Biodiversity at risk under future cropland expansion and intensification

Abstract

Agriculture is the leading driver of biodiversity loss. However, its future impact on biodiversity remains unclear, especially because agricultural intensification is often neglected, and high path-dependency is assumed when forecasting agricultural development—although the past suggests that shock events leading to considerable agricultural change occur frequently. Here, we investigate the possible impacts on biodiversity of pathways of expansion and intensification. Our pathways are not built to reach equivalent production targets, and therefore they should not be directly compared; they instead highlight areas at risk of high biodiversity loss across the entire option space of possible agricultural change. Based on an extensive database of biodiversity responses to agriculture, we find 30% of species richness and 31% of species abundances potentially lost because of agricultural expansion across the Amazon and Afrotropics. Only 21% of high-risk expansion areas in the Afrotropics overlap with protected areas (compared with 43% of the Neotropics). Areas at risk of biodiversity loss from intensification are found in India, Eastern Europe and the Afromontane region (7% species richness, 13% abundance loss). Many high-risk regions are not adequately covered by conservation prioritization schemes, and have low national conservation spending and high agricultural growth. Considering rising agricultural demand, we highlight areas where timely land-use planning may proactively mitigate biodiversity loss. The authors predict biodiversity loss under potential future agricultural change. Agricultural expansion threatens species richness and abundance worldwide (up to one-third in some areas), often with little overlap between protected areas and high-risk expansion areas.

Full text

Articles
DOI: 10.1038/s41559-017-0234-3
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
1 Geography Department, Humboldt-University Berlin, Berlin 10099, Germany. 2 Biology Department, University of Victoria, Victoria, BC V8P 5C2, Canada.
3 Department of Conservation Biology, Estación Biológica de Doñana-CSIC, Sevilla 41092, Spain. 4 Civil and Environmental Engineering, Princeton University,
Princeton, NJ 08544, USA. 5 Woodrow Wilson School, Princeton University, Princeton, NJ 08544, USA. 6 Graduate School of Geography, Clark University,
Worcester, MA 01581, USA. 7 Biodiversity, Macroecology and Biogeography, Georg-August-University Göttingen, Göttingen 37077, Germany.
8 Integrative Research Institute on Transformations of Human–Environment Systems (IRI THESys), Humboldt-University Berlin, Berlin 10099, Germany.
*e-mail: laurajkehoe@gmail.com
Agriculture dominates more than 38% of the world’s terrestrial
surface1, poses the single largest threat to Red List threatened
species2 and is likely to remain the primary driver of terrestrial
biodiversity loss throughout the twenty-first century3. With global
crop demand for food, livestock feed and biofuels estimated to double
by 20504, agricultural expansion and intensification are projected to
increase rapidly. Despite this, few studies have assessed the potential
impacts of future agricultural land-use change on biodiversity, with
most research focusing on the impacts of climate change5. Of the few
studies that have assessed future agricultural impacts5, most have
focused on expansion6. Agricultural intensification, if considered,
is modelled using limited data or single indicators (typically yields
only)3,7–13. This may miss crucial land-use intensity impacts on biodiver-
sity14,15—especially as many agricultural practices can increase yields,
but different practices may have different outcomes for biodiversity16,17.
Most studies that explore future agricultural land-use change
rely on integrated assessment models (IAMs)3,9–13. The central
strength of IAMs lies in their complexity, in which development
pathways can be derived from a wide range of factors18,19. However,
this strength comes at the cost of a high uncertainty associated with
model outcomes20. Moreover, agricultural land-use change often
happens rapidly, following high-impact shock events21,22, which are
nearly impossible to predict in an IAM setting23. Finally, the resolu-
tion of IAMs is typically coarse (0.5°), making the assessment of
local biodiversity impacts challenging24. High-resolution investi-
gations of the full option space of possible agricultural change are
therefore needed to complement IAM approaches25.
We take modelled estimates from over 1 million data points26 of
biodiversity responses to diverse land uses27 in order to assess the
relative percentage change in species richness and abundance on a
1-km grid, and to estimate the potential average absolute change
in species richness numbers on a 110-km grid (as downscaling of
global species richness ranges to under 110 km × 110 km leads to an
over-estimation of species occurrences28). We examine three main
agricultural development pathways: (1) low-intensity expansion in
areas suitable for cropland, (2) intensification within existing crop-
lands and (3) a combination of intensification and expansion.
Results
The tropics, particularly the Amazon basin and Sub-Saharan Africa,
generally had the highest risk of biodiversity loss, were only par-
tially covered by currently protected areas (IUCN category I to VI)
and conservation prioritization schemes, and were characterized by
relatively low conservation spending and high agricultural growth.
Expansion effects on biodiversity. Mosaic grasslands, natural
grasslands and dense forests showed the highest overall relative bio-
diversity losses in expansion pathways (up to 25% loss of species
richness, in terms of the maximum percentage lost on a 1-km grid,
Fig.1, 95% confidence interval (CI) 18–31%, Supplementary Table4
and Supplementar y Fig.2). Areas of highest species richness loss due
to expansion (that is, the top 5% of grid cells with the highest num-
bers of species lost, with > 150 species potentially lost per 110-km
grid cell) were overwhelmingly found in the Amazon Basin and
the Northeastern Congolian forests and savannahs (Supplementary
Fig.3). In terms of abundance loss, many forest regions in the trop-
ics and the boreal zone were found to be at particularly high risk (up
to 21% loss of abundance, Supplementary Table4, 95% CI: 14–33%).
Biodiversity at risk under future cropland
expansion and intensification
Laura Kehoe 1,2*, Alfredo Romero-Muñoz 1, Ester Polaina3, Lyndon Estes 4,5,6, Holger Kreft 7 and
Tobias Kuemmerle1,8
Agriculture is the leading driver of biodiversity loss. However, its future impact on biodiversity remains unclear, especially because
agricultural intensification is often neglected, and high path-dependency is assumed when forecasting agricultural development—
although the past suggests that shock events leading to considerable agricultural change occur frequently. Here, we investigate
the possible impacts on biodiversity of pathways of expansion and intensification. Our pathways are not built to reach equivalent
production targets, and therefore they should not be directly compared; they instead highlight areas at risk of high biodiversity
loss across the entire option space of possible agricultural change. Based on an extensive database of biodiversity responses to
agriculture, we find 30% of species richness and 31% of species abundances potentially lost because of agricultural expansion
across the Amazon and Afrotropics. Only 21% of high-risk expansion areas in the Afrotropics overlap with protected areas (com-
pared with 43% of the Neotropics). Areas at risk of biodiversity lossfrom intensification are found in India, Eastern Europe and
the Afromontane region (7% species richness, 13% abundance loss). Many high-risk regions are not adequately covered by con-
servation prioritization schemes, and have low national conservation spending and high agricultural growth. Considering rising
agricultural demand, we highlight areas where timely land-use planning may proactively mitigate biodiversity loss.
NATURE ECOLOGY & EVOLUTION | www.nature.com/natecolevol

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles Nature ecology & evolutioN
Intensification effects on biodiversity. Overall, intensification
effects on biodiversity were less pronounced than the effects of
expansion. Intensification produced a maximum of 7% loss in spe-
cies richness (95% CI: 4–10%) and 13% loss in abundance (95% CI:
7–18%) spanning across cropland in much of Eastern Europe, India,
Nepal and sub-Saharan Africa (Fig.1, Supplementary Table4). In
terms of the maximum number of species potentially lost per 110-km
grid cell, up to ~20 species may be lost owing to intensification
across a large region of Eastern Europe, and up to ~30 species in
India and Nepal. More concentrated regions of species richness loss
were found in Mesoamerica, the Gran Chaco, and the Chiquitano
dry forests of Bolivia, where up to ~40 species may potentially be
lost. The top 5% of losses in species richness due to intensification
would see up to ~60 species potentially lost per grid cell in many
regions of sub-Saharan Africa, including the Eastern Guinean
forest–savannah mosaic and West Sudanian savannah (~40 species),
and the Afromontane and the African Great Lakes Region (~60 spe-
cies; Supplementary Fig.3).
Combined effects of intensification and expansion on biodiversity.
As can be expected, the effect of the combination of expansion and
intensification had the greatest negative effect on biodiversity (Fig.1).
For most of the globe, because the risk of biodiversity loss due to inten-
sification is much less than the risk due to expansion (Fig.1), our com-
bined pathway highlights many of the same areas as our pathway of
expansion, especially when looking at the top 5% and 10% of the distri-
bution of species loss. In terms of the top 5% of species lost, > 180 spe-
cies were found to be potentially lost per 110-km grid cell (a loss of up
to 30% of relative species richness, 95% CI: 21–37%). Such areas include
the majority of the Amazon Basin, a large area in the northeast of the
Democratic Republic of the Congo (DRC), smaller areas in Zambia
and southern Tanzania, along with border regions of the Central
African Republic, the Republic of the Congo, Cameroon and Gabon
(Supplementary Fig.3). Many parts of these areas also would have rela-
tive abundance losses of 31% (95% CI: 20–36%) under this pathway
(Fig.1). The 110-km grid cell in the analysis at most risk globally was
found in the Southwest Amazon moist forests in Peru, where a com-
bination of expansion and intensification would result in the potential
loss of > 300 species of terrestrial vertebrates (Supplementary Fig.3).
Comparison with protected areas and conservation prioritiza-
tion schemes. Many regions of potentially high biodiversity risk
due to our pathways of intensification and expansion are currently
outside protected area networks (Fig. 2, Table 1, Supplementary
Fig.3, Supplementary Table5). The Afrotropics stood out as having
large areas at risk of biodiversity loss related to expansion, only one-
fifth of which overlapped with protected areas (Fig.2). We found the
northeast DRC to be particularly at risk, with > 200 estimated spe-
cies lost per grid cell and up to a 31% loss of biodiversity abundance
(95% CI: 20–36%). Very few protected areas exist in this region,
with only 16% of our at-risk areas overlapping with protected areas
(Supplementary Table5), and of those, even fewer have a strict des-
ignation according to the IUCN categorization29 (Supplementary
Fig.3). Compared with the Afrotropics, the Neotropics have twice
the protected area coverage of areas at risk of expansion (21% and
43% respectively, Fig. 2). Although the Amazon has an extensive
network of protected areas, it is far from fully protected against
the impacts of future potential cropland expansion (52% overlap in
the Brazilian Amazon; Supplementary Table5). In sum, many of the
areas most at risk that we identify here were not fully protected.
In terms of conservation prioritization schemes and their over-
lap with our high-risk regions (top 10% of species richness loss
per pathway), we found that, on average, less than half of our
high-risk regions were included under some form of conservation
317
31%
0
0
Both Expansion Intensification
Number of species lost
Percentage loss
Figure 1 | Potential biodiversity loss due to three agricultural development pathways. The loss is shown in terms of the estimate of terrestrial vertebrate
species (mammals, birds, amphibians and reptiles) lost per 110-km grid cell(left column), relative per cent of species richness lost(middle column),
andrelative per cent of abundance lost(right column). Our pathways are not forecasts, they instead highlight areas at risk of high biodiversity loss across
the entire option space of possible agricultural change.
NATURE ECOLOGY & EVOLUTION | www.nature.com/natecolevol

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles
Nature ecology & evolutioN
prioritization. Conservation International (CI) hotspots overlapped
with 5% of our high-risk regions of expansion and 36% of our
high-risk regions of intensification, mostly in pockets of Southeast
Asia and Mesoamerica (Fig.2). The Global 200 ecoregions coin-
cided most notably across the majority of our high-risk regions of
expansion (75%, Fig.2) and with around one-third of our high-
risk regions of intensification (35%). Finally, the areas identified in
the Last of the Wild initiative coincided with 60% of our high-risk
regions of expansion, mostly in the Neotropics, and less than 1% of
our high-risk regions of intensification (Fig.2).
National level summaries. Sub-Saharan African and Latin American
countries dominated the top ten ranks in terms of average species
and abundance loss. Suriname had the overall highest potential
species richness loss (average > 200 species lost per 110-km grid
cell under the expansion and combined pathways; Supplementary
Table6). Rwanda was worst affected by our intensification pathway,
with an average of ~50 species potentially lost per 110-km grid cell.
The Republic of Congo and the DRC ranked highly across path-
ways of expansion. Outside the tropics, the only countries to reach
the top ten were located in Eastern Europe, where large tracts of
extensive cropland with few livestock in Moldova and the Ukraine
were found to potentially lose on average 10% in species abun-
dance under the intensification pathway (Supplementary Table 6,
Supplementary Fig.4).
The highest risk for species loss due to agricultural land-use
change occurred in less economically developed and highly biodi-
verse countries that also have lower conservation spending, gener-
ally lower overall governance and often high agricultural growth
rates (Fig.3). Countries at highest risk, in terms of concurrent high
agricultural growth30, low conservation spending31 and high poten-
tial species loss, include: Peru, Paraguay and Suriname in Latin
America; Cambodia in Asia; and the Republic of Guinea, Sierra
Leone, Liberia, Ghana, Republic of Congo, Cameroon, Tanzania,
Mozambique and Malawi in Africa (Fig.3, Supplementary Fig.4).
Conversely, countries predominantly found in North America and
Western Europe, in which most agricultural land is already used,
exhibited low agricultural growth rates and spent most on conser-
vation per square kilometre. Such countries had a comparatively
low risk of species loss due to future agricultural land-use change
for all three agricultural pathways that we investigated (Fig. 3,
Supplementary Tables6 and 7), probably because such regions have
relatively low species richness compared with the tropics.
Discussion
With a growing consensus that both expansion and intensifica-
tion are likely to be key drivers of biodiversity loss throughout the
twenty-first century, investigating which areas are most at risk has
become central to conservation research8,32. Although we do not
directly compare the impact of our pathways, as these pathways
were not designed to be equivalent in terms of reaching similar
future agricultural production targets, we instead provide a robust
and transparent mechanism for estimating which areas are most
susceptible to what forms of potential agricultural development in
terms of species richness and abundance loss. A major gap in sus-
tainability research lies in linking the local-scale effects of land use
on biodiversity to the impact of broader-scale pathways of potential
future changes in land use. We bridge this gap by estimating fine-
scale relative biodiversity loss at the 1-km scale, and then aggregat-
ing and translating these losses to an approximate estimate of the
potential number of species lost on a 110-km grid.
Although using absolute species richness loss has the advantage
of highlighting particularly biodiverse regions at risk, relative values
can operate on a finer scale and highlight areas with considerable
threats. Relative values of local biodiversity loss do not overlook the
importance of cultural landscapes: for instance, many landscapes in
Eastern Europe, although not hugely relevant for global biodiversity
(as shown by absolute measures of species richness), would never-
theless lose valuable farmland biodiversity if industrialized intensi-
fication were to occur33,34.
Our at-risk areas had little overlap with many conservation pri-
ority schemes, particularly the more reactive approaches, such as CI
hotspots35. On the one hand, this could be seen as fortunate; many
areas of high and endemic biodiversity were outside our areas of
high species loss due to cropland expansion. On the other hand,
CI hotspots were defined on the basis of a single taxonomic group
(plants) and the past loss of natural habitat, but, perhaps crucially
here, do not factor in potential future habitat loss. A substantial pro-
portion of conservation funding is directed towards CI hotspots36,
but our results suggest that they may not be particularly effective
in protecting vertebrates from potential future changes in agri-
cultural land use. Many at-risk areas were also not covered by the
more proactive conservation prioritization schemes. Globally, the
Last of the Wild scheme overlapped with 60% of our at-risk areas
for cropland expansion, with most overlap found in the Neotropics
(78%). However, very little overlap occurred in the Afrotropics
(21%; Fig.2). In particular, the Congolian forests and savannahs
were highly susceptible to expansion impacts, host large expanses of
intact habitat and were not well covered under the Last of the Wild
scheme. Updating of conservation prioritization methods, along
with better incorporation into conservation management, is needed
to avoid overlooking such highly biodiverse arable areas.
Our results identify countries at high future risk of biodiver-
sity loss from agricultural land-use change due to high agricultural
0
20
40
60
80
100
Global Afrotropics Indomalay Neotropics
Expansion at-risk areas
0
20
40
60
80
100
Global Afrotropics Indomalay Neotropics
Intensification at-risk areas
Protected areas CI hotspots Global 200 Last of the Wild
Overlap (%)Overlap (%)
Figure 2 | Many regions at risk of expansion and intensification currently
lie outside protected areasand conservation prioritization schemes.
Percentage overlap of our at-risk areas (in terms of the top 10% of
species potentially lost) with protected areas (IUCN category I to V)29,
Conservation International (CI) hotspots35, Global 200 Ecoregions72 and
Last of the Wild73.
Table 1 | Overlap of our at-risk areas with protected areasand
conservation prioritization schemes
Protected
areas CI hotspots Global
200 Last of
the Wild
Expansion 36% 5% 75% 60%
Intensification 10% 36% 35% 1%
Percentage overlap of our at-risk areas with protected areas at risk of expansion and
intensification (in terms of the top 10% of species potentially lost) with protected areas (IUCN
category I to V)29, Conservation International (CI) hotspots35, Global 200 Ecoregions72 and Last
of the Wild73.
NATURE ECOLOGY & EVOLUTION | www.nature.com/natecolevol

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles Nature ecology & evolutioN
growth rates, conservation underfunding and high potential future
biodiversity loss. Most of these countries were found in Latin America
and sub-Saharan Africa, with the exception of Cambodia, the only
country in Asia to exhibit such risk at a national level. Some of these
high-risk countries overlap with the 2013 list of the 40 most under-
funded countries31, which was based on threatened mammals’ range
fractions per country. However, other countries that we identified as
high risk, in terms of high potential species loss, high agricultural
growth and conservation underfunding, were not present in this list,
indicating that many high-risk countries may be under-recognized in
terms of their need for timely conservation action (Fig.3, shown in
orange: Peru, Suriname, Paraguay, Republic of Guinea, Sierra Leone,
Liberia, Ghana, Cameroon, Tanzania, Mozambique, Malawi and
Cambodia)31. Given that most conservation investment and plan-
ning take place at national scale31, these results show that proactive
conservation and land-use planning, plus higher conservation invest-
ments in these countries, are particularly important to avoid signifi-
cant future biodiversity loss, especially given that delayed mitigation
efforts are likely to have a lower rate of success, take longer to imple-
ment and cost more than prompt action37,38.
Our results expand on previous broad-scale studies that assess
potential conservation conflict due to future agricultural land-use
change7,8 by not only highlighting regions, but also by quantifying
spatially explicit biodiversity loss. When comparing our at-risk
areas with previous research, we found many regions of India and
the African Great Lakes to be areas of high potential conserva-
tion conflict, based on high levels of human appropriation of net
primary productivity and cropland extent alongside high levels of
vertebrate endemism richness14, high numbers of threatened verte-
brates and the proportion of non-cropland that could be degraded7,
and one of the highest priorities for both intensification and bird
conservation—particularly for the African Great Lakes region8.
This emphasizes the importance of ramping up conservation
efforts in these regions to curb possible future biodiversity loss.
Latin America and sub-Saharan Africa still contain relatively intact
and highly diverse natural areas that are suitable for cropland.
Thus, the potential for agricultural activity and subsequent biodi-
versity loss is high. These areas, particularly sub-Saharan Africa,
are at the crossroads of economic, demographic and agricultural
growth, making the minimization of the potential impacts of agri-
cultural change an urgent task39,40.
Our results show that compared with other studies7, Southeast
Asia has relatively low potential biodiversity loss. This may be
because our crop suitability data do not account for some tree crops
such as rubber, which can grow in areas normally not suitable for
cereal cropland, pose significant threats to biodiversity and are
prominent in Southeast Asia41, suggesting that we underestimated
biodiversity loss in such situations. Furthermore, primarily because
of issues with data availability, our analysis could not include areas
suitable for livestock, calculate time-lagged responses of biodiver-
sity loss, or assess the potential geographical variation in estimates
of biodiversity loss.
On the other hand, our results may overestimate biodiversity
loss for two reasons: first, scaling local biodiversity loss estimates
to larger scales may overestimate loss42, and second, biodiversity
change estimates of non-industrialized agroecological intensifica-
tion techniques (not available for this study) may not be detrimental
to biodiversity16,17 and thus may hold great promise for increasing
yields with little to no net loss to biodiversity (see Methods for
expanded limitations section). Finally, our prioritization is based on
maximizing for species richness and abundance. This cannot and
should not be the only way to prioritize for nature conservation,
especially when considering the importance of intrinsic and cultural
values along with ecosystem resilience and human well-being43.
In sum, our results can inform land-use planning policy and future
prioritization analysis by highlighting the most at-risk areas from a
1-km grid to the national level, where the threat to species richness
and abundance related to potential future agricultural change was
Conservation
spending
Species
loss
Agricultural
growth
Figure 3 | Conservation spending, agricultural growth and potential species loss. High conservation spending (> US$50 km–2)31 is shown in blue, high
agricultural growth (> 3% average growth between 2009–2013)30 in pink, and high potential species loss under the pathway of both expansion and
intensification (national average of > 50 species lost per grid cell) in yellow. Combinations of factors are shown in: orange (low spending, high growth and
high species loss; that is, highest risk for all three factors); purple (high spending, high growth and low species loss); and green (high spending, low growth,
and high species loss). Dark brown indicates high values for all three factors: that is, concordance of high spending, growth and species loss. Countries
with low values for all three factors are shown in grey. Countries and territories with missing values are shown in white (Puerto Rico, French Guiana,
Greenland, Zimbabwe and Somalia).
NATURE ECOLOGY & EVOLUTION | www.nature.com/natecolevol

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles
Nature ecology & evolutioN
previously unknown. We provide a comprehensive overview of areas
most at risk under the main modes of agricultural change and, in
doing so, fill gaps related to coarse spatial resolution and a neglect
of sudden, severe and unpredictable events in scenario-based
assessments. In a world of rapidly increasing demand for natural
resources, and in which marked changes in coupled human–natural
systems are both becoming more frequent21 and global in reach44,
our proactive approach indicates where timely conservation action
could avert future biodiversity loss. Although our results should not
be the only resource used to guide conservation action, they support
previous calls that potential future threats due to land-use change
should be incorporated in conservation prioritization schemes13,45–47.
Methods
Land-use data. We used a land-systems-based approach and developed possible
agricultural development pathways in terms of shis between land systems (for
example from low-intensity to high-intensity cropland). Land systems represent the
interface between the majority of human activities and the natural environment,
and consist of hierarchical categorical classications that combine metrics of land
cover, land-use intensity and livestock density48.
We first developed an updated global land-systems map to make use of the
most recent land-cover and land-use datasets available, and to work at a finer
spatial resolution than previously possible (from an original resolution of ~9.25 km2
to our 1-km grid; Supplementary Fig.1). We achieved this by following a decision-
tree classification48 to map land systems globally while using updated datasets.
We compiled six datasets related to cropland extent49, tree cover50, urban and bare
areas51, livestock density52 and yield gaps53 (Supplementary Table1). To investigate
biodiversity loss due to potential agricultural expansion, we included a biophysical
crop suitability condition53 to the natural classes. Cropland suitability was based on
agro-climatic and agro-edaphic conditions in five main steps of data processing:
(i) agro-climatic indicators, (ii) crop-specific agro-climatic assessment and yield
calculations, (iii) agro-climatic constraints, (iv) edaphic assessment including soil
and terrain limitations, and (v) integration of steps (i) to (iv) to form a unified
spatial database53. We included areas that are ‘very high’ to ‘marginally’ suitable
for high-input-level cropland (in terms of optimum applications of nutrients,
and chemical control of pests, diseases and weeds) along with low-input areas
(characterized by labour-intensive techniques, no application of nutrients, no use of
chemicals for pest and disease control) in order to avoid being overly conservative
in terms of where cropland is possible. In saying this, the difference in spatial
extent between high-input and low-input areas suitable for cropland was minor53.
However, the underlying land-systems map is built on land-cover maps that can
have significantly different spatial extents, and some lack formal validation54,55,
which can translate into errors in the assessment of potential biodiversity impacts.
No global datasets are currently available that indicate areas suitable for tree
plantations or livestock; therefore, neither could be explicitly included in the
analysis. Moreover, data protection and confidentiality legislation make it difficult
to map industrialized livestock units52 and assess their impact on biodiversity.
Land inside protected areas was not excluded from the analysis for two
main reasons: first, to explore whether areas with high risk of biodiversity loss
due to potential future agricultural development are currently protected against
such development; second, although current protected areas play a central role
in conservation, such areas are human-controlled land-cover classes, are thus
susceptible to change, and may not continue to be protected in the future56.
Therefore, currently protected areas that are found in our areas of high risk may
particularly merit ongoing support.
Agricultural development pathways. We developed three explorative pathways
of potential future agricultural land-use change: (1) expansion into all natural
areas suitable for cropland; (2) intensification on all existing cropland; and (3) a
combination of both intensification and expansion. These pathways explore the
full option space of potential agricultural development in terms of expansion and
intensification in order to highlight areas most at risk of biodiversity loss. These
pathways are therefore not designed as predictive scenarios, do not have a specific
timeline or endpoint, and are not based on a specific socio-economic scenario. In
reality, a mix of all three pathways is likely to occur.
Our expansion pathway does not intensify current cropland but extends low-
intensity cropland systems into all natural areas deemed suitable for cropland53.
This pathway represents continued loss of natural ecosystems due to the lack
of yield gains on existing land systems stemming from inadequate resources or
capital, or available inexpensive land for expansion, as is currently apparent in
many tropical deforestation frontiers of South America, sub-Saharan Africa and
Southeast Asia57,58.
On the other hand, our intensification pathway consisted of all land systems
currently under any form of agricultural activity transforming to the highest level
of cropland intensity for that class without any expansion into natural areas. For
example, an area classified as extensive (low-intensity) cropland would become
intensive cropland (Table2). This pathway represents a global push to close yield
gaps in less-developed regions, as is currently happening across much of Europe,
the United States and parts of South America58. Many crop yields are heavily
dependent on fertilizer use and irrigation, with substantial production increases
(45% to 70% for most crops) possible if yield gaps were closed58. Northern and
western European countries are already close to their maximum attainable yield,
with North America, Southeast Asia and Oceania achieving more than 60% of their
potential production. However, Africa and Eastern Europe are currently producing
only 40% of their potential59. In our intensification pathway, closing such yield
gaps would be achieved by moving from low-intensity farms with little to no
fertilizer, pesticide, irrigation or mechanization to highly intensified, conventional
monocultures that are characterized by high inputs and large fields, in line with the
classification system of ref. 26.
Finally, the third pathway is a combination of both expansion and
intensification, in which cropland expands into natural areas wherever possible and
intensification takes place in all agricultural areas (both long-standing and newly
converted). This represents the pathway of most severe change, in which both
processes of intensification and expansion increase unabated in response to rising
demand. Current land-use change in parts of Argentina, Paraguay or Brazil reflects
this trend60,61. This pathway also accounts for Jevon’s paradox, in which yield
improvements spur further expansion owing to better opportunities for farming,
and is a likely pathway for some regions of sub-Saharan Africa62.
Our pathways did not include changes in livestock density on pastures.
For example, if a system was extensive cropland with few livestock, in the
intensification pathway it would become intensive cropland with few livestock
(Supplementary Table2). This decision was based on the assumption that
intensified livestock management will join the ‘livestock revolution’—a shift away
from pasture-based management towards industrialized feedlots that depend on
crop-based feeds rather than local land resources. This process is already under
way in many rapidly growing economies of Asia and South America63,64. Including
climate change scenarios in our consideration of the suitability of potential future
cropland was outside the scope of the current study, but could prove useful for
assessing the synergistic impacts of climate and land-use change.
Biodiversity data. We calculated total species richness by overlaying extent-of-
occurrence maps for birds65, mammals, amphibians and reptiles (where available)66
with an equal-area grid (approximately 110 × 110 km or 1 degree at the Equator).
Global-scale biodiversity data based on extent of occurrence should not be
downscaled to less than 1 degree, as finer resolutions lead to an over-estimation of
species occurrences28. Therefore, although our land-system maps allow us to assess
relative percentage change on a 1-km grid, we use a 110-km grid to provide an
estimate of the average absolute plot-level loss of biodiversity.
Estimating the impact of agricultural development pathways on biodiversity.
We assessed species responses to various forms of land use and land-use intensity
change. The data that allow for this originate from the PREDICTS project26, an
initiative that collates local-scale studies from around the globe with the goal of
quantifying species-level and community-level responses to a range of human
activities including agriculture, hunting, deforestation, introduction of invasive
species and human population expansion26. Using this data, earlier work27 predicted
changes in species richness, rarefied species richness, and abundance percentage
change from a natural baseline to various levels of land-use intensity. By using an
approach that substitutes space for time, generalized linear mixed effects models
analysed 320,924 records of species richness and 1,130,251 records of abundance
at 11,525 sites27. Random intercept terms were included to account for study-level
differences, with final inclusion based on the models’ Akaike Information Criterion
values. A backward stepwise model simplification was then used to select the final
fixed-effects structure27. The results allow for estimates of biodiversity loss per
land-use intensity class relative to a natural unimpacted baseline.
To investigate the spatial patterns of biodiversity loss for each agricultural
development pathway, we first matched our land-systems classes to levels of high,
Table 2 | Three pathways of land-use change
Original land system Expansion Intensification Both
Forest and grassland Converted
to extensive
cropland
No change Converted
to intensive
cropland
Extensive to
medium-intensity
cropland
No change Intensified Intensified
Intensive cropland No change No change No change
The table shows (i) expansion of low-intensity (extensive) cropland into uncultivated areas, (ii)
intensification of existing cropland, and (iii) a combination of both, in which existing cropland and
newly converted regions are intensified.
NATURE ECOLOGY & EVOLUTION | www.nature.com/natecolevol

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles Nature ecology & evolutioN
medium, and low intensity for each main land-use type (for detailed conversion
table, see ref. 27). This allowed us to calculate average biodiversity loss per land
system (relative to an unimpacted baseline) by taking the mean model estimates
of biodiversity loss per land-use intensity class from previous work27. The result
gives average relative biodiversity gain or loss per land system. We then calculated
the biodiversity percentage change per land-system conversion in each of our
three pathways described above (Supplementary Tables3 and 4). To estimate the
biodiversity loss associated with land-system conversions, we took the difference
between the loss associated with the original land system and the loss associated
with the new land system that our pathway converted to. We then divided this by
the loss estimate for the original land system. This gave the relative biodiversity
change (on a 1-km grid). Therefore, whereas the original PREDICTS estimates
are from an unimpacted baseline, our estimates represent the biodiversity loss
associated with shifting from one land system to another. For example, in the
intensification pathway, estimates of biodiversity loss are the difference between
(1) the loss that has already taken place in converting from natural land to low or
medium intensity and (2) the biodiversity loss associated with high-intensity land
use. We calculated mean 95% confidence intervals per land-system conversion
from the original confidence intervals by ref. 27 in which a ten-fold cross-validation
was used by excluding data from ~10% of all studies at a time.
To calculate biodiversity loss in terms of the potential number of species
lost per 110-km grid cell, we first calculated the area-weighted mean value of
biodiversity percentage change across all converted land systems per 110-km grid
cell. We then multiplied this by the number of species present in each grid cell as
derived from the extent of occurrence maps66. Spatially explicit biodiversity data
on abundance is not available on a global scale, so these values were not converted
from relative (% per land system) to absolute values (number of individuals or
biomass lost). We also calculated the top 5% and 10% of the distribution of species
richness loss in order to highlight the highest risk regions for each pathway.
Six sources of uncertainty related to this approach merit attention. First, the
sites that were used to model biodiversity loss varied in maximum linear extent
with a median extent of 106 m (interquartile range 50–354 m)27. Our estimates
of biodiversity loss are scaled to much larger areas and therefore should not be
taken as direct predictions of species loss, but approximate loss estimates from
the linear mixed-effects models27 which are based on the most comprehensive
dataset currently available. The upscaling from a local scale to a 1-km grid likely
overestimates species loss42. Our estimates of species richness loss on a 110-km grid
should therefore be seen as a metric to highlight biodiversity-rich areas that are
most susceptible to absolute biodiversity loss, but absolute values should be treated
with caution owing to the uncertainties related to upscaling. Second, although
we minimize the over-estimation of species occurrences by using a 1-degree
grid, it remains possible that over-estimation of species occurrence has occurred,
especially for taxa for which extant distributions are weakly understood, such as
for many amphibians and reptiles28. Third, our analyses ignores lagged responses
and thus possible extinction debt as historical data are rare27, and therefore may
underestimate biodiversity loss67. For example, the Living Planet Index68 uses
time-series data for vertebrate abundance and finds a higher rate of loss than
the PREDICTS estimates that we use here27. Fourth, although our method for
estimating biodiversity impacts is based on the largest available dataset of land-
use-related biodiversity change, some countries and ecosystems remain under-
represented26. Fifth, our intensification pathways may overestimate biodiversity
loss, as they assume that conventional intensification will take place. Intensification
is complex and multidimensional14,69, and there are ways in which yield increases
are possible with lower or no net loss of species richness and abundance16,17. To
estimate biodiversity change associated with ‘sustainable’ intensification, it is
necessary to develop a better definition of this concept along with more empirical
data, particularly when considering fragile ecosystems70. Sixth and finally, using
species richness and abundance as metrics of biodiversity is problematic as it can
be over-representative of common, widespread species and can overshadow rare or
small-ranged species, which are often most in need of conservation71. With better
data, for example, on the effect of land-use change on threatened species, we could
highlight areas that are particularly sensitive to change in a more nuanced way.
Analysing spatial patterns of biodiversity loss due to agricultural land-use
change. We calculated the proportion of overlap between our at-risk areas of high
biodiversity loss (top 10% of the distribution of species richness loss per pathway)
with IUCN category protected areas29 and three global conservation priority
schemes: (1) the CI biodiversity hotspots, as a reactive approach that targets areas
that have already lost 70% of native habitat35, (2) the Global 200 ecoregions, as a
mixture of both proactive and reactive approaches that identifies ecoregions of
exceptional biodiversity in terms of irreplaceability and distinctiveness72, and (3)
the Last of the Wild, a reactive approach that shows the ten largest contiguous
wilderness areas by terrestrial biome and realm73. Together, this allowed us to
ascertain the percentage of overlap (on a global to national level) between our
high-risk areas and global proactive and reactive conservation priority schemes
or, in the case of protected areas, the regions that are secure against potentially
rising land-use pressure. Alternatives such as Key Biodiversity Areas (KBA)74
were excluded as their pattern is similar to protected areas at broad spatial scales.
However, future, finer-scale studies may benefit from KBAs’ more targeted,
conservation investment-based approach. Likewise, taxon-specific schemes such
as Important Bird Areas75 may address the conservation prioritization of a single-
taxon approach.
Highlighting at-risk areas is a crucial first step in effectively pinpointing
regions most in need of conservation attention. Although identifying such areas
on a global scale can be useful for cross-boundary conservation nongovernmental
organizations (NGOs) and institutions, most conservation funding originates from
domestic spending (US$14.5 billion out of US$16 billion, with approximately
US$1 billion from international NGOs)31. Therefore, conveying results on a national
level allows us to better understand policy-relevant conservation opportunities.
We summarized our results at the national level by calculating the average
country-level species richness lost per 110-km grid cell in each agricultural
development pathway. To emphasize countries that are at higher risk, not only of
potential species loss, but also in terms of national support for conservation and
agricultural trends, we compared our results against (1) conservation spending,
averaged per square kilometre and corrected by each country’s proportional dollar
costs of a fixed basket of goods and services31, and (2) the average percentage
agricultural economic growth from 2009 to 201330. National agricultural growth
rates give an indication of where conservation planning may be more urgent in the
face of rapid agricultural change.
Data availability. The datasets generated during the current study are available
from the HU-Box (https://box.hu-berlin.de/d/053f45f377/).
Received: 25 November 2016; Accepted: 8 June 2017;
Published: xx xx xxxx
References
1. Ramankutty, N., Evan, A. T., Monfreda, C. & Foley, J. A. Farming the planet:
1. Geographic distribution of global agricultural lands in the year 2000.
Global Biogeochem. Cycles 22, GB1003 (2008).
2. Maxwell, S., Fuller, R., Brooks, T. & Watson, J. e ravages of guns, nets and
bulldozers. Nature 536, 143 (2016).
3. Sala, O. E. et al. Global biodiversity scenarios for the year 2100. Science 287,
1770–1774 (2000).
4. Tilman, D., Balzer, C., Hill, J. & Befort, B. L. Global food demand and the
sustainable intensication of agriculture. Proc. Natl Acad. Sci. USA 108,
20260–20264 (2011).
5. Titeux, N. et al. Biodiversity scenarios neglect future land-use changes.
Glob. Change Biol. 22, 2505–2515 (2016).
6. Pereira, H. M. et al. Scenarios for global biodiversity in the 21st century.
Science 330, 1496–1501 (2010).
7. Shackelford, G. E., Steward, P. R., German, R. N., Sait, S. M. & Benton, T. G.
Conservation planning in agricultural landscapes: hotspots of conict
between agriculture and nature. Diversity Distrib. 21, 357–367 (2015).
8. Phalan, B., Green, R. & Balmford, A. Closing yield gaps: perils and
possibilities for biodiversity conservation. Phil. Trans. R. Soc. B Biol. Sci. 369,
20120285 (2014).
9. Rondinini, C. & Visconti, P. Scenarios of large mammal loss in Europe for the
21st century. Conserv. Biol. 29, 1028–1036 (2015).
10. Visconti, P. et al. Projecting global biodiversity indicators under future
development scenarios. Conserv. Lett. 9, 5–13 (2015).
11. B ellard, C. et al. Vulnerability of biodiversity hotspots to global change.
Glob. Ecol. Biogeogr. 23, 1376–1386 (2014).
12. de Baan, L., Alkemade, R. & Koellner, T. Land use impacts on biodiversity in
LCA: a global approach. Int. J. Life Cycle Assessment 18, 1216–1230 (2013).
13. Visconti, P. et al. Future hotspots of terrestrial mammal loss. Phil. Trans. R.
Soc. B Biol. Sci. 366, 2693–2702 (2011).
14. Kehoe, L. et al. Global patterns of agricultural land-use intensity and
vertebrate diversity. Diversity Distrib. 21, 1308–1318 (2015).
15. Foley, J. A. et al. Solutions for a cultivated planet. Nature 478,
337–342 (2011).
16. Clough, Y. et al. Combining high biodiversity with high yields in tropical
agroforests. Proc. Natl Acad. Sci. USA 108, 8311–8316 (2011).
17. Tscharntke, T. et al. Global food security, biodiversity conservation and
the future of agricultural intensication. Biol. Conserv. 151,
53–59 (2012).
18. Alkemade, R. et al. GLOBIO3: a framework to investigate options for
reducing global terrestrial biodiversity loss. Ecosystems 12, 374–390 (2010).
19. Müller, C. & Lotze-Campen, H. Integrating the complexity of global change
pressures on land and water. Glob. Food Secur. 1, 88–93 (2012).
20. Rosen, R. A. IAMs and peer review. Nat. Clim. Change 5, 390–390 (2015).
21. Ramankutty, N. & Coomes, O. T. Land-use regime shis: an analytical
framework and agenda for future land-use research. Ecol. Soc. 21, 1 (2016).
22. Jepsen, M. R. et al. Transitions in European land-management regimes
between 1800 and 2010. Land Use Policy 49, 53–64 (2015).
23. Müller, D. et al. Regime shis limit the predictability of land-system change.
Glob. Environ. Change 28, 75–83 (2014).
NATURE ECOLOGY & EVOLUTION | www.nature.com/natecolevol

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved. © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles
Nature ecology & evolutioN
24. Nadeau, C. P., Urban, M. C. & Bridle, J. R. Coarse climate change projections
for species living in a ne-scaled world. Glob. Change Biol. 23, 12–24 (2016).
25. Erb, K.-H. et al. Exploring the biophysical option space for feeding the world
without deforestation. Nat. Commun. 7, 11382 (2016).
26. Hudson, L. N. et al. e PREDICTS database: a global database of how
local terrestrial biodiversity responds to human impacts. Ecol. Evol. 4,
4701–4735 (2014).
27. Newbold, T. et al. Global eects of land use on local terrestrial biodiversity.
Nature 520, 45–50 (2015).
28. Hurlbert, A. H. & Jetz, W. Species richness, hotspots, and the scale
dependence of range maps in ecology and conservation. Proc. Natl Acad. Sci.
USA 104, 13384–13389 (2007).
29. e World Database on Protected Areas (WDPA) (UNEP-WCMC,
Cambridge, 2016).
30. World Bank GroupWorld Development Indicators (World Bank
Publications, 2013).
31. Waldron, A. et al. Targeting global conservation funding to limit immediate
biodiversity declines. Proc. Natl Acad. Sci. USA 110, 12144–12148 (2013).
32. Green, R. E., Cornell, S. J., Scharlemann, J. P. W. & Balmford, A. Farming and
the fate of wild nature. Science 307, 550–555 (2005).
33. Donald, P. F., Pisano, G., Rayment, M. D. & Pain, D. J. e Common
Agricultural Policy, EU enlargement and the conservation of Europe’s
farmland birds. Agric. Ecosyst. Environ. 89, 167–182 (2002).
34. Hartel, T., Schweiger, O., Ollerer, K., Cogalniceanu, D. & Arntzen, J. W.
Amphibian distribution in a traditionally managed rural landscape of Eastern
Europe: probing the eect of landscape composition. Biol. Conserv. 143,
1118–1124 (2010).
35. Mittermeier, R. A. et al. Hotspots Revisited: Earth's Biologically Richest and
Most Endangered Terrestrial Ecoregions(CEMEX, Mexico City, 2004).
36. Brooks, T. M. et al. Global biodiversity conservation priorities. Science 313,
58–61 (2006).
37. Cook, C. N., Wintle, B. C., Aldrich, S. C. & Wintle, B. A. Using strategic
foresight to assess conservation opportunity. Conserv. Biol. 28, 1474–1483 (2014).
38. Oliver, T. H. & Roy, D. B. e pitfalls of ecological forecasting. Biol. J.
Linnean Soc. 115, 767–778 (2015).
39. Searchinger, T. D. et al. High carbon and biodiversity costs from converting
Africa’s wet savannahs to cropland. Nat. Clim. Change 5, 481–486 (2015).
40. Estes, L. D. et al. Reconciling agriculture, carbon and biodiversity in a
savannah transformation frontier. Phil. Trans. R. Soc. Lond. B Biol. Sci. 371,
20150316 (2016).
41. Wilcove, D. S., Giam, X., Edwards, D. P., Fisher, B. & Koh, L. P. Navjot’s
nightmare revisited: logging, agriculture, and biodiversity in Southeast Asia.
Trends Ecol. Evol. 28, 531–540 (2013).
42. Laurance, W. F. et al. e fate of Amazonian forest fragments: a 32-year
investigation. Biol. Conserv. 144, 56–67 (2011).
43. Fischer, J. et al. Land sparing versus land sharing: moving forward. Conserv.
Lett. 7, 1755–2630 (2014).
44. Scheer, M. et al. Anticipating critical transitions. Science 338, 344–348 (2012).
45. Lee, T. M. & Jetz, W. Future battlegrounds for conservation under global
change. Proc. R. Soc. B Biol. Sci. 275, 1261–1270 (2008).
46. Tingley, M. W., Estes, L. D. & Wilcove, D. S. Ecosystems: Climate change
must not blow conservation o course. Nature 500, 271–272 (2013).
47. Estes, L. D. et al. Using changes in agricultural utility to quantify future
climate-induced risk to conservation. Conserv. Biol. 28, 427–437 (2014).
48. van Asselen, S. & Verburg, P. H. A land system representation for global
assessments and land-use modeling. Glob. Change Biol. 18, 3125–3148 (2012).
49. Fritz, S. et al. Mapping global cropland and eld size. Glob. Change Biol. 21,
1980–1992 (2015).
50. Hansen, M. C. et al. High-resolution global maps of 21st-century forest cover
change. Science 342, 850–853 (2013).
51. ESA-CCI. CCI Land Cover Project http://www.esa-landcover-cci.org/?q=
node/158 (ESA, 2016).
52. Wint, W. & Robinson, T. Gridded Livestock of the World(FAO: Rome, 2007).
53. Global Agro-ecological Zones (GAEZ v3.0) http://www.fao.org/nr/gaez/en/
(IIASA/FAO, 2012).
54 Verburg, P., Neumann, K. & Nol, L. Challenges in using land use and land
cover data for global change studies. Glob. Change Biol. 17, 974–989 (2011).
55 Fritz, S. et al. Highlighting continued uncertainty in global land cover maps
for the user community. Environ. Res. Lett. 6, 044005 (2011).
56 Mascia, M. B. & Pailler, S. Protected area downgrading, downsizing, and
degazettement (PADDD) and its conservation implications. Conserv. Lett. 4,
9–20 (2011).
57 Gibbs, H. K. et al. Tropical forests were the primary sources of new
agricultural land in the 1980s and 1990s. Proc. Natl Acad. Sci. USA 107,
16732–16737 (2010).
58 Mueller, N. D. et al. Closing yield gaps through nutrient and water
management. Nature 490, 254–257 (2012).
59 Pradhan, P., Fischer, G., van Velthuizen, H., Reusser, D. E. & Kropp, J. P.
Closing yield gaps: how sustainable can we be? PLoS ONE 10,
e0129487 (2015).
60 Nepstad, D. C. & Stickler, C. M. Managing the tropical agriculture revolution.
J. Sustainable For. 27, 43–56 (2008).
61 Angelsen, A. & Kaimowitz, D. in Agricultural Technologies and Tropical
Deforestation (eds. Angelsen, A. & Kaimowitz, D.) 1–18 (CABI,
Wallingford, 2001).
62 Hertel, T. W., Ramankutty, N. & Baldos, U. L. C. Global market integration
increases likelihood that a future African Green Revolution could increase
crop land use and CO2 emissions. Proc. Natl Acad. Sci. USA 111,
13799–13804 (2014).
63 Delgado, C., Rosegrant, M., Steinfeld, H., Ehui, S. & Courbois, C. Livestock
to 2020: the next food revolution. Outlook Agric. 30, 27–29 (2001).
64 Naylor, R. et al. Losing the links between livestock and land. Science 310,
1621–1622 (2005).
65 Bird Species Distribution Maps of the World (BirdLife International,
Cambridge, and NatureServe, Arlington, 2015).
66 IUCN Red List of reatened Species Version 2016-1 http://www.iucnredlist.org
(IUCN; accessed 2 February 2016).
67 Kuussaari, M. et al. Extinction debt: a challenge for biodiversity conservation.
Trends Ecol. Evol. 24, 564–571 (2009).
68 Living Planet Report 2012 (World Wildlife Fund, Gland, 2012).
69 Kuemmerle, T. et al. Challenges and opportunities in mapping land use
intensity globally. Curr. Opin. Environ. Sustainability 5, 484–493 (2013).
70 Loos, J. et al. Putting meaning back into ‘sustainable intensication’. Front.
Ecol. Environ. 12, 356–361 (2014).
71 Grenyer, R. et al. Global distribution and conservation of rare and threatened
vertebrates. Nature 444, 93–96 (2006).
72 Olson, D. M. & Dinerstein, E. e Global 200: priority ecoregions for global
conservation. Ann. Missouri Bot. Gard. 89, 199–224 (2002).
73 Sanderson, E. W. et al. e human footprint and the last of the wild.
Bioscience 52, 891–904 (2002).
74 A Global Standard for the Identication of Key Biodiversity Areas Version 1.0,
1st edn (IUCN, Gland, 2016).
75 Spotlight on Important Bird and Biodiversity Areas http://www.birdlife.org/
datazone (BirdLife International, 2014).
Acknowledgements
We thank T. Newbold for comments and insights on an earlier version of the manuscript,
and P. Verburg, D. Eitelberg and D. Müller for constructive discussions. We thank
F. Pötzschner and B. Jakimow for technical support. L.K. and T.K. acknowledge funding
by the Einstein Foundation Berlin (Germany).
Author contributions
L.K. and T.K. conceived the study. L.K. collected and analysed the data, and prepared
the manuscript. A.R.M. and E.P. assisted in analysing the data. All authors discussed the
results and commented on the manuscript.
Competing interests
The authors declare no competing financial interests.
Additional information
Supplementary information is available for this paper at doi:10.1038/s41559-017-0234-3.
Reprints and permissions information is available at www.nature.com/reprints.
Correspondence and requests for materials should be addressed to L.K.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
NATURE ECOLOGY & EVOLUTION | www.nature.com/natecolevol