Note that this availability of a given type of area is the factor with the largest uncertainty. All these steps are described in detail below, first for land-based resources, then for sea-based resources, and formalised in Eq. (2).

where A = total available area for a given technology and period; al = availability factor per land type; Ai = suitable area in km2 in grid cell i for a given technology; p = the study period (2010, 2030, 2070); t = the technology, i.e. CSP, wind on land or sea, PV on land. 2.1.1. Land-based solar and wind resource The area available for land-based wind and solar electricity installations is restricted by the following factors. These factors have been used to estimate suitable area onshore based on the data sources in Table 1 in the following steps:

Exclusion of Antarctica: The land area of Antarctica was excluded.

Elevation:

Wind: Areas above 2000 m were excluded due to the significantly lower power density at such elevation.

Land cover–Urban area: This was excluded for all technologies based on a combination of data on urban land cover and on population density. Note that for PV potential on buildings, a different approach, not based on GIS, was used (see below).

Land cover–Forests:

Solar: All forest areas were excluded for solar electricity production. Note that we also excluded mixed land covers (cropland or grassland with some forest) in this exclusion step.

Wind: Non-protected forests were not excluded for wind power production (but received a lower availability in the final step).

Land cover – Ice, Water, Coast, Cliffs, Dune, Rock: Areas of these land cover types were excluded for all technologies.

Protected areas:

From the World Database on Protected Areas (WDPA) data set (see Table 1) all classified protected areas were excluded (Natura 2000 and Cat. I–VI). Non-classified areas were not included in this study due to insufficient information.

All land cover classes of ‘rain forest’ and ‘tropical forest’ have been fully excluded.

Slope:

Solar – PV and Wind: Grid cells (1 km? 1 km) with an average slope of more than 158 (?27%) were excluded.

Solar – CSP: Grid cells (1 km? 1 km) with an average slope of more than 28 (?4%) were excluded.

Resource intensity:

Solar – CSP: Areas with a direct normal irradiance value of less than 1900 kWh/m 2 /a were excluded for CSP. Availability of nearby water was not used as an exclusion criterion as technological developments are currently underway to minimise reliance on water for CSP.

Solar – PV: No areas were excluded based on resource intensity for PV as current practice shows that latitude is not a strong factor in determining where PV is installed and there is still significant price reduction potential in PV which would likely render all areas within reach of cost-effectiveness. However, we note that the lowest horizontal irradiation levels found at the latitudes included in this study are around 800 kWh/m 2 /a.

Wind: To reflect the fact that areas of low wind speed are unlikely to see significant investment in wind power installations we excluded areas with low average wind speeds in our calculations. The economically feasible minimum average wind speed is heavily dependent on local factors. For this global study, we implemented a critical cut-off of 6 m/s at hub height, with an assumption of average hub-height rising from 80 m now to 90 m for 2030 and 2070 (see also Supplementary Information online). Only areas with average wind speeds below this cut-off in four out of four quarters within a year were excluded.

2.1.2. Sea-based wind resource

The area available for sea-based wind power installations is restricted by the following factors which have been used to estimate suitable offshore area based on the data sources listed in Table 1:

Economic zone attribution: Only offshore areas attributed to a country jurisdiction have been included in this study. In case of joint jurisdiction the area has been split evenly between the administrating countries. Disputed areas have been completely excluded from the analysis.

Sea ice: Areas likely to be impacted by sea ice have been excluded by following the winter sea ice line around the Arctic and by implementing a general cut-off at 608 latitude in the Antarctic.

Ocean floor depth: Current technology for offshore wind power primarily relies on pylons driven into the sea-bed. This technology is already in use up to 50 m ocean depth. At (much) higher depths it is expected that floating turbine technology will be deployed. This technology is expected to be deployed within the study time horizon (up to 2070). There is no established upper limit for the depths this technology can reach; we have used a cautious cut-off of 1000 m depths; i.e. offshore areas with deeper ocean floor depths have been excluded.

The distance from shore is not a technical limitation per se, but areas very far from the coast have been assumed to be too costly to connect to land for the foreseeable future. We have used a critical cut-off of 200 km. It is noted, however, that most areas further than 200 km offshore would have been excluded in the previous steps on ocean floor depth and exclusive economic zones anyway.

Protected zones: Where data on ecosystems and protected zones was available, these have been excluded, but freely available data is sparse. Most of the areas excluded in this step were identified through the WDPA data set with some additional exclusions of salty marshes and rocky reefs from NCEAS (see Table 1).

Maritime use: Data on maritime use is not comprehensive. We have excluded the shipping lanes, areas of artisanal fishing and oil rigs (including buffer zones) contained in a dataset provided by NCEAS (see Table 1) to estimate areas unavailable for offshore wind parks due to prior uses. This will lead to underestimates in some areas (only 10% of heavy shipping traffic is logged in the NCEAS data set) and overestimates in other areas (all existing oil rigs are excluded but it is likely that several of them will have been dismantled by 2030/2070).

Resource intensity – wind speed: In analogy with the approach for onshore wind above, we implemented a critical cut-off of 8 m/s at hub height, with an assumption of average hub-height

Table 1 Data sets used for the GIS modelling of land- and sea-based resources. Category Data Resolution Source Reference Land Elevation 1 km USGS (-EROS) GTOPO30 USGS (1996) Land cover/use 1 km USGS GLCC v2 Loveland et al. (2000) Population 5 km NASA–SEDAC v3 CIESIN et al. (2005) Country borders?2 km VMAP0 NIMA (2000) Railways 10 km NGA/USGS VMAP0/VMAP1 NIMA (2000) Solar irradiance 40 km NASA LaRC ASDC (2012) Wind speed (land) * 19 km CRU, CL2.0 New et al. (2002) Land/sea Wind speed (all) * 120 km CISL CISL (2012) Protected areas 1 km WDPA IUCN and UNEP (2011) Sea Coastlines?2 km VMAP0 NIMA (2000) Ocean depth 2 km NOAA Amante and Eakins (2009) Maritime use 1 km NCEAS impacts Halpern et al. (2008) Reefs & marshes 1 km NCEAS ecosystems Halpern et al. (2008) Offshore borders?2 km Exclusive Economic Zones VLIZ (2012)

The wind speed on land was determined as an average between these two datasets as the CRU dataset had a better resolution but suffered from sparse data in some world regions (China, Africa, South America). rising from 90 m now to 110 m for 2030 and 2070 (see online Supplementary Information for more detail). Only areas with average wind speeds below this cut-off in four out of four quarters were excluded.

2.1.3. From suitable to available area The exclusions above yielded the suitable area for solar and wind power. To assess the area which will likely actually be available for electricity production we used a geographically non- specific availability factor (see Table 2). This can be interpreted as representing the average share of the suitable area in a given grid cell, region, or country, which is likely to be actually available for a PV, CSP or wind power park in the given period. Because this availability factor is highly uncertain but has a large bearing on the overall potential, we spanned a range of possible values in our assumptions, denoted Low, Medium, High and varied the factor for the High case between industrialised and developing countries, and different land types. For solar electricity, availability factors were based on Trieb et al. (2009) for the Medium case and variations around this number for the Low and High cases. For on-shore wind, we tried to derive availability factors from case studies in Germany, Denmark and the Netherlands. We estimated currently installed wind power capacities to cover around 1–2% of suitable area in these countries (see Supplementary Information for the calculation behind these numbers). Given that these countries are not, on average, at the limit of their wind power density (although individual regions within the countries may be reaching the limits acceptable to the local population), we have assumed that this represents around 20–30% of the possible maximum available area, resulting in an availability factor of 5–6%. This is in line with another European study (EEA, 2009) postulating 4–5% availability. Based on these considerations, we used a range of 3–10% for our three scenarios and a much lower availability for forests. Since availability for offshore wind parks is today primarily impacted by competing use and perceived visual impact, the availability factor was varied by distance from shore (see Table 2). The factors were based on the following considerations:

Social acceptance is only expected to play a role <50 km from shore as wind farms at higher distances are not visible, therefore availability has been heavily restricted for distances <50 km from shore.

The Low case is based on the values used in another study (EEA, 2009), but leads to a more restricted area since that study started

with a much larger (less constrained) suitable area.

The Medium and High cases assume full availability (100%) minus an (upper) estimate of areas for protection (Medium Case) or protection and shipping (High Case) which we did not have sufficient data for to exclude above.

2.1.4. Resource cut-offs

Although this study is not attempting to calculate economic potentials, it is instructive to translate the resource intensity cut- offs into approximate production cost values at least for our base year and 2030. These are shown in Table 3 with inputs derived from literature (Teske et al., 2011; Schlo¨mer et al., 2014) and assuming a lifetime of 20 years and a weighted cost of capital of 10%. Note that these approximate costs are higher than typical current costs, as they represent costs for low resource intensity areas and that actual production costs would be minimised on a project-by-project basis.

2.1.5. Building-based solar resource

A GIS approach for assessing roof and fac¸ade area would carry a large uncertainty as the global GIS dataset only identifies artificial surfaces at a 1 km? 1 km scale, not differentiating between roofs on buildings and other structures. Instead we have used floor space to derive the roof and fac¸ade area suitable for solar resources as shown in Eq. (3).

where Aroof = area of building roofs suitable for energy harvesting using PV; Afacade = area of building facades suitable for energy harvesting using PV; Pop = population for each country per period and urbanisation level (United Nations, 2009); p = the study period (2010, 2030, 2070); c = country; u = urbanisation level; Afloor/Pop = floor area per capita for each country per period and urbanisation level; Froof = roof to floor ratio, dependent on building type and urbanisation level; Ffacade = facade to floor ratio, dependent on building type, urbanisation level and orientiation; b = building type; o = orientation; Sroof/facade = suitability factor on roofs/fac¸ades, i.e. share of full roof/fac¸ade deemed suitable and available for energy harvesting.

2.1.5.1. Floor area. We calculated floor area for each country based on the floor area per capita and the population per country. The floor area per capita was first estimated for ten reference countries and all other countries were mapped to these ten reference countries based on several characteristics (region, climate zone, GDP per population). The mapping is shown in the Supplementary Information online. Table 2 Availability factor for technologies on land (Ind. = industrialised, Dev. = developing country). Technology Land class or distance from shore class Availability (by case and country type) Low Med. High Ind./Dev. Ind./Dev. Ind. Dev. Wind on-shore Forest 0.5% 1% 2% 2% Agriculture 3% 6% 10% 20% Desert Grassland Barren land PV/CSP Agriculture 0.1% 0.5% 2% 5% Grassland 0.5% 1% 3% 5% Barren land Desert 2% 5% 8% 15% Wind off-shore 0–10 km 4% 5% 5% 10–50 km 10% 30% 40% 50–200 km 25% 60% 80% For more information about how these factors were established, see the Supplementary Information online. Table 3 Estimated production costs for the medium term at the resource cut-offs used here. Technology Resource cut-off Approximate levelised costs in EUR 2012 /kWh 2010 2030 Wind on-shore 6 m/s at hub height 0.08–0.12 0.05–0.08 Wind off-shore 8 m/s at hub height 0.10–0.18 0.07–0.14 CSP 1900 kWh/m 2 /a DNI 0.24–0.68 0.12–0.33 PV [?800 kWh/m 2 /a GHI] 0.53–0.80 0.11–0.21

For the ten reference countries, we differentiated between building types (multi-family home, single-family home, non- residential buildings), urbanisation level (rural and urban), period (2010, 2030, 2070) and new or existing buildings and also tried to account for the increasing share of the population living in multi- family homes and urban areas in the future. We first estimated base year (2010) values for floor area per capita in all of these categories based on existing statistics, where available (de la Rue du Can, 2009; DoE, 2009; Diefenbach and Loga, 2011; EIA, 2006; IFO Institute, 1999; IBGE, 2005; Milford, 2009; Ministerio de Fomento, 2010; Ministerio de industria tourismo y comercio, 2007; Schlomann et al., 2009; Statistisches Bundesamt, 2009; Zhou et al., 2007). For future years, we trended historic growth, but moderated this growth where necessary to achieve consistency across countries. This iterative approach yielded values for the floor area per capita from 10 to 76 m 2 in 2010 and 23 to 71 m 2 in 2070 for residential buildings and 1–29 m 2 in 2010 and 5–37 m 2 in 2070 for non-residential buildings based on modelling of building stock development in each of the reference countries. For more detail, see the Supplementary Information online.

2.1.5.2. Roof to floor ratio/fac¸ade to floor ratio. We determined roof- to-floor and fac¸ade-to-floor (North, South, East and West) ratios, differentiating between single-family and multi-family homes in rural and urban areas and non-residential buildings. We estab- lished these ratios for three reference countries and mapped all other countries to these reference countries, based on the same characteristics as above. The roof-to-floor ratios ranged from 0.17 to 1.00. The fac¸ade-to-floor ratios varied between 0.10 and 0.33 or 0.00 and 0.47 depending on orientation. A zero value represents a wall connected to the wall of an adjacent building. For more detail, see the Supplementary Information online.

2.1.5.3. Suitability. From the total roof and fac¸ade area, the suitable area (synonymous with available area here) was estimated. For this we used a suitability factor of 33% for roofs and 10% for fac¸ades, rising to 30% in 2070. These factors were derived from literature (Bergamasco and Asinari, 2011; Ghosh and Vale, 2006; IEA-PVPS, 2002; Izquierdo et al., 2008; Montavon et al., 2004; NREL, 2008; Pillai and Banerjee, 2007; Scartezzini et al., 2002).

Potential per area

In the final step, the conversion of raw resource incident on the available area was converted to usable electricity through an effective conversion efficiency. Note that we do not differentiate further between different types of sub-technologies, e.g. mono- vs. polycrystalline solar cells, etc., but characterise each technology by one efficiency which is representative of the whole sub-technology spectrum.

Solar electricity production

The potential per area for solar electricity production can be calculated from the solar energy incident upon the surface per year and the conversion efficiency of the harvesting technology, as shown in Eq. (4).

where P/A = potential per area in EJ/km2 per study period and technology (PV vs CSP); p = the study period (2010, 2030, 2070); t = the technology (PV or CSP); I = resource intensity in J/m2 for solar irradiation; E = conversion efficiency in % by technology.

2.2.1.1. Resource intensity. For solar photovoltaics (PV) on land or building roofs, the global horizontal irradiation (GHI) per grid cell or country was derived from the solar irradiance data source in kWh/km2/a listed in Table 1. The GHI includes the total gross solar energy from both direct and diffuse radiation incident on a horizontal plane. For buildings, we calculated an average irradiation value per country across all grid cells, weighted by population density.

For solar photovoltaics (PV) on building fac¸ades we also calculated the gross vertical irradiation (GVI) in kWh/km2/a by orientation (North, South, East, West). For concentrated solar power (CSP), the direct normal irradiance (DNI) per grid cell in kWh/km2/a was derived from the solar irradiance data source listed in Table 1. The DNI is the total direct gross solar energy incident on an area perpendicular to the incoming radiation.

2.2.1.2. Conversion efficiency for PV. The overall conversion efficiency for PV is the product of these factors:

The module efficiency of the individual solar modules (Em)

The performance ratio, capturing the system’s conversion efficiency from the module’s output to usable electricity (PR)

(For land-based PV only): A ground coverage value representing the share of land capturing energy, i.e. the share of the total area of a PV plant actually covered with PV cells (GC)

Table 4 shows the values used for each of these three factors as well as the overall resulting conversion efficiency for buildings (Em⋅PR) and for land (Em⋅PR⋅GC).

2.2.1.3. Conversion efficiency for CSP. The overall conversion efficiency for CSP is the product of two factors:

the net efficiency which represents the internal conversion efficiency 1 of the CSP system and is understood to include the (small) efficiency losses in plants with storage capacity.

a space factor which represents the additional factors to take into account when converting from gross resource over the entire plant area to final electricity produced from the full plant.

The values used for these two factors and the overall resulting conversion efficiency are shown in Table 5. Note that assumptions on the development of technologies (in the form of conversion efficiencies for PV and CSP and hub heights for wind) were not explicitly linked across technologies in a prior scenario, e.g. on installed capacities, but were deemed consistent as they were derived from similar considerations of continued growth.

2.2.2. Electricity from wind on land and sea

The wind power potential per area per year for each grid cell was calculated for each period according to Eq. (5) (based on Held, 2010, p.60) which includes the approximate relationship between the full load hours per year and average annual wind speed, based on a range of turbines, as well as an average power density:

where P/A = potential per area in EJ/km2 per period and technology; p = the study period (2010, 2030, 2070); t = the technology (on-shore vs off-shore wind); H = full load hours in a given timeframe; D = power density, here 7 MW/km2; E = conversion efficiency in % by technology; n2= the average wind speed at hub height in that timeframe; a = 728; b = 2368.

2.2.2.1. Conversion efficiency. There are also small efficiency losses in wind power production which lower the overall resource intensity. The intensity is therefore moderated (multiplied) by the following additional efficiency factors, which range from 90% to 100%.

Operational efficiency: The share of total possible operation time that the wind turbine is actually producing electricity, i.e. not stopped for maintenance. We have varied this by country (industrialised vs. developing) and by distance from shore for offshore wind.

Array efficiency: Multiple turbines can cause interference. The amount of interference depends on land use for onshore wind

(not applicable to solitary turbines, as assumed here for agriculture and forest).

The values used for efficiencies for wind power are shown in Table 6.

2.3. Distance from infrastructure

With the exception of offshore wind, renewable electricity installations today do not usually require additional electricity transmission capacity to be connected to existing transmission and distribution grids. When assessing long-term potentials across larger areas, however, the distance from existing electricity lines may be important, depending on the share of the cost of connection in the overall cost for the installation. To get a first impression of the distribution of our calculated potentials across differing distances from existing electricity grids we have tried to assess the distance of each grid cell from an electricity grid that connects to a significant demand for electricity. Since a global data set for electricity lines was not readily available, we have estimated a ‘distance from infrastructure’ as a marker for electricity grids based on the approach described in the following. Each grid cell was either classed as ‘with infrastructure’ or ‘without infrastructure’, based on land cover, population density and railways as detailed below. For each grid cell without infrastructure we then calculated the distance to the nearest grid cell with infrastructure and then grouped grid cells into distance classes based on the results.

2.4. Classification of grid cells with infrastructure as a marker of grids

To assess whether the potential would be able to supply power into an electricity grid, we used a combination of data sets to estimate the distance from ‘‘infrastructure’’ which we use as a marker for the distance from electricity grids. The presence of infrastructure was based on a combination of: