2.1. Identification of recently abandoned cropland

The European Space Agency (ESA) Climate Change Initiative Land Cover (CCI-LC) provides global annual data of land cover classes from 1992 to 2015 at a spatial resolution of 10 arc seconds (300 m at the equator) [32]. By combining multiple remote sensing products with ground-truth observations, the ESA CCI-LC maps describe terrestrial surface characteristics through 37 different land cover classes. Six of these classes are cropland or mosaic cropland. We identified abandoned cropland by tracking grid cells transitioning from cropland in 1992 to any non-cropland (and non-urban) class in 2015. In other words, abandoned cropland includes all grid cells that were registered as cropland in 1992 and not in 2015. The cropland grid cells that transitioned to urban land are excluded and not considered as available for renewable energy production, as this is a non-reversible transition.

2.2. Primary bioenergy potential

The bioenergy crops considered here are three promising perennial grasses: switchgrass (Panicum virgatum), miscanthus (Miscanthus x giganteus), and reed canary grass (Phalaris arundinacea). They were selected because they have proven to thrive well on marginal lands due to their suitability to a variety of soil and climate conditions as well as their relatively high yields, low cost and various environmental co-benefits [[44], [45], [46]]. Switchgrass has a C₄ photosynthetic pathway for carbon fixation and can survive in a variety of environmental conditions [[47], [48], [49]]. Switchgrass is native to North America [45,50], but has been introduced in China [46,51,52], Europe [44,48], and many other regions. It is highly adaptive to soil with less fertile, erosive, or acidic conditions, as it has a well-developed root system [46]. However, the species does not cope well under drought conditions [53,54]. The harvesting of switchgrass typically happens throughout autumn [47]. Miscanthus also has a C₄ photosynthetic pathway [47,48]. Miscanthus originated in South-East Asia, but has been introduced in many other regions because of its suitability for establishment in a wide climatic range [49]. Today, miscanthus is present in Europe [48,55], the UK [56], China [52], Turkey [45], among many other regions. Usually, harvesting of miscanthus occurs in late winter or early spring [57,58]. Reed canary grass has a C₃ photosynthetic pathway. It is especially suitable in temperate climates with cool and humid conditions [57,59]. Reed canary grass is currently mainly established in Europe where the species originated [48]. Even in extreme climatic conditions such as floods and droughts, reed canary grass can survive due to its well-adaptive water regime [60]. Similar to miscanthus, the harvest of reed canary grass occurs in late winter or early spring [57,58].

We applied the parameterized crop yield model Global Agro-Ecological Zones 3.0 (GAEZ) to model local yields of bioenergy crops at five arc minutes resolution [61]. GAEZ was developed by the Food and Agriculture Organization (FAO) of the United Nations in collaboration with the International Institute for Applied Systems Analysis (IIASA). GAEZ models dry mass yields based on a variety of crop-specific parameters related to crop characteristics, local soil quality and terrain, agricultural management practices, and site-specific climatic conditions (e.g., radiation, precipitation, and temperature) [33]. It considers the response in crop productivity to yearly variability in soil moisture, pests, frosts and constraints to soil workability. Furthermore, it models the effects of fertilizer use, pesticides, and agricultural conservation measures. Additionally, GAEZ quantify crop water balances, water deficits, and evapotranspiration using the optimal crop growth calendar at rain-fed conditions. Based on this, it also models irrigated crop yields by assuming no water deficits throughout the crop growth cycle (i.e. water losses by evapotranspiration do not exceed water absorption at any point). We quantify bioenergy yields for a modern, mechanized and market oriented agricultural management system, with high yielding varieties, and optimal fertilizer and pesticide use. Maximum attainable dry mass yields for switchgrass, miscanthus, and reed canary grass were computed for both rain-fed and irrigated conditions.

We considered a mixed water supply system with rain-fed water supply in areas threatened by physical water scarcity and irrigation elsewhere. The Aquamaps database [62] provides datasets on physical water scarcity (i.e., when the water demands of the available renewable water resources is not met [63]) at a river basin level. While multiple methods exist to assess water scarcity, the physical water scarcity indicator is especially suitable to address local potentials of new water infrastructure development and measures to increase water use efficiency [64,65]. Additional water withdrawals in areas threatened by water scarcity have higher risks of causing sustainability trade-offs on water depletion. Areas with annual agricultural water use as <10%, 10–20% and >20% of renewable water resources are classified as low, moderate and high physical water scarce, respectively [66]. As an upper irrigation potential, we constrained irrigation deployment to low water scarce areas only.

Applying lower heating values of 17.82 MJ kg⁻¹ for switchgrass, 18.55 MJ kg⁻¹ for miscanthus, and 18.06 MJ kg⁻¹ for reed canary grass [67], we convert dry mass yields to bioenergy yields. By performing an energy-based optimization of the global crop distribution with mixed water supply, each grid is assigned the bioenergy crop with the largest potential. This approach allows to maximize bioenergy yields locally with reduced risks of additional water depletion in water scarce areas.

2.3. Solar photovoltaic primary energy potential

Local PV potentials depend on climatic conditions and nominal installed PV capacity [68]. The incoming solar radiation is the most significant factor for determining the site-specific PV potential [12,69]. Other studies have included additional meteorological conditions such as humidity [[70], [71], [72], [73]], diffuse radiation [74,75], or cloud-cover [76]. The effects of ambient temperature and wind speed are also important for a proper quantification of PV potentials [77,78]. Some previous studies have estimated PV potentials on abandoned cropland at the local and national scale [14,30], but the global PV potential on abandoned cropland is still unclear.

We used the climate data from the Centro Euro-Mediterraneo sui Cambiamenti Climatici Climate Model (CMCC-CM) atmosphere-ocean general circulation model to assess local PV potentials on abandoned cropland [34]. In CMCC-CM, the global circulation is represented by coupling the atmospheric model ECHAM5 [79] and the ocean model OPA8.2 [80], with a resolution of 0.75° and 2°, respectively. Interface between the atmospheric and ocean components is performed by the Ocean Atmosphere Sea Ice Soil version 3 (OASIS3) [81], with a coupling period of 160 min for all parameters. Previous applications of CMCC-CM include assessments of precipitation and cyclones [82], extreme weather [83], vegetation response to climate change [84], and variability of wave power [85]. Spatially explicit climate data on surface downwelling shortwave radiation (RSDS), near-surface air temperature (TAS), and near-surface wind speed (sfcWind) is obtained from the “Atmos” collection of CMCC-CM. Climate model output for the stabilization scenario for Representative Concentration Pathway of 4.5 Wm^-2 radiative forcing (RCP4.5), with monthly ensemble means from 2010 to 2040 which were aggregated to yearly means served as an estimate for the current climatic conditions. While nominal installed PV capacity is typically provided at standardized conditions (i.e. at a PV cell temperature (T_cell) of 25 °C, and RSDS of 1000 Wm^-2), the efficiency of PV cells changes with T_cell based on local climatic conditions. We apply a parameterized approach to quantify the primary energy potential of a PV deployment on abandoned cropland [35]. The method is widely used and validated through a range of applications [68,[86], [87], [88], [89]].(1)Tcell(i,j)=c1+c2⋅TAS(i,j)+c3+c4⋅sfcWind(i,j)(2)PR(i,j)=1+γ[Tcell(i,j)−Tstd](3)PVpot(i,j)=PR(i,j)⋅RSDS(i,j)RSDSstc(4)PVpri(i,j)=PVpot(i,j)⋅PN⋅carea⋅8760

2.4. Development Potential Index for the renewable energy options

Optimal distribution of bioenergy or solar PV on the identified abandoned cropland areas is not only dependent on the primary energy potential, but also on many other factors including biophysical aspects, local land-use or administrative contexts and socio-economic feasibility constraints associated with the intended renewable energy option [9,10,94]. For example, evaluation of local soil characteristics, topography of the landscape and the prevailing climate, as well as grid-specific accessibility to infrastructure, markets, and other socio-economic conditions, are key factors that affect priorities among different renewable energy options [11].

In order to take these aspects into account, Oakleaf et al. [36] introduced the Development Potential Index (DPI), which incorporates site-specific development constraints and criteria when determining the land suitability for development of the given renewable energy option. Spatially explicit DPIs are available for 13 renewable energy sources, including bioenergy and PV. By applying a multi-criteria decision analysis technique integrated with geographic information systems [93], site-specific suitability is mapped globally with a 1 km resolution. DPIs include various development constraints associated with the specific resource, land-use and biophysical criteria, technical yields, feasibility criteria (e.g., major roads, electrical grid, and demand centers), and feasibility measures (e.g., logistics), which differ by location and type of renewable energy option considered. All these factors are integrated in terms of relative importance through an analytic hierarchy process [95] combined with a weighted linear combination method [93], so that the computed normalized spatial DPI values range between 0 and 1. For PV, energy yields and distance to electrical grids contribute the most to DPI values (weighted 0.447 and 0.228, respectively). Landcover characteristics (0.115), distance to urban areas (0.069), and distance to railways and ports (0.026) are also included in the parameterization. Bioenergy DPIs are quantified based on bioenergy crop yields (0.493), market accessibility (0.311) and land supply elasticity (0.196) [36].

We overlapped the DPIs for bioenergy and PV based on Oakleaf et al. [36] with our identified maps of abandoned cropland to quantify their development potentials. We used ESRI ArcGIS Desktop 10.8 [96] to remap DPI data from the Mollweide coordinates system to the WGS 1984 coordinate system. It was possible to directly assign DPI values for bioenergy and PV to 47 Mha and 56 Mha of abandoned cropland, respectively, and there were 40 Mha of abandoned cropland for which it was possible to assign DPI values for both renewable energy sources. For some areas of abandoned cropland, it was thus not possible to assign a given DPI value owing to missing entries in the original DPI dataset, e.g., for latitudes 60–90 °N. We applied a data filling method to quantify DPI values of these abandoned cropland areas by taking the moving mean of surrounding grid cells for a range of different window sizes (from 3 × 3 to 31 × 31 cell windows). With a window size of 3 × 3 cells that only considers the nearest neighbor cells, the extent of abandoned cropland with DPI values for bioenergy and PV increases from 40 Mha to 67 Mha. Across all assessed window sizes, we obtain DPI values for both bioenergy and PV for an additional 27–40 Mha of abandoned cropland. This increases total DPI coverage up to 67–80 Mha. Supplementary Fig. 1 shows that while there are rapid gains in abandoned cropland with DPI values at smaller window sizes, the curve quickly starts to converge towards 80 Mha. The remaining 3 Mha of abandoned cropland are located within larger clusters of missing values such as seen in parts of northern Africa and eastern Asia. We have thus identified the 11 × 11 cell window (about 1 ° × 1 °) as the best option to balance trade-offs between lower areas with DPI coverage and increased uncertainties related to data filling by taking DPI values from far areas within larger clusters. This choice results in 78 Mha (out of the 83 Mha identified globally) of abandoned cropland with assigned DPI values for both bioenergy and PV.

By comparing the spatially explicit DPIs for bioenergy and PV on abandoned cropland, the optimal combination for renewable energy production on abandoned cropland is determined. The energy potentials described in section 2.2 Primary bioenergy potential, 2.3 Solar photovoltaic primary energy potential are applied to each given grid cell so that the bioenergy potential is computed for areas where the bioenergy DPI is higher than the DPI for PV, and the PV potential for areas where DPI for PV is higher than bioenergy DPI. Further, a sensitivity analysis explores the land requirements and energy potentials across different DPI thresholds (i.e., deploying energy production on all areas with DPI values above a given threshold), and the effect of the data filling method on the global potentials. The DPI-optimized land allocation to bioenergy and PV, the corresponding mean energy yields, and total energy potentials on abandoned cropland is therefore computed for a range of window sizes between 3 × 3 and 31 × 31 cells.

3.1. Abandoned cropland

A total of 83 Mha of abandoned cropland globally can be identified from the ESA CCI-LC land cover maps between 1992 and 2015 (Fig. 1). Most of these areas are located in Asia (30%), followed by Americas (28%), Africa (22%), Europe (20%), and, lastly, Oceania (5%). Drivers behind abandonment of agricultural land are many and include environmental, socio-economic and political factors [97,98]. Natural geophysical constraints, such as a decline in soil quality and land degradation, may motivate the cropland abandonment [15,98] as well as topography (e.g., mountain regions and dry areas) [99]. Nevertheless, socio-economic drivers are of larger importance. Lack of availability of critical infrastructure (such as roads) and missing subsidies for farming may force landowners to give up the land [98]. Abandonment of cropland may also result from changes in internal or external politics (e.g., migration of land-use activities to less developed countries [100]), and the shift towards market-driven economies (e.g., the dissolution of the Soviet Union) [99].

 

Fig. 1. Globally identified abandoned cropland between 1992 and 2015. The color bar shows the grid cell fraction at 1° resolution (aggregated from the resolution of the original dataset for visualization purposes). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

In Asia, we find intensive cropland abandonment in Southeast regions close to the equator (e.g., Malaysia and Indonesia). Smaller fractions of abandoned cropland are found in North and Central America, and along the coast of South America. In Africa, regions just south of the equator, such as Congo, Angola, and Tanzania, have large areas of abandoned cropland. In Europe, large areas of abandoned cropland are found in the former Soviet Union (e.g., Russia, Ukraine, and Kazakhstan) and eastern parts of Europe. Also, abandonment of farming occurs in Oceania, mainly along the coastlines of Australia and New Zealand.

3.2. Primary energy potential of bioenergy and PV

Annually, global bioenergy potentials on abandoned cropland are estimated to be 35 EJ (Fig. 2a), assuming modern agricultural practices and irrigation in areas with no physical water scarcity. This means that in water scarcity areas where the abandoned cropland is not suitable to rain-fed bioenergy crops (about 5 Mha), there is no bioenergy potential. On the other hand, PV farms on abandoned cropland can produce 179 EJ year⁻¹ (Fig. 2b), which is more than five times higher than the bioenergy potential. The mean primary energy yield of bioenergy is 0.4 TJ ha⁻¹ year⁻¹ (Fig. 2c) and for PV 2.1 TJ ha⁻¹ year⁻¹ (Fig. 2d).

 

Fig. 2. Primary energy potentials on abandoned cropland for bioenergy (a) and PV-energy (b), in PJ year⁻¹ and energy yield (i.e., energy output per hectare of land) for bioenergy (c) and PV-energy (d) in GJ ha⁻¹ year⁻¹. Note that scales of the color bars are different for each panel. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

For all locations of abandoned cropland, the potential primary energy output is higher for PV compared to bioenergy; however, the magnitude by which PV performs better than bioenergy differs regionally. A clear latitudinal trend can be identified: bioenergy performs better in the tropics (where yields are high) compared to higher latitudes, even though it does not out-compete PV. The highest bioenergy yields are located in the west coast of South America, Africa, and Southeast Asia. However, a significant bioenergy potential is also found in eastern parts of Europe and some locations in Central America. Areas with high PV potential are widely scattered. The main productive regions include the east coast of South America, Central America, parts of Africa, mid-Europe and South East Asia. The lowest PV yields are found in Scandinavia and North America.

3.3. Development Potential Index for bioenergy and PV

We mapped Development Indices (DPIs), a measure of technical and socioeconomic feasibility, of bioenergy and PV from Oakleaf et al. [36] on the identified areas of abandoned cropland (Fig. 3). Average DPI for bioenergy across abandoned cropland is found to be 0.53 (Fig. 3a), while it is 0.58 for solar PV (Fig. 3b). This result indicates that globally the development of solar PV farms on abandoned cropland has a slightly higher development potential compared to bioenergy. However, bioenergy is more suitable compared to PV in some areas due to several local biophysical and socio-economic feasibility factors considered in the DPI. In general, the highest DPI for bioenergy on abandoned cropland are located in tropical and mild temperate regions, while the development potential for PV is higher in dryer climatic conditions. By investigating the differences between DPI for PV and DPI for bioenergy on abandoned cropland per grid cell (Fig. 3c) we find a mean global difference of 0.05. The difference in DPI per grid cell is closer to zero around the equator (mainly in Africa and South America) and in Central-Europe, among other regions (light blue and yellow color in Fig. 3c). The DPI of PV is higher than the DPI of bioenergy in large parts of sub-Saharan Africa, North America, the eastern Part of South America, Central Asia and the eastern parts of Australia (red color in Fig. 3c). In contrast, the DPI of bioenergy is higher than the DPI of PV for areas such as the western and middle parts of South America, South-East Asia, and some clusters in Europe, Africa and Australia (dark blue color in Fig. 3c).

 

Fig. 3. Development Potential Index (DPI) on abandoned cropland for bioenergy (a), PV-energy (b) and (c) difference between DPI for PV and DPI for bioenergy on (cells of) abandoned cropland. Range for DPI is between 0 and 1, where 1 represent maximum development potential. DPIs are based on Oakleaf et al. [36] and applied to the identified abandoned cropland areas. The difference in DPI is calculated as DPI of PV minus the DPI of bioenergy per grid cell. This means that if a grid cell has a value greater than zero it has a higher DPI for PV than for bioenergy.

3.4. Optimal deployment of bioenergy and PV based on DPIs

By comparing the site-specific DPIs, an optimal distribution of bioenergy and PV deployment for renewable energy production is determined (Fig. 4) for 78 Mha of the initial 83 Mha abandoned cropland. The remaining 5 Mha has no DPI due to the selected size of the window for the data filling methods. The influence of alternative window sizes is explored in a sensitivity analysis. The results show that 68% (53 Mha) of the abandoned cropland have higher suitability for establishment of PV infrastructure, while the other 32% (25 Mha) is more suitable for bioenergy deployment (Fig. 4a). An optimal DPI-based deployment of these two renewable energy options has a primary energy potential of 125 EJ year⁻¹ at a global level, of which 114 EJ year⁻¹ are from PV and 11 EJ year⁻¹ from bioenergy. Therefore, a significantly higher share of the total energy generated will come from PV deployment (91%) compared to bioenergy (9%). We find that the optimal land management of abandoned cropland combining PV and bioenergy will have DPI values greater than 0.1 for all locations (Fig. 4b and c). However, most of the primary energy potential is associated with land areas with a DPI ranging from 0.4 to 0.8, and only a tiny fraction of the land will have a DPI close to 1 (i.e., maximum development potential).

 

Fig. 4. Optimal allocation to abandoned cropland of PV and bioenergy based on the Development Potential Index (DPI). Areas identified with higher DPI for PV in yellow and bioenergy in green (a). Amount of hectares of land (b) and primary energy production (c) for the optimal management of PV (yellow) and bioenergy (green) by ranges of DPIs. Primary energy potential per year (in PJ year⁻¹) (d) and energy yield (in GJ ha⁻¹ year⁻¹) (e) for optimal management of abandoned cropland. (f) Development Potential Index (DPI) for optimal land management on abandoned cropland for renewable energy production by bioenergy and PV. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

While the overall optimal deployment for PV and bioenergy is quite fragmented (Fig. 4a), some trends can be observed. North America and a large part of Central and South America have a higher development potential for PV compared to bioenergy. In Europe, the Mediterranean countries are more suitable for PV than bioenergy deployment. In contrast, bioenergy has larger DPI values in Great Britain, Ireland, Poland, and the former Soviet states except for Russia. In the African continent, bioenergy is a better option in the eastern part of the Horn of Africa, western and central Africa, while PV is the preferred option under the equator. Large parts of Asia have a higher development potential for PV, but some exceptions are in Pakistan, China, and Southeastern Asia. Further, we find that the eastern states of Australia and New Zealand have a higher potential for PV. Along a latitudinal gradient, the areas between 30 and 60°N have a higher PV potential while most of the tropical areas situated around the equator have a higher bioenergy potential. In total, 125 EJ year⁻¹ of primary energy can be generated, at a global mean primary energy potential of 1.6 TJ ha⁻¹ year⁻¹ for an optimal deployment of abandoned cropland areas (Fig. 4d and e). This potential is produced with an average DPI of 0.61 (Fig. 4f). This value is higher than the mean DPI when only deployment of bioenergy (mean DPI 0.53) or PV (mean DPI 0.58) is considered (see Fig. 3a and b), meaning that an optimal allocation of renewable energy production to abandoned cropland on the basis of the DPI offers higher potentials of development.

This primary energy potential from abandoned cropland can be compared to the expected renewable energy deployment in future climate change mitigation scenarios. The projected median renewable primary energy needed to limit global warming to 1.5 °C across different SSPs in 2100 is 224–412 EJ year⁻¹ from biomass and 54–243 EJ year⁻¹ from solar energy [5,6]. Comparing these estimates to our result of bioenergy (11 EJ year⁻¹ under DPI-based optimal land allocation), it is clear that limiting bioenergy production to abandoned cropland is not sufficient. More resources from forest residues, waste products, or bioenergy crops produced on current cropland or pastureland are needed to reach the projections across all SSPs. However, deployment of perennial grasses on abandoned cropland is a promising near-term option that can facilitate a gradual upscale of bioenergy production from second generation feedstocks. On the other hand, our estimate of 114 EJ year⁻¹ for PV represents a large share of the expected PV demand, and in one scenario (SSP4) it can cover all the projected solar energy demand. It also covers large parts of the 2100 demand in SSP1 and SSP2 (83% and 78%, respectively), and 47% of the demand in the energy intensive SSP5 scenario where a delayed energy transition leads to an extensive deployment of renewable energy sources in the second half of the century [6].

3.5. Development Potential Index thresholds and land-energy interactions

There are non-linear relationships between DPI-thresholds, land availability and energy potentials (Fig. 5). Half of the abandoned cropland is located in areas where DPI is higher than 0.53 and 0.60 for bioenergy (Fig. 5a) and PV (Fig. 5b), respectively. With DPI optimized land management, there is relatively higher land availability for energy production at a given DPI threshold (Fig. 5c). Half of the abandoned cropland can be utilized at a DPI threshold of 0.63, with 13 Mha and 26 Mha allocated to bioenergy and PV, respectively. Abandoned cropland with DPI above 0.8 is limited for both bioenergy and PV (2.4 Mha and 1.7 Mha, respectively). Even in the DPI optimized scenario, only 3.8 Mha of the abandoned cropland has DPI above 0.8 (2.3 Mha and 1.5 Mha allocated to bioenergy and PV, respectively). At the same time, 85%, and 95% of abandoned cropland has DPI above 0.4 for bioenergy and PV, respectively. The optimized land management make use of 99% of abandoned cropland above a 0.4 DPI threshold, with 24 Mha and 53 Mha allocated to bioenergy and PV, respectively.

 

Fig. 5. Relationship between DPI levels, land availability and achievable primary energy potentials assuming all abandoned cropland above a given DPI is utilized for energy production. Land availability [Mha] are shown for (a) bioenergy, (b) PV, and (c) for the DPI optimized allocation of bioenergy and PV on abandoned cropland. Maximum land availability differs slightly between scenarios due to variations in DPI data availability. Energy potentials [EJ year⁻¹] are shown for (d) bioenergy, (e) PV, and (f) the DPI optimized allocation of bioenergy and PV on abandoned cropland. Note different scales on y-axis for energy potentials in (d), (e) and (f).

Reaching half of the bioenergy potential requires a DPI threshold of 0.55 and 37 Mha of abandoned cropland, with a mean bioenergy yield of 460 GJ ha⁻¹ year⁻¹ (Fig. 5d). Half the PV potential is achievable with a DPI threshold of 0.61 by utilizing 39 Mha of abandoned cropland, and a mean PV energy yield of 2349 GJ ha⁻¹ year⁻¹ (Fig. 5e). With optimal land management, half of the energy potential is reached at a higher DPI threshold of 0.64 (Fig. 5f). This requires 12 Mha and 25 Mha to produce 5.8 EJ year⁻¹ and 59.5 EJ year⁻¹ for bioenergy and PV, respectively. Mean energy yields increase slightly at a 0.64 DPI threshold to 467 GJ ha⁻¹ year⁻¹ and 2380 GJ ha⁻¹ for PV and bioenergy, respectively.

The top 10 percentiles of the energy potentials are achievable at DPI thresholds of 0.71, 0.74 and 0.76, with land requirements of 8.1 Mha, 7.5 Mha and 9.5 Mha to produce 3.6 EJ year⁻¹, 19 EJ year⁻¹ and 13 EJ year⁻¹ for bioenergy, PV and optimal land management, respectively. In the optimal land management scenario, 1.7 EJ year⁻¹ and 11 EJ year⁻¹ is produced with mean energy yields of 439 GJ ha⁻¹ year⁻¹ and 2481 GJ ha⁻¹ year⁻¹ for bioenergy and PV, respectively. On the other hand, 90% of the total potential is reachable with DPI threshold of 0.38, 0.44 and 0.47, thereby producing 30 EJ year, 161 EJ year⁻¹ and 114 EJ year⁻¹ with land requirements of 69 Mha, 74 Mha and 70 Mha for bioenergy, PV and optimal land management, respectively. With optimal land management, 10 EJ year⁻¹ and 103 EJ year⁻¹ is produced, with mean energy yields of 454 GJ ha⁻¹ year⁻¹ and 2196 GJ ha⁻¹ year⁻¹ for bioenergy and PV, respectively.

While PV energy yields peak at a 0.96 DPI threshold for the optimal land management scenario (2825 GJ ha⁻¹ year⁻¹), we find that mean bioenergy yields peak at a DPI threshold of 0.63 (466 GJ ha⁻¹ year⁻¹). This indicates that market accessibility or land supply elasticity (i.e. the relationship between land use change, and land and crop prices) might be a constraint in some of the highest yielding areas for bioenergy production. Furthermore, we deploy irrigated bioenergy production in areas that are not threatened by water scarcity, which is not directly reflected by the DPI values.

3.6. Uncertainty and limitations

We performed a sensitivity analyses to assess the effects of the DPI data filling method on areas of abandoned cropland with an assigned DPI value and the resulting primary energy potentials. Supplementary Table 1 shows how results vary according to different sizes of the moving window (between 3 × 3 to 31 × 31 cell windows). Across all the window sizes considered, we obtained DPI values for an additional 27–40 Mha. Including these additional cells brings total abandoned cropland with both bioenergy and PV DPIs to 67–80 Mha, with optimal energy potentials ranging between 107 and 129 EJ year⁻¹. While a larger window size increases both the considered land availability and energy potentials, it does not affect the share of land distributed between bioenergy and PV. Allocated shares of abandoned cropland to bioenergy and PV are 31–32% and 68–69% across all considered window sizes, respectively. This indicates an even spread of missing values across the abandoned cropland grids. Likewise, mean bioenergy and PV energy yields show small variations (454–457 GJ ha⁻¹ and 2130–2153 GJ ha⁻¹, respectively). Contributions to optimal total energy production stay at 9% and 91% for bioenergy and PV, respectively.

The recently released ESA CCI-LC products aim to give a more realistic representation of land cover dynamics compared to former datasets by resolving some of their limitations [101,102]. Several previous studies validated the accuracy of the ESA CCI-LC dataset. The global overall accuracy is 71% [103], but with some regional differences. Overall accuracies are 72% in China [104], 62% in Africa [105], 70% in South America [105], 62% in the Arctic [106] and 84% in coastal Eurasia [107]. Cropland classes have the highest global accuracy. Median global user and producer accuracies of cropland classes are 89% and 80%, respectively [103]. The ESA CCI-LC dataset is also found to be broadly consistent with other cropland identification approaches [105,108]. This makes the ESA CCI-LC especially useful for cropland monitoring, with both a relatively high precision in total cropland extent, and a high spatial cropland match. Moreover, the estimates of a detected change in terrestrial surface are based on changes registered over a two year period, so that temporary changes (e.g., a short period of extreme events such as draughts) will not affect the data, and this is considered a strength of the CCI-LC database. An increasing number of studies rely on the ESA CCI-LC products for analysis of effects of land cover changes. Examples include effects of recent land-use and land-cover changes on the terrestrial carbon stocks [102,109], air temperature and humidity effects [110], surface energy budget [109], assessments of global cropland sparing potentials [111], global urban expansion [112], and land cover changes within biodiversity hotspots [113].

The GAEZ model relies on complex databases to provide spatial agro-climatic yields, and both the databases and the assumptions in the model itself have intrinsic uncertainties [33]. Despite this, the GAEZ model is widely tested, validated and used in multiple applications [[114], [115], [116], [117], [118]]. In order to estimate the bioenergy potential on abandoned cropland, the energy-based optimal solution for perennial grasses allocation in combination with high agricultural management levels and mixed water supply system is utilized. Bioenergy potentials are lower if a low or medium agriculture management intensity is considered (e.g., lower inputs of nutrients, lower mechanization of field work, etc.), but higher if irrigation is deployed on all abandoned cropland. More refined estimates of bioenergy potentials can be achieved with regional analysis that considers local trade-offs with economic and environmental factors to identify the best management practice.

Meteorological data from CMCC-CM rely on the atmospheric model ECHAM5 [79] and the ocean model OPA8.2 [80] coupled by the OASIS3 interface [81]. As each of the models are combined to create the CMCC-CM, the datasets from CMCC-CM include the intrinsic uncertainties related to each model and those from their combination [82]. The primary energy potential for PV is estimated based on two main assumptions that are subject to uncertainties: the nominal installed power [27,28,92,119] and the amount of area covered by solar panels [[27], [28], [29]]. These parameters can be adjusted to represent advancements in technologies or by accounting for site-specific conditions. The primary energy potential of PV also depends on solar radiation, ambient temperature, cell temperature, and wind speed, and future changes in background climatic conditions can influence productivity of PV. Nominal installed production, area covered with solar cells and technical parameters like orientation, tilt, and incline of the PV modules also affect estimates of potentials [120]. Typically, PV systems using a tracking system that follow the sun path compared to fixed orientation can gain higher efficiency [121].

Regarding the DPI data, the uncertainty of model outputs as well as the sensitivity to model inputs have been investigated in the original study that produced the DPI dataset [36]. The study also validated the DPIs by investigating recent or planned developments by using more than 6000 data-points and 200,000 km² of mapped locations. A Monte Carlo approach determined the uncertainty, and the mean coefficient of variation for bioenergy DPI (0.137) is higher compared to the mean coefficient of variation for PV DPI (0.048). In general, lower uncertainty is found for higher DPI areas. For PV, high uncertainty is typically associated with areas in remote regions that lacked supporting infrastructure or market access but had higher than average resource potential (i.e., yield criteria in all DPIs). For bioenergy, the highest uncertainty values mainly occurred in regions with lower yields and longer distances to markets or infrastructure.

Additional work can explore the optimal DPI land management on abandoned cropland under the changing climate niche [122]. As energy yield is the most important parameter in the DPI methodology for PV and bioenergy [36], and are depending on climate conditions, future climate changes will likely affect the optimal land allocation and energy potentials. Bioenergy dry mass yields are projected to decrease globally in the 21st century for rain-fed bioenergy crops in RCP4.5 and RCP8.5, with the magnitude of change being almost double in RCP8.5 relative to RCP4.5 [123]. The largest declines are projected in the tropics and subtropics, but there are also increasing yields projected at higher latitudes in parts of the northern hemisphere. Global crop yield losses can to some extents be mitigated through irrigation deployment and genetic modifications [[123], [124], [125]]. For PVs, cell efficiency decreases with increased cell temperatures and cloud cover but increase with increased incoming solar radiation [126]. Global PV energy yields are expected to decrease globally as a response to climate change [69]. For example, mean PV potentials are projected to decrease towards 2100 in RCP8.5 with −10% and −4% in Europe [68] and Africa [89], respectively.

Although our analysis only focused on renewable energy potentials, there are multiple local environmental effects associated with bioenergy or solar PV deployment that are important to consider, and that can differ significantly between the two options. There are several environmental benefits that can be obtained by introducing dedicated bioenergy crops on former cropland. For example, lower nutrient and soil losses [50], and supporting several ecosystem services [44]. On the other hand, producing solar energy with photovoltaic panels on open land is sometimes heavily contested because of nature protection, accessibility, and cultural heritage [42].

Deployment strategies for renewable energy options are highly influenced by political and economic factors [16]. Local, national or international laws, regulations, and subsidies can either promote or inhibit renewable energy deployment. The integration of ESA CCI-LC and DPIs allows assessing opportunity costs and the potential of a land-use conversion from abandoned cropland to energy production. The results of our spatially explicit analysis offer the opportunity to zoom into different local and regional contexts. As the DPI methodology is especially suitable to identify priority regions for new deployment of renewable energy infrastructure [36], our findings show the potential hidden at low risks for rising competition for land and provide a preliminary screening of the key regions with extensive cropland abandonment that can be prioritized for bioenergy or PV deployment. Implementation of these solutions should consider the specific local factors for sustainable land management and energy planning at the local level, and involve environmental and socio-economic indicators other than those included in the analysis (e.g., resource availability, material requirements, biodiversity, social factors, political factors, energy system characteristics or property rights). For example, bioenergy crops are found to improve many indicators of ecosystem services relative to cropland (including biodiversity), whereas PV panels would have a likely more detrimental effects on terrestrial ecosystems on which they are established. Further, the inclusion of other renewable energy sources in the analysis may change the optimal allocation of renewable energy options to specific grid cells, and likely lead to overall increased DPI values on abandoned cropland. For example, abandoned cropland may be more suitable for wind energy production [14], or could contribute to the deployment of new hydropower reservoirs [127].