Agrivoltaics provide mutual benefits across the food–energy–water nexus in drylands

Abstract

The vulnerabilities of our food, energy and water systems to projected climatic change make building resilience in renewable energy and food production a fundamental challenge. We investigate a novel approach to solve this problem by creating a hybrid of colocated agriculture and solar photovoltaic (PV) infrastructure. We take an integrative approach—monitoring microclimatic conditions, PV panel temperature, soil moisture and irrigation water use, plant ecophysiological function and plant biomass production within this ‘agrivoltaics’ ecosystem and in traditional PV installations and agricultural settings to quantify trade-offs. We find that shading by the PV panels provides multiple additive and synergistic benefits, including reduced plant drought stress, greater food production and reduced PV panel heat stress. The results presented here provide a foundation and motivation for future explorations towards the resilience of food and energy systems under the future projected increased environmental stress involving heat and drought. Agrivoltaics can achieve synergistic benefits by growing agricultural plants under raised solar panels. In this article, the authors showed that growth under solar panels reduced tomato and pepper drought stress and increased production, while simultaneously reducing photovoltaic panel heat stress.

Full text

Articles
https://doi.org/10.1038/s41893-019-0364-5
1School of Geography & Development, University of Arizona, Tucson, AZ, USA. 2Office of Research, Development, and Innovation, Biosphere 2, University
of Arizona, Tucson, AZ, USA. 3Department of Environmental Science & Technology, University of Maryland, College Park, MD, USA. 4Tucson Unified School
District, Tucson, AZ, USA. 5The School of Landscape Architecture and Planning, University of Arizona, Tucson, AZ, USA. 6The Southwest Center,
University of Arizona, Tucson, AZ, USA. 7Energy Analysis and Decision Support, National Renewable Energy Laboratory, Golden, CO, USA.
*e-mail: gregbg@email.arizona.edu
A
key challenge to building resilience under a changing and
uncertain climate is maintaining and improving both energy
and food production security. Such efforts are hampered,
in part, by conventional understanding of land use that asserts an
inherent ‘zero-sum-game’ of competition between some forms of
renewable energy—particularly solar PV installations—and agri-
cultural food production. While some farms have adopted renew-
able energy production to assist with their function1, this either–or
discourse drives many policies and development decisions around
conservation practices, land and water allotments for agriculture,
and permitting the establishment of large-scale renewable energy
installations2–5. However, we may require a more holistic and inte-
grated approach centred at the nexus of food, energy and water sys-
tem studies that simultaneously meets increasing energy demands
through decentralized technologies, reduces impacts from the land
use footprint of energy development or immobilization of land
resources for biofuel production (termed ‘energy sprawl’6) and
addresses the need for more efficient food production in diverse
landscapes, all while minimizing water use and environmental
impacts7–10. This type of nexus thinking and research emphasizes
links among water, energy and food resource systems and extends
beyond single-sector approaches to resource management11,12.
Globally, our food systems are vulnerable to projected changes in
climate—primarily changes in the timing and amount of precipita-
tion and rising air temperatures13,14. We grow non-dryland-adapted
food within a dryland climate through an over-reliance on irriga-
tion15,16, and models predict a northward migration in potential
rain-fed agricultural areas based on projected climate change17. In
fact, within the United States, water scarcity alone was a major driver
in the conversion of more than 20,000 acres of former croplands in
southern California to renewable energy development in a single
year18, as the lack of water was making agriculture non-economi-
cally viable. Many areas across the globe, including North, Central,
and South America, the Middle East and North Africa, have seen a
shift to increased aridity and are projected to see continued aridity
throughout the century19,20. These regions are also facing increas-
ing water scarcity that places conventional agriculture and farmland
at risk, and projected climate change has been estimated to reduce
food production by 8–45% across Africa and Southeast Asia21–24.
The resulting increases in demand for irrigation will probably com-
pound existing water insecurities experienced globally. Our already
strained freshwater supply is likely to see additional extraction, not
only for future agricultural land use to keep pace with population
and economic growth, but also to match the increased atmospheric
demand under these projected climate changes25.
Our energy system may not be resilient under forecasted cli-
mate change26. Higher air temperatures can reduce the efficiency
and maximum capacity of thermal power plants27,28, and changes
in the seasonality, availability and temperature of water resources
can constrain the operations of hydropower29–31 and thermal power
plants32–34. Globally climate impacts could reduce thermal and hydro-
power capacity by 20% for individual power plants35–37, although
grid reliability metrics indicate a smaller impact38. Drought-proof
technologies that require no water for operations, such as wind and
PV, could provide a solution for enhanced resilience under uncer-
tain water resource conditions while also cutting down on green-
house gases emission—a primary cause of climate change.
Electricity production from large-scale PV installations has
increased exponentially in recent decades4,39,40, signifying an increase
in the cost-effectiveness and grid suitability of this technology2,41. In
the United States, solar development is projected to grow substan-
tially. By 2030, solar installation could reach 330 GW of installed
Agrivoltaics provide mutual benefits across the
food–energy–water nexus in drylands
Greg A. Barron-Gafford 1,2*, Mitchell A. Pavao-Zuckerman3, Rebecca L. Minor1,2, Leland F. Sutter1,2,
Isaiah Barnett-Moreno1,2, Daniel T. Blackett1,2, Moses Thompson1,4, Kirk Dimond 5,
Andrea K. Gerlak1, Gary P. Nabhan6 and Jordan E. Macknick7
The vulnerabilities of our food, energy and water systems to projected climatic change make building resilience in renewable
energy and food production a fundamental challenge. We investigate a novel approach to solve this problem by creating a
hybrid of colocated agriculture and solar photovoltaic (PV) infrastructure. We take an integrative approach—monitoring micro-
climatic conditions, PV panel temperature, soil moisture and irrigation water use, plant ecophysiological function and plant
biomass production within this ‘agrivoltaics’ ecosystem and in traditional PV installations and agricultural settings to quantify
trade-offs. We find that shading by the PV panels provides multiple additive and synergistic benefits, including reduced plant
drought stress, greater food production and reduced PV panel heat stress. The results presented here provide a foundation
and motivation for future explorations towards the resilience of food and energy systems under the future projected increased
environmental stress involving heat and drought.
NATURE SUSTAINABILITY | www.nature.com/natsustain

Articles NaTure SuSTaiNabiliTy
capacity (to meet 14% of national demands), with 209 GW expected
to be ground-mounted solar, which would require approximately
8,000 km2 of land, including agricultural land42,43.
Drylands of the southwestern United States are among the best
positioned for supporting renewable energy through PV because of
the abundance of sunlight44,45, but projections of increasing ambi-
ent temperatures dampen this renewable energy source potential
because of PV panel sensitivity to increases in temperature. While
PV panels vary, their temperature coefficient—the rate of perfor-
mance decline for every 1 °C increase in temperature >25 °C—
illustrates that PV panel efficiency decreases by an average of
~0.6% oC–1)46–48. Additionally, recent research has found that larger
PV arrays cause a ‘heat island’ effect that warms the area within the
installations49, creating a negative feedback of additional warming.
As with the urban heat island effect, landscape change shifts eco-
system structure from one dominated by vegetation to one char-
acterized by a blend of built structures and vegetation, which alters
the energy balance of absorption, storage and release of short- and
long-wave radiation50. Incoming solar energy is either reflected
back to the atmosphere or absorbed, stored and later re-radiated
in the form of latent or sensible heat51 (Fig. 1, blue or red and
orange arrows, respectively). Within natural ecosystems, vegetation
reduces heat gain and storage in soils by creating surface shading,
though the degree of shading varies among plant types (Fig. 1a)52.
Transitions of liquid water-to-water vapour loss to the atmosphere
through evapotranspiration—the combined water loss from soils
(evaporation) and vegetation (transpiration)—use energy absorbed
by vegetation and surface soils. Because many PV installations have
gravel as groundcover, with little to no vegetation, they have little to
no means of energy dissipation through latent heat exchange (Fig. 1,
transition from a to b), and thus are subjected to more sensible heat.
Sustainable development practices of low-impact urban design
counter the urban heat island effect with strategic planting that
reintroduces latent heat exchange of energy by way of plant tran-
spiration. How might a similar model be applied to a PV heat
island? Restoration ecology suggests that there is an important role
for ‘novel ecosystems’—non-historical assemblages of species that
result from the combined effects of shifts in abiotic conditions, land
cover change and environmental management53. Novel ecosystems
serve important functional roles, contribute to the provision of eco-
system services and enhance well-being in human-dominated and
built landscapes54,55.
Q* + QF = QH + QE + ∆Qs + ∆QPV (WM
–2
)
Q* = net all-wave radiation (solar and terrestrial)
QF = anthropogenic heat flux
a b
c
QH = sensible heat flux (atmospheric heating)
QE = latent heat flux (or evapotranspiration)
∆Qs = net storage heat flux
QPV = energy transferred through
energy production
Soil moisture
Fig. 1 | Illustration of changes in midday energy exchange with transitions from natural systems, solar PV arrays and a colocated agrivoltaic system.
a,b, Assuming equal rates of incoming energy from the sun (broken yellow arrows), a transition from a vegetated ecosystem (a) to a solar PV installation
(b) will significantly alter the energy flux dynamics of the area because of the removal of vegetation, and thus the latent heat fluxes (blue arrows). This
leads to greater sensible heat fluxes (red and orange arrows), which yield higher localized temperatures. c, Reintroduction of vegetation, in this case
agricultural plants, restores latent heat fluxes and should reduce sensible heat loss to the atmosphere. Energy re-radiation from PV panels (teal arrows)
and energy transferred to electricity (green arrows) are also shown. Arrow size and abundance correspond to the magnitude of the effect.
Credit: Illustration modified from ref. 49, Springer Nature Ltd.
NATURE SUSTAINABILITY | www.nature.com/natsustain

Articles
NaTure SuSTaiNabiliTy
We present here a novel ecosystems approach—agrivoltaics—
to bolster the resilience of renewable energy and food production
security to a changing climate by creating a hybrid of colocated agri-
culture and solar PV infrastructure, where crops are grown in the
partial shade of the solar infrastructure12,41,56–66 (Fig. 1c). We suggest
that this energy- and food-generating ecosystem may become an
important—but as yet quantitatively uninvestigated—mechanism
for maximizing crop yields, efficiently delivering water to plants
and generating renewable energy in dryland environments. We
demonstrate proof of concept for agrivoltaics as a food–energy–
water system approach in drylands by simultaneously monitoring
the physical and biological dimensions of the novel ecosystem. We
hypothesized that colocating solar and agricultural could yield sev-
eral significant benefits to multiple ecosystem services, including
(1) water: maximizing the efficiency of water used for plant irri-
gation by decreasing evaporation from soil and transpiration from
crop canopies49,67 and (2) food: preventing depression in photosyn-
thesis due to heat and light stress57,68,69, thus allowing for greater
carbon uptake for growth and reproduction. An additional benefit
might be (3) energy: transpirational cooling from the understorey
crops lowering temperatures on the underside of the panels, which
could improve PV efficiency49.
We focused on three common agricultural species that represent
different adaptive niches for dryland environments: chiltepin pepper
(Capsicum annuum var. glabriusculum), jalapeño (C. annuum var.
annuum) and cherry tomato (Solanum lycopersicum var. cerasi-
forme). We created an agrivoltaic system by planting these species
under a PV array—3.3 m off the ground at the lowest end and at
a tilt of 32°—to capture the physical and biological impacts of this
approach. Throughout the average three-month summer growing
season we monitored incoming light levels, air temperature and
relative humidity continuously using sensors mounted 2.5 m above
the soil surface, and soil surface temperature and moisture at 5-cm
depth. Both the traditional planting area (control) and agrivoltaic
system received equal irrigation rates, and we tested two irrigation
scenarios—daily irrigation and irrigation every 2 d.
Results
The amount of incoming photosynthetically active radiation (PAR)
was consistently greater in the traditional, open-sky planting area
(control plot) than under the PV panels (Fig. 2a). This reduction
in the amount of incoming energy under the PV panels yielded
cooler daytime air temperatures, averaging 1.2 + 0.3 °C lower in the
agrivoltaics system over the traditional setting. Night-time tem-
peratures were 0.5 + 0.4 °C warmer in the agrivoltaics system over
the traditional setting (Fig. 2b). Photosynthetic rates, and therefore
growth and reproduction, are also regulated by atmospheric dry-
ness, as represented by vapour pressure deficit (VPD) where lower
VPD indicates more moisture in the air. VPD was consistently lower
in the agrivoltaics system than in the traditional growing setting,
averaging 0.52 + 0.15 kPa lower across the growing season (Fig. 2c).
Having documented that an agrivoltaic installation can significantly
reduce air temperatures, direct sunlight and atmospheric demand
for water relative to nearby traditional agricultural settings, we
address several questions regarding impacts of the food–energy–
water nexus system.
Potential impacts for food production. We found that agrivoltaic
system conditions impacted every aspect of plant activity, though
results and significance varied by species. We used three different
agricultural plants from the same family (Solanaceae) that repre-
sent different adaptive niches for dryland environments (Fig. 3).
Capsicum annuum var. glabriusculum is native to southern North
America and northern South America, and has the adaptation of
growing in the shade of overstorey ‘nurse trees’ due to the high irra-
diance and temperatures characteristic of the region70. Cumulative
CO2 uptake in chiltepin was 33% greater in the agrivoltaic instal-
lation (Fig. 3a), but there was no difference in the water use effi-
ciency (ratio of daily CO2 uptake to transpirational water loss) of
the plants (Fig. 3b), indicating that transpiration was equally greater
in chiltepin grown in the agrivoltaic system. As a result, total chilt-
epin fruit production was three times greater under the PV panels
in an agrivoltaic system (Fig. 3c). This matches the adaptation of
this small-leaved desert shrub and previous studies growing chilt-
epin under artificial shade (but not in an agrivoltaic system)70. We
also chose C. annuum var. annuum), a sister variety to chiltepin that
has been widely domesticated across large biogeographic space and
is of greater commercial value71. Cumulative CO2 uptake in jala-
peño was 11% lower in the agrivoltaic system than in the traditional
growing area (Fig. 3a), suggesting a light limitation in this setting.
However, water use efficiency was 157% greater in the agrivol-
taic system (Fig. 3b). Ultimately, total fruit production was nearly
equal between treatments (Fig. 3c), but this was attained with 65%
less transpirational H2O loss. Finally we chose S. lycopersicum var.
cerasiforme) because it is very heat sensitive, in that summer flower
production is accompanied by abortion due to excessive tempera-
tures. Cumulative CO2 uptake was 65% greater in the agrivoltaic
installation than in the traditional growing area, and water use
efficiency was also 65% greater, indicating that transpirational
water loss was equal between the treatment areas, so the increased
productivity we find in an agrivoltaic system is probably due to an
alleviation of multiple stress interactions from heat and atmospheric
drought. Ultimately, total fruit production was twice as great under
the PV panels of the agrivoltaic system than in the traditional grow-
ing environment.
Potential impacts for water savings. We assessed the impacts of
irrigation water savings in terms of the relative amount of moisture
1,200
a
b
c
1,000
Midday PAR
(
µ
mol m–2 s–1)
800
600
400
200 Control
Agrivoltaics
1
Difference in air temp.
(agrivoltaic – control, °C)
0
–2
–1
Daytime maximum
Night-time maximum
0.0
Difference in VPD
(agrivoltaic – control, kPa)
–0.2
–0.6
–0.4
–0.8
–1.0
May Jun
2017
Jul Aug
Fig. 2 | Micrometeorological impacts of colocation of agriculture and solar
PV panels (agrivoltaic) over traditional (control) installations. a, Average
daily light levels in terms of PAR. b, Differences in ambient air temperature
between daytime and night-time. c, Atmospheric dryness, VPD.
NATURE SUSTAINABILITY | www.nature.com/natsustain

Articles NaTure SuSTaiNabiliTy
that remained in the soil after each irrigation event in a tradi-
tional, or control, growing area versus under an agrivoltaic system
(Fig. 4). We detected the greatest influence of the agrivoltaic sys-
tem on soil moisture savings when we irrigated every 2 d, as soil
moisture remained ~15% greater (3.2% volumetric units) than
in the control setting (Fig. 5a,c) before the subsequent irrigation
event. Nevertheless, even with daily irrigation the agrivoltaic sys-
tem remained 5% greater (1.0% volumetric units) before the sub-
sequent irrigation event than in the control setting (Fig. 5b,d).
Importantly, soil moisture levels in the agrivoltaic setting after 2 d
remained above the driest points seen in the control setting after
daily irrigation events, suggesting that even more reduced irrigation
in an agrivoltaic system may be possible. The potential reduction
in water use within agrivoltaics could be substantial and warrants
further research in future studies, especially given the uncertainties
in projected future rainfall and water allocations.
Potential impacts for improved renewable energy production.
Given the inherent sensitivity of PV panels to temperature, any
cooling of panels below current daytime temperatures >30 °C will
positively impact its efficiency46,47. We found that PV panels in a
traditional ground-mounted array were significantly warmer dur-
ing the day and experienced greater within-day variation than
those over an agrivoltaic understorey (Fig. 6a). We attribute these
lower daytime temperatures in PV panels in the agrivoltaic system
(Fig. 6b) to the greater balance of latent heat energy exchange from
plant transpiration relative to sensible heat exchange from radia-
tion from bare soil (the typical installation method). Across the core
growing season, PV panels in an agrivoltaic system were ~8.9 + 0.2 °C
cooler in daylight hours. This reduction in temperature can lead to
an increase in system performance. Using the system advisor model
(SAM) for a traditional and a colocation PV system in Tucson, AZ,
we calculated that temperature reductions documented here in the
growing months of May–July from the colocation system led to a
3% increase in generation over those months, and a 1% increase in
generation annually.
Discussion
Together, these results suggest that the novel colocation of agricul-
ture and PV arrays could have synergistic effects that support the
production of ecosystem services such as crop production, local
climate regulation, water conservation and renewable energy pro-
duction. Our results suggest that an agrivoltaic colocation design
not only mitigates energy balance challenges associated with the
12
a
*
*
*
Control
Agrivoltaics
10
8
6
Cumulative uptake
(gCO2 m–2 d–1)
4
2
0
10
Daily water use efficiency
(gCO2 d–1/gH2O d–1)
8
6
4
2
Fruit production
(cumulative, per individual)
0
40
30
20
10
0Jalapeno TomatoChiltepin pepper
*
*
*
*
b
c
~
Fig. 3 | Plant ecophysiological impacts of colocation of agriculture and
solar PV panels versus traditional installations. a, Average daily cumulative
CO uptake through photosynthesis per unit leaf area. b, Daily water use
efficiency, as estimated by the amount of carbon uptake relative toleaf-
level water loss through transpiration (per unit leaf area). c, Cumulative
fruit production—number of fruits per individual plant. Results are shown
for chiltepinpeppers, jalapeños and tomatoes grown in a traditional setting
(control) and in the colocated agrivoltaics system. Bars represent +1 s.e.m.,
and an asterisk indicates a significant difference (P<0.001).
5.5 m
1 m spacing
9.1 m9.1 m 6 m
N
18.2 m
18.2 m
Agriculture control site Agrivoltaics system
Photovoltaics control site
Fig. 4 | Map of the experimental area, which consisted of an agricultural
control site, a traditional ground-mounted PV installation and an
agrivoltaic system site. The solid red, blue and hashed circles represent
tomato, jalapeño and chiltepin plants, respectively. The black square
represents the location of the meteorologicalmeasurement station, and
the solid lines extending from the square represent the locations of the soil
moisture sensors. The dashed lines extending from the square represent
the locations of the thermistors adhered to the PV panels for temperature
measurement.
NATURE SUSTAINABILITY | www.nature.com/natsustain

Articles
NaTure SuSTaiNabiliTy
development of a PV site, but also increases the collective ecosys-
tem services associated with an area2. We should no longer follow
the narrow understanding of land use that has averred a zero-sum-
game of competition between renewable energy and agricultural
food production. In fact, we have shown that each portion of the
food–energy–water nexus can respond positively to the coloca-
tion of these seemingly disparate needs. In this novel ecosystem,
plants growing in an agrivoltaic setting (under PV) receive less
light, but this has now been shown to be associated with positive
trade-offs in terms of reduced evaporative loss of soil moisture in a
dryland area. The efficacy and extent of positive effect was depen-
dent on the plant species. Growing food crops in an agrivoltaic
system led to increased CO2 uptake and fruit production in two
of three species, and the one species that did not exhibit higher
production achieved equal production with only about 35% of the
transpirational water loss. At the same time, that transpirational
water loss also created an energy balance shifted more towards
latent heat exchange and less sensible heat flux to the atmosphere
in the move from traditional agriculture to an agrivoltaic system.
This resulted in the PV panels in an agrivoltaic setting being sig-
nificantly cooler in the daytime—a positive trade-off for shading
a vegetative understorey, which should lead to increased renew-
able energy production—and longer retention times of irrigation
waters within the soil.
Additional species should be explored to capture a wider under-
standing of which plant functional types are most appropriate
for this new type of food production; additional solar infrastruc-
ture designs and configurations should be considered, to better
understand trade-offs in energy output and plant productivity;
and additional installations around a biogeographic gradient
should be explored to quantify the relative impacts, as have been
documented here. It is possible that alternative solar configura-
tion, crop and location combinations could lead to disadvanta-
geous outcomes, such as excess soil moisture, crop yield reductions
or increased risks to solar infrastructure. Future field sites could
explore these variations and thereby contribute to a more com-
prehensive understanding of agrivoltaics opportunities. Hitherto
uninvestigated for agrivoltaics are the potential for the restoration
of endemic plant communities to provide increases in solar panel
efficiencies, the retrofitting potential of the groundcover of existing
solar facilities to accommodate food crops or endemic plant com-
munities, and how pollinator habitat planted underneath arrays
could benefit local agriculture. Also unexplored are issues tied to
the physical and social impacts of farm labourers working in an
agrivoltaic system. To date we have found no ill effects, and future
studies could quantify these impacts through the metric of human
thermal comfort (HTC)72, which takes into account not only air
temperature but also sun exposure. Given the milder microclimate
under PV panels within an agrivoltaic system, we hypothesize that
HTC would be greater than in a control system, and this could be
particularly important in dryland environments where rates of
heat stroke and heat-related death among farm workers are espe-
cially high. Economically, this novel microclimate may also extend
growing seasons and protect against untimely frosts. The land-
leasing opportunity may additionally provide revenue to farmers
to ward off development pressures and keep food costs down. All
of these impacts—water scarcity, environmental sensitivity of our
food crops, the efficiency of PV panels and the HTC of the people
that bring food to market—are especially vulnerable to projected
climate change patterns and extremes.
This study represents the first experimental and empirical exam-
ination of the potential for an agrivoltaic system to positively impact
60
b
a
Traditional PV panels
Agrivoltaic PV panels
50
40
30
20
10
Photovoltaic panel surface
temperature (°C)
0
0
–5
–10
–15
–20
1 May 8 May 15 May 22 May 29 May 5 Jun
2017
19 Jun12 Jun
Daytime PV panel surface temp.
(agrivoltaic – traditional, °C)
Fig. 6 | Impacts of colocation of agriculture and solar PV panels
(agrivoltaic) over traditional ground-mounted installations on the surface
temperature of PV panels. a, Thirty-minute average photovoltaic panel
temperature, as measured by a thermistor placed on the rear of panels.
b, Differences in panel temperature between the agrivoltaic and traditional
ground-mounted settings. Negative values indicate the degree to which
panels in the agrivoltaic setting were cooler. Positive values—indicating
warmer conditions—occurred during the night-time when the PV systems
were not operating.
35
a b
c d
Control
Agrivoltaics
Control
Agrivoltaics
Soil moisture (%)
30
25
20
15
12
10
8
6
4
Soil moisture
(agrivoltaic/control, %)
2
0
1 May
5 May
9 May
13 May
17 May
2017
30 Jun
4 Jul
8 Jul
12 Jul
Fig. 5 | Impacts of colocation of agriculture and solar PV panels
(agrivoltaic) over traditional (control) installations on irrigation
resources, as indicated by soil moisture. a,b, Thirty-minute average
volumetric water content (soil moisture) in the top 5 cm of the soil in the
agrivoltaic and control settings. c,d, Differences between soil moisture in
an agrivoltaic setting and in control plots, where positive values indicate
additional moisture in the agrivoltaic setting. a,c, A period when plots were
watered every two days. b,d, A period when plots were watered every day.
NATURE SUSTAINABILITY | www.nature.com/natsustain

Articles NaTure SuSTaiNabiliTy
each component of the food–energy–water nexus. Our results from
a dryland system indicate a reduction in daytime temperatures of
the solar panels (energy) and microclimate under the panels (food),
and a dampening in the diurnal fluctuations of each and day-to-
day fluctuations in soil moisture in irrigated agriculture (water).
Together, our findings suggest that a dryland agrivoltaic system may
be a resilient energy and food system that has reduced vulnerabili-
ties to future climate variability. However, there are probable barriers
to wider adoption, which include challenges associated with some
forms of mechanized farming and harvest and the additional costs
associated with elevating PV arrays to allow for food production
in the understorey. An integrated approach to the physical and
social dimensions of our food and energy systems is key in sup-
porting decision making regarding PV development and sustain-
able food and energy production in a changing world. The results
presented here provide a foundation, direction and motivation for
future explorations towards resilience of food and energy systems
in the future.
Methods
Site description. We established the Biosphere 2 Agrivoltaics Learning Lab in
August 2016. e site is operated by the University of Arizona, and is situated
on the ground of the Biosphere 2 research centre north of Tucson, AZ, USA
(32.578989 °N, 110.851103 °W, elevation 1,381 m above sea level). e climate
in Tucson is hot desert (Koppen classication BWh), which experiences a mean
annual temperature of 21.6 °C and is characterized as having bimodal precipitation
with a summer monsoon and winter rains. Average precipitation is limited to
<30 cm, but the magnitude and timing of storms have increasingly varied in recent
decades73. Summers are hot, with air temperature regularly averaging 38 °C and
soil surface temperatures exceeding 45 °C. Winter temperatures are moderate,
with occasional light frosts in January. Summer season agriculture in this region
is primarily a mix of vegetables, with tomato, pepper, herbs, eggplant and melon
being most prominent.
The site involves replicated rows of agricultural crop species growing in either
traditional, open-sun growing conditions or under a raised solar PV panel array
(agrivoltaics; Fig. 4). Each of these areas is approximately 9.1 × 18.2 m2. The fixed
panels within the agrivoltaic system are 3.3 m above the soil surface at their lowest
point, whereas they are only about 0.3–1.0 m above the ground in the traditional
PV configuration. All panels face south, at a latitude angle of 32°48. There is 1 m of
spacing between each row of PV panels.
Both sites were excavated down to a depth of approximately 25.5 cm, and the
native soil was replaced and amended with an organic ‘Garden Blend’, which is
a mixture of 75% organic compost and 25% sandy soil (Tank’s Green Stuff ). An
irrigation system was established that delivered equal amounts—in terms of rate
and cumulative application—across the control and agrivoltaics plots. Equality
in irrigation delivery was confirmed on two occasions by collecting drip water
and measuring total volumes at random emitters across the plots. We planted
42 replicate plants of each of three agricultural species from the same family
(Solanaceae) that represent different adaptive niches for dryland environments:
C. annuum var. glabriusculum, C. annuum var. annuum and S. lycopersicum var.
cerasiforme; Fig. 4). All of the replicate individuals originated from the same seed
source—a homogenized collection of seeds from fruits produced by the previous
year’s growth. Seeds were planted within a matrix of the same ‘Garden Blend’ in
February 2017 in a greenhouse, and were then transplanted to the research site
in March 2017. We were cognisant of the potential for inducing error by studying
plants along the border of our treatment due to ‘edge effects’74. We avoided this
issue by selecting plants that were at least three rows in from any edge of the
experimental array.
For most of the experiment, irrigation was delivered daily at a rate of
3.79 l min–1 for 30 min by a multi-valve irrigation system (Rain Bird ESP-Modular
Controller), but in May 2017 we reduced irrigation to an alternate-day schedule
to quantify rates of water use under this water-saving schedule. Irrigation was
supplied through standard 1.27-cm polyethylene drip irrigation tubing. We
conducted calibrations on the uniformity of the irrigation system’s delivery twice
per year. To confirm that the irrigation emitters were delivering equal amounts of
water despite variable distances from the irrigation control box, we used graduated
cylinders to calculate rates in terms of volumes per minute. We maintained these
measurement installations for one full growing season, to capture variation due
to sun angle and extremes associated with hot and cold periods at the edge of the
growing season.
At the end of the growing season in August 2017, we harvested all of the fruits
for each of the ten replicate study plants per species. In so doing, we captured the
productivity of each plant in terms of marketable produce. We present here the
production per individual, to underscore the impacts of a changing microclimate
due to the agrivoltaics system approach on productivity at an organismal level.
Monitoring equipment and variables monitored. Ambient air temperature (°C)
and relative humidity (%) were measured with a shaded, aspirated temperature
probe 2.5 m above the soil surface (Vaisala HMP60). Importantly, the temperature
probes used within the agrivoltaic and control settings were cross-validated for
precision (closeness of temperature readings across all probes) at the onset of
the experiment. We also monitored incoming PAR (LI-190R, LI-COR) at 2.5 m
above the soil surface. Both of these probes were mounted on a post placed within
the centre of each installation, to avoid any variance due to edge effects around
each plot. We monitored volumetric water content and soil temperature at 5-cm
depth (ECH2O 5TM, METER Group) at six points across each of the control and
agrivoltaic system sites (Fig. 4). Data across the six points were averaged to give a
single representative value for each time period for each site.
Finally, we monitored solar panel temperatures using precision integrated-
circuit temperature sensors (LM35CA thermistor, Texas Instruments) adhered
to the rear of each of six different solar panels75. To compare the temperatures
of panels in the agrivoltaic setting to traditional ground-mounted solar panels,
we replicated this measurement scheme on six panels within a solar array
approximately 10 m away in the same research area of Biosphere 2 (Fig. 4). All
of these measurements were recorded at 30-min intervals throughout a 24-h
day, and data were recorded on a data-logger (CR1000, Campbell Scientific).
We calculated averages of daytime PV panel temperatures using all data from
daylight hours, when PAR was >10 μmol m−2 s−1. We used SAM parameterized for
Tucson, AZ, USA to quantify normal power generation for an example 200 kW
DC system76,77. For this simulation, we used standard SAM defaults for a PV
array of SunPower-X21-335 (mono-crystalline Silicon) modules with a nominal
efficiency of 20.5521%. The only variable that differed between the adjacent
traditional and agrivoltaic installations was PV panel temperature. A traditional
PV installation would generate 373 MWh per year (21.4% capacity factor),
whereas the agrivoltaic installation, with the reduced temperatures shown here,
would generate 377 MWh per year (21.6% capacity factor). This equates to an
increase of approximately 1% per year in annual generation based on only these
three months of documented cooling.
Leaf-level measurements of CO2 and water exchange. Measurements of leaf-
level net photosynthesis (Anet) and transpiration were conducted using a LI-6400
portable photosynthesis system (LI-COR)78–81,82. A red–blue light source
(LI-6400-02b) attached to the leaf cuvette provided constant irradiance
of ambient light levels for each measurement area (open sun versus shade
under the PV panels). The cuvette (CO2) was held constant at 400 ppm
across all measurements. Cuvette air temperature was set to match that of
ambient conditions at each measurement time point. Hourly measurements
were conducted between 05:00 and 21:00 mean solar time—for a total of
17 measurement periods—to capture the full daily carbon assimilation and
water loss period. Once chamber conditions and gas-exchange rates of Anet had
stabilized, the two infrared gas analysers within the instrument were matched,
and gas-exchange data were logged five times across a 1-min period and
averaged. For each of the treatments (control and agrivoltaics), we measured
five replicates of each of the three plant types—for a total of 30 individuals. We
conducted these measurements over the course of a 5-d period in the middle of
the growing season, to capture instantaneous rates of leaf-level gas exchange and
to gain insight into plant performance. All leaves that were within the 2 × 3-cm
cuvette for gas-exchange measurements were harvested after measurements
and stored in paper envelopes in a chilled cooler for transport to the laboratory,
so that we could correct our measurements on a per-unit leaf area basis. We
obtained wet leaf mass, and then sampled leaf area was determined using an LI-
3100C area meter (LI-COR). Samples were then air-dried to obtain dry leaf mass.
These measurements of CO2 assimilation and water loss were used to infer
daily cumulative by first taking the instantaneous measures, which have units of
μmol m−2 s−1, and up-scaling these to the hourly estimates. We then accumulated
these hourly values for the entire daytime period.
Throughout this procedure, we captured the productivity of each plant in terms
of carbon uptake at the leaf scale.
Statistical analysis. Comparisons of cumulative CO2 uptake, daily water use
efficiency (WUE; daily CO2 uptake/daily transpirational water loss) and fruit
production between the control and agrivoltaic treatment were made using Tukey’s
honestly significant difference (HSD) test. Standard errors to calculate HSD were
made using pooled values at either the daily (CO2 uptake and WUE) or growing
season (fruit production) scale. We used midnight and noon values to examine
maximum and minimum, respectively, differences in PV panel temperatures
between the agrivoltaic and traditional ground-mounted settings. Comparisons
among the sites were made using the same Tukey’s HSD test.
Reporting Summary. Further information on research design is available in the
Nature Research Reporting Summary linked to this article.
Data availability
The data that support the findings of this study are available from the
corresponding author on request.
NATURE SUSTAINABILITY | www.nature.com/natsustain

Articles
NaTure SuSTaiNabiliTy
Received: 28 June 2018; Accepted: 22 July 2019;
Published: xx xx xxxx
References
1. Beckman, J. & Xiarchos, I. M. Why are Californian farmers adopting
more (and larger) renewable energy operations? Renew. Energy 55,
322–330 (2013).
2. Hernandez, R. R. etal. Environmental impacts of utility-scale solar energy.
Renew. Sustain. Energy Rev. 29, 766–779 (2014).
3. Hernandez, R. R., Hoacker, M. K. & Field, C. B. Land-use eciency of big
so lar. Environ. Sci. Technol. 48, 1315–1323 (2014).
4. Renewables 2014 Global Status Report (REN21 Secretariat, 2014).
5. Hoacker, M. K., Allen, M. F. & Hernandez, R. R. Land-sparing opportunities
for solar energy development in agricultural landscapes: a case study of the
great Central Valley, CA, United States. Environ. Sci. Technol. 51,
14472–14482 (2017).
6. Trainor, A. M., McDonald, R. I. & Fargione, J. Energy sprawl is the
largest driver of land use change in United States. PLoS ONE 11,
7896–7906 (2016).
7. Bazilian, M. etal. Considering the energy, water and food nexus: towards an
integrated modelling approach. Energy Policy 39, 7896–7906 (2011).
8. Albrecht, T. R., Crootof, A. & Scott, C. A. e water-energy-food nexus: a
systematic review of methods for nexus assessment. Environ. Res. Lett. 13,
043002 (2018).
9. Harwood, S. A. In search of a (WEF) nexus approach. Environ. Sci. Policy 83,
79–85 (2018).
10. Biggs, E. M. etal. Sustainable development and the water-energy-food nexus:
a perspective on livelihoods. Environ. Sci. Policy 54, 389–397 (2015).
11. Bogardi, J. J. etal. Water security for a planet under pressure: interconnected
challenges of a changing world call for sustainable solutions. Curr. Opin.
Environ. Sustain. 4, 35–43 (2012).
12. Dupraz, C. etal. Combining solar photovoltaic panels and food crops for
optimising land use: towards new agrivoltaic schemes. Renew. Energy 36,
2725–2732 (2011).
13. Jaeger, W. K. etal. Finding water scarcity amid abundance using human-
natural system models. Proc. Natl Acad. Sci. USA 114, 11884–11889 (2017).
14. Lobell, D. B., Schlenker, W. & Costa-Roberts, J. Climate trends and global
crop production since 1980. Science 333, 616–620 (2011).
15. Elliott, J. etal. Constraints and potentials of future irrigation water availability
on agricultural production under climate change. Proc. Natl Acad. Sci. USA
111, 3239–3244 (2014).
16. Liu, J. etal. Achieving sustainable irrigation water withdrawals: global
impacts on food security and land use. Environ. Res. Lett. 12, 104009 (2017).
17. Bradford, J. B. etal. Future soil moisture and temperature extremes imply
expanding suitability for rainfed agriculture in temperate drylands. Sci. Rep.
7, 12923 (2017).
18. Kasler, D. More California farmland could vanish as water shortages loom
beyond drought. e Sacramento Bee (26 November 2015).
19. Waha, K. etal. Climate change impacts in the Middle East and Northern
Africa (MENA) region and their implications for vulnerable population
groups. Reg. Environ. Change 17, 1623–1638 (2017).
20. Reyer, C. P. O. etal. Climate change impacts in Latin America and the
Caribbean and their implications for development. Reg. Environ. Change 17,
1601–1621 (2017).
21. Tumbo, S. D. etal. Impact of projected climate change on agricultural
production in semi-arid areas of Tanzania: a case of same district. Afr. Crop
Sci. J. 20, 453–463 (2012).
22. Adhikari Umesh, N., Pouyan, A. & Woznicki, S. A. Climate change and
eastern Africa: a review of impact on major crops. Food Energy Secur. 4,
110–132 (2015).
23. Knox, J., Hess, T., Daccache, A. & Wheeler, T. Climate change impacts
on crop productivity in Africa and South Asia. Environ. Res. Lett. 7,
034032 (2012).
24. Schlenker, W. & Lobell, D. B. Robust negative impacts of climate change on
African agriculture. Environ. Res. Lett. 5, 014010 (2010).
25. Jiménez Cisneros, B. E. etal. in Climate Change 2014: Impacts, Adaptation,
and Vulnerability (eds Field, C. B. etal.) Ch. 3, 229–269 (IPCC, Cambridge
Univ. Press, 2014).
26. Bevan, G. Renewable energy and climate change. Signicance 9, 8–12 (2012).
27. Jaglom, W. S. etal. Assessment of projected temperature impacts from
climate change on the U.S. electric power sector using the Integrated
Planning Model. Energy Policy 73, 524–539 (2014).
28. Gonzalez-Diaz, A. etal. Eect of the ambient conditions on gas turbine
combined cycle power plants with post-combustion CO2 capture. Energy 134,
221–233 (2017).
29. Hamlet, A. F., Lee, S.-Y., Mickelson, K. E. B. & Elsner, M. M. Eects of
projected climate change on energy supply and demand in the Pacic
Northwest and Washington State. Clim. Change 102, 103–128 (2010).
30. Turner, S. W. D., Ng, J. Y. & Galelli, S. Examining global electricity supply
vulnerability to climate change using a high-delity hydropower dam model.
Sci. Total Environ. 590, 663–675 (2017).
31. Madani, K. & Lund, J. R. Estimated impacts of climate warming on
California’s high-elevation hydropower. Clim. Change 102, 521–538 (2010).
32. U.S. Energy Sector Vulnerabilities to Climate Change and Extreme Weather
(United States Department of Energy, 2013).
33. Rogers, J. etal. Water-Smart Power: Strengthening the U.S. Electricity System
in a Warming World (2013).
34. Macknick, J. etal. Water and Climate Impacts on Power System Operations:
e Importance of Cooling Systems and Demand Response Measures (2016).
35. van Vliet, M. T. H., Wiberg, D., Leduc, S. & Riahi, K. Power-generation
system vulnerability and adaptation to changes in climate and water
resources. Nat. Clim. Change 6, 375–380 (2016).
36. van Vliet, M. T. H. etal. Vulnerability of US and European electricity supply
to climate change. Nat. Clim. Change 2, 676–681 (2012).
37. Liu, L., Hejazi, M., Li, H., Forman, B. & Zhang, X. Vulnerability of US
thermoelectric power generation to climate change when incorporating
state-level environmental regulations. Nat. Energy 2, 17109 (2017).
38. Miara, A. etal. Climate and water resource change impacts and adaptation
potential for US power supply. Nat. Clim. Change 7, 793–798 (2017).
39. IPCC Special Report on Renewable Energy Sources and Climate Change
Mitigation (IPCC Working Group III, Cambridge Univ. Press, 2011).
40. US Energy Information Administration June 2016 Monthly Energy Review
U.S. Department of Energy (Oce of Energy Statistics, 2016).
41. Bazilian, M. etal. Re-considering the economics of photovoltaic power.
Renew. Energy 53, 329–338 (2013).
42. SunShot Vision Study, DOE/GO-102012-3037 (US Department of Energy,
2012).
43. Hartmann, H. M. etal. Understanding Emerging Impacts and Requirements
Related to Utility-scale Solar Development (US Department of Energy, 2016).
44. Roberts, B. J. Solar Production Potential across the United States (Department
of Energy, National Renewable Energy Laboratory, 2012).
45. Brown, P. & Whitney, G. U.S. Renewable Electricity Generation: Resources and
Challenges (Congressional Research Service, 2011).
46. Kaldellis, J. K., Kapsali, M. & Kavadias, K. A. Temperature and wind speed
impact on the eciency of PV installations. Experience obtained from
outdoor measurements in Greece. Renew. Energy 66, 612–624 (2014).
47. Said, S. A. M., Hassan, G., Walwil, H. M. & Al-Aqeeli, N. e eect of
environmental factors and dust accumulation on photovoltaic modules and
dust-accumulation mitigation strategies. Renew. Sustain. Energy Rev. 82,
743–760 (2018).
48. Cronin, A. etal. Measuring degradation rates of PV systems without
irradiance data. Prog. Photovolt. 22, 851–862 (2014).
49. Barron-Gaord, G. A. etal. e solar heat island eect: larger solar power
plants increase local temperatures. Sci. Rep. 6, 35070 (2016).
50. Solecki, W. D. etal. Mitigation of the heat island eect in urban New Jersey.
Environ. Hazards 6, 39–49 (2005).
51. Campbell, G. S. & Norman, J. M. An Introduction to Environmental Biophysics
2nd edn (Springer, 1998).
52. Breshears, D. D. e grassland-forest continuum: trends in ecosystem
properties for woody plant mosaics? Front. Ecol. Environ. 4, 96–104 (2006).
53. Hobbs, R. J. etal. Novel ecosystems: theoretical and management aspects of
the new ecological world order. Glob. Ecol. Biogeogr. 15, 1–7 (2006).
54. Hobbs, R. J., Higgs, E. & Harris, J. A. Novel ecosystems: implications for
conservation and restoration. Trends Ecol. Evol. 24, 599–605 (2009).
55. Standish, R., Hobbs, R. J. & Miller, J. Improving city life: options for
ecological restoration in urban landscapes and how these inuence
interactions between humans and nature. Landsc. Ecol. 28, 1213–1221 (2012).
56. Nabhan, G. P. Our Coming Food Crisis. New York Times http://www.nytimes.
com/2013/07/22/opinion/our-coming-food-crisis.html (21 July 2013).
57. Nabhan, G. P. Growing Food in a Hotter, Drier Land: Lessons from Desert
Farmers on Adapting to Climate Uncertainty (Chelsea Green Publishing,
2013).
58. Ravi, S. etal. Colocation opportunities for large solar infrastructures and
agriculture in drylands. Appl. Energy 165, 383–392 (2016).
59. Amaducci, S., Yin, X. & Colauzzi, M. Agrivoltaic systems to optimise land
use for electric energy production. Appl. Energy 220, 545–561 (2018).
60. Aroca-Delgado, R., Perez-Alonso, J., Jesus Callejon-Ferre, A. & Velazquez-
Marti, B. Compatibility between crops and solar panels: an overview from
shading systems. Sustainability 10, 743 (2018).
61. Valle, B. etal. Increasing the total productivity of a land by combining mobile
photovoltaic panels and food crops. Appl. Energy 206, 1495–1507 (2017).
62. Macknick, J., Beatty, B. & Hill, G. Overview of Opportunities for Co-Location
of Solar Energy Technologies and Vegetation (National Renewable Energy
Laboratory, 2013).
63. Majumdar, D. & Pasqualetti, M. J. Dual use of agricultural land: introducing
‘agrivoltaics’ in Phoenix Metropolitan Statistical Area, USA. Landsc. Urban
Plan. 170, 150–168 (2018).
NATURE SUSTAINABILITY | www.nature.com/natsustain

Articles NaTure SuSTaiNabiliTy
64. Dinesh, H. & Pearce, J. M. e potential of agrivoltaic systems. Renew.
Sustain. Energy Rev. 54, 299–308 (2016).
65. Goetzberger, A. & Zastrow, A. On the coexistence of solar-energy conversion
and plant cultivation. Int. J. Sol. Energy 1, 55–69 (1982).
66. Marrou, H., Wery, J., Dufour, L. & Dupraz, C. Productivity and radiation
use eciency of lettuces grown in the partial shade of photovoltaic panels.
Eur. J. Agron. 44, 54–66 (2013).
67. Adeh, E. H., Selker, J. S. & Higgins, C. W. Remarkable agrivoltaic inuence
on soil moisture, micrometeorology and water-use eciency. PLoS ONE 13,
e0203256 (2018).
68. Marrou, H., Dufour, L. & Wery, J. How does a shelter of solar panels
inuence water ows in a soil-crop system? Eur. J. Agron. 50, 38–51 (2013).
69. Marrou, H., Guilioni, L., Dufour, L., Dupraz, C. & Wery, J. Microclimate
under agrivoltaic systems: is crop growth rate aected in the partial shade of
solar panels? Agric. For. Meteorol. 177, 117–132 (2013).
70. López-España, R. G. etal. Geographical dierentiation of wild pepper
(Capsicum annuum L. var. glabriusculum) populations from northwestern
Mexico. Phyton 85, 131–141 (2016).
71. Bosland, P. & Votava, E. Peppers: Vegetable and Spice Capsicums (CABI
Publishing, 2000).
72. Coutts, A. M., Tapper, N. J., Beringer, J., Loughnan, M. & Demuzere, M.
Watering our cities: the capacity for water sensitive urban design to support
urban cooling and improve human thermal comfort in the Australian
context. Prog. Phys. Geogr. 37, 2–28 (2013).
73. Luong, T. M. etal. e more extreme nature of North American monsoon
precipitation in the Southwestern United States as revealed by a historical
climatology of simulated severe weather events. J. Appl. Meteorol. Climatol.
56, 2509–2529 (2017).
74. Schmidt, M., Jochheim, H., Kersebaum, K.-C., Lischeid, G. & Nendel, C.
Gradients of microclimate, carbon and nitrogen in transition
zones of fragmented landscapes—a review. Agric. For. Meteorol. 232,
659–671 (2017).
75. Martinez, M. A., Andujar, J. M. & Enrique, J. M. Temperature measurement
in PV facilities on a per-panel scale. Sensors 14, 13308–13323 (2014).
76. Blair, N. etal. System Advisor Model (SAM) General Description (Version
2017.9.5) (National Renewable Energy Laboratory, 2018).
77. Blair, N. etal. System Advisor Model, SAM 2014.1.14: General Description
(National Renewable Energy Laboratory, 2014).
78. Barron-Gaord, G. A. etal. Impacts of hydraulic redistribution on grass-tree
competition vs facilitation in a semi-arid savanna. New Phytol. 215,
1451–1461 (2017).
79. Hamerlynck, E. P., Scott, R. L. & Barron-Gaord, G. A. Consequences of
cool-season drought-induced plant mortality to Chihuahuan Desert grassland
ecosystem and soil respiration dynamics. Ecosystems 16, 1178–1191 (2013).
80. von Caemmerer, S. & Farquhar, G. D. Some relationships between the
biochemistry of photosynthesis and the gas exchange of leaves. Planta 153,
376–387 (1981).
81. Barron-Gaord, G. A. etal. Photosynthetic temperature responses of
co-occurring desert winter annuals with contrasting resource-use eciencies
and dierent temporal patterns of resource utilization may allow for species
coexistence. J. Arid Environ. 91, 95–103 (2013).
82. Barron-Gaord, G. A., Scott, R. L., Jenerette, G. D., Hamerlynck, E. P. &
Huxman, T. E. Temperature and precipitation controls over leaf- and
ecosystem-level CO2 ux along a woody plant encroachment gradient.
Glob. Change Biol. 18, 1389–1400 (2012).
Acknowledgements
This research and data were supported by (1) the Water, Environmental, and Energy
Solutions initiative at the University of Arizona; (2) the Accelerate For Success Grants
Program at the University of Arizona; (3) NSF EAR No. 1659546, REU Site: Earth
Systems Research for Environmental Solutions at Biosphere 2; and (4) the Department
of Energy’s National Renewable Energy Lab through No. REJ-7-70227, Meeting
SunShot Cost and Deployment Targets through Innovative Site Preparation and Impact
Reductions on the Environment programme. The authors thank J. Adams and the
Biosphere 2 team for their assistance in maintenance of the Biosphere 2 Agrivoltaics
Learning Lab.
Author contributions
G.A.B.-G., R.L.M., L.F.S., I.B.-M., D.T.B. and M.T. established research sites and installed
monitoring equipment. G.A.B.-G. directed research. R.L.M., L.F.S., I.B.-M., D.T.B. and
M.T. conducted most of the site maintenance. G.A.B.-G., M.A.P.-Z., G.P.N. and J.E.M.
led efforts to secure funding for the research. All authors discussed the results and
contributed to the manuscript.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
s41893-019-0364-5.
Reprints and permissions information is available at www.nature.com/reprints.
Correspondence and requests for materials should be addressed to G.A.B.-G.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature Limited 2019
NATURE SUSTAINABILITY | www.nature.com/natsustain