Nature Geoscience | Volume 18 | August 2025 | 761–768 761 nature geoscience https://doi.org/10.1038/s41561-025-01742-z Article Limited carbon sequestration potential from global ecosystem restoration
Csaba Tölgyesi 1,2,14 , Nándor Csikós 1,3,14, Vicky M. Temperton4,
Elise Buisson 5, Fernando A. O. Silveira 6, Caroline E. R. Lehmann 7,8,
Péter Török 2,9,10, Zoltán Bátori 1,11 & Ákos Bede-Fazekas 12,13
Ecosystem restoration is increasingly recognized as a means of climate
change mitigation. Recent global-scale studies have suggested that
ecosystem restoration could offset a substantial fraction of human
carbon emissions since the Industrial Revolution. However, global carbon
sequestration potential remains uncertain due to the tree-centric view
of some models and difficulties in modelling restoration across different
ecosystem types. Here we applied a model-based prediction workflow
to estimate the carbon capture potential of restoring forest, shrubland,
grassland and wetland ecosystems until 2100. We found that the maximum
sequestration potential is 96.9 Gt of carbon, equivalent to 17.6% of the
anthropogenic emissions to date, or 3.7–12.0% if taking into account future
emissions until 2100. Our results suggest that ecosystem restoration has
limited potential for climate change mitigation even if orchestrated with a
pervasive shift towards sustainable, low-emissions economies globally. In
addition, if we plan restoration targets to match future climatic conditions
and consider state transitions of currently natural ecosystems due to climate
change, the potential for natural climate solutions related to ecosystem
restoration is close to zero. Therefore, we recommend that ecosystem
restoration is pursued primarily for restoring biodiversity, supporting
livelihoods and resilience of ecosystem services, as the climate mitigation
potential will vary depending on the state transitions that occur between
vegetation types.
In 2024, the European Parliament passed the ambitious Nature
Restoration Law to address the biodiversity crisis by initiating eco-
system restoration in 20% of the European Union’s land and maritime
areas until 2030 and in all degraded habitats of member states by 20501.
The EU initiative aligns with global movements, including the UN Decade
on Ecosystem Restoration, launched to halt biodiversity loss and secure
human well-being on a long-term basis2,3. Most biodiversity initiatives
are linked to the mitigation of anthropogenic climate change through
ecosystem-scale carbon storage. The EU Nature Restoration Law expli-
citly states that accelerating and up-scaling ecosystem restoration
will contribute to climate change mitigation.
There is wide consensus that reducing greenhouse gas emissions
is central to climate change mitigation4, but recapturing atmospheric
CO2 is also necessary to reach climate targets 5–7. T echnological solu-
tions for atmospheric CO2 removal are today unavailable at relevant
scales8. Conversely, natural climate solutions (NCSs) are considered
straightforward and rely, among others, on the ability of plants to
capture CO2 and store carbon in their tissues or later in the soil.
Ecosystem restoration includes revegetation of degraded land
to a reference state, thus qualifying as an important NCS via captur -
ing CO2. However, the potential of ecosystem restoration to offset
anthropogenic emissions remains controversial. Given the urgency
Received: 3 March 2024
Accepted: 10 June 2025
Published online: 31 July 2025
Check for updates
A full list of affiliations appears at the end of the paper. e-mail: [email protected]
Nature Geoscience | Volume 18 | August 2025 | 761–768 762
Article https://doi.org/10.1038/s41561-025-01742-z
to terrestrial locations using climatic, soil and topographic predictors
(Extended Data Table 1). We then assessed the potential restorable area
for each ecosystem. Using published carbon sequestration rates for
each ecosystem type, we calculated expected global carbon gain until
2100. Our estimates may be more realistic than previous ones because
we account for (1) all major terrestrial ecosystem types, (2) above- and
belowground carbon storage, where relevant, (3) carbon sequestration
rates (instead of total stocks, which often require > 70 years, that is the
length of the 2030–2100 planning period, to develop), (4) the carbon
sequestration rate of current ecosystem type (because the net carbon
gain is the difference between current and post-restoration rates), (5)
biogeographic differences, (6) socio-economic considerations that
exclude built-up and intensive agricultural areas from restoration and
(7) sustainable land-use practices in restoration targets (that is, certain
high-nature-value farming landscapes, such as wood pastures, can also
be predicted as targets), (8) future carbon emissions, (9) a schedule of
restoration implementation and (10) current and future (2061–2080)
climatic conditions to predict restoration targets. Using our model, we
estimate how global ecosystem restoration can potentially contribute
to climate change mitigation until 2100.
Available areas for restoration
Using current climate, we predicted a total of 42.48 million km2 of for-
est, 14.14 million km2 of shrubland, 36.07 million km2 of grassland and
3.10 million km2 of wetland on Earth’s land surface as potential natural
ecosystems (Fig. 2a–d). The majority has been greatly altered by human
actions30, but according to our model, 28.76 million km2 are available
for ecosystem restoration. Of this, 11.66 million km2 (40.5% of the total
area) is potential forest (Fig. 2e), slightly higher than the 9 million km2
predicted by Bastin et al. 9, but lower than the 15.50 million km 2 of
Strassburg et al.10. We found large potential areas for forest restoration
(including restoring the forested component of mosaic landscapes)
across the northern temperate and boreal zones and across subtropi-
cal and tropical regions. Nearly 4.91 million km2 (17.1%) was suitable for
shrubland worldwide as the target of ecosystem restoration (Fig. 2f),
particularly in eastern Australia and southern-central United States,
a figure similar to the Strassburg et al.10 prediction (4.11 million km2).
The total area for grassland restoration is 9.37 million km2 (32.6%),
~30% higher than Strassburg et al. ’s10 7.17 million km2, implying over-
estimation of forest expansion at the expense of grassland. Potential
grassland restoration in our model is concentrated in North America,
Eurasia, Asian highlands and tropical mosaics (Fig. 2g). Large potential
grassland restoration targets were predicted at the northern edge of
boreal forests, potentially indicating misplaced tree plantations or
other forms of degradation due to grassland overuse or a lagging of
climate-change-driven expansion of grassland on currently sparsely
vegetated areas.
T otal area predicted for wetland restoration including all freshwa-
ter and saltwater herbaceous wetlands (excluding permanent water
and wooded wetlands) was 2.83 million km 2 (9.8%), concentrated in
the American Midwest and Eastern Asia, where wetlands have been
extensively drained for agriculture31,32. Potential wetland restoration
was also identified in many floodplains and coastal habitats, such as the
Euphrates River and Gulf of Bengal (Fig. 2h). Strassburg et al.10 calcu-
lated potential wetland restoration at 0.57 million km2, and they were
explicit wetland restoration was probably underestimated. However,
our model’s predictive power was also comparatively low for wetlands.
Using a climate scenario projected for 2061–2080 (that is, the middle
of our planning period) may potentially provide more relevant target
ecosystems as in the future, sites may become available for restoration
if a current native ecosystem type will no longer be a potential predicted
ecosystem type. These state transitions may happen spontaneously
(for example, tree encroachment in present-day tundra or grassland
expansion due to excessive fires in dry tropical forests) but can also
be actively assisted if the transition is for some reason favourable.
to act and the opportunity to up-scale ecosystem restoration from
the mid-2020s, a clear picture of the realistic impact of ecosystem
restoration is required.
Influential studies provided promising model outputs 9,10, sug-
gesting up to two-thirds of the anthropogenic carbon burden
can be recaptured with restoration-based NCS. However, Bastin et al.9
received criticism for their tree-centric view of global ecosystems
that ignored diverse ecosystem types and overlooked negative
afforestation impacts on biodiversity and ecosystem functioning of
non-forest ecosystems11–13. Open ecosystems (for example, grasslands
and savannahs) with their unique biodiversity and ecosystem services,
also sequester considerable amounts of carbon 14. Unlike in forests,
open ecosystem carbon stores are mostly belowground, out of reach
from fire and drought, processes to which these ecosystems have
a high resilience, and that in forests substantively reduce above -
ground carbon15–17. Into the future, open ecosystems may be a more
secure land-cover type to store carbon in fire-prone regions15. Empiri-
cal studies have shown that fire-suppression-driven increases in tree
cover of historically open ecosystems has limited impact on total eco-
system carbon stocks18. Furthermore, open ecosystem afforestation
increases water scarcity19,20, alters fire regimes, reduces biodiversity21
and albedo, which can offset or outweigh climate benefits22,23.
Strassburg et al.10 addressed some of the above issues by modelling
the potential carbon gain of restoring a range of ecosystems and
found values similar to Bastin et al. 9. However, they used a method
to predict potential ecosystem types earmarked for restoration that
generates high uncertainty. Examining the current composition
of every 4.96 × 4.96 km pixel on Earth’s terrestrial surface, they identi-
fied natural-looking ecosystem patches and used the proportion of
each to extrapolate to the entire pixel, an approach appropriate for
homogeneous landscapes but yields high uncertainty in mosaic land-
scapes (for example, open-forest mosaic, topographically or hydrologi-
cally complex landscapes)24–26 where uniform anthropogenic impacts
were highly unlikely. Furthermore, partially degraded areas are largely
overlooked27, and woody vegetation, including non-native plantations,
were considered natural looking. Therefore, non-forested landscapes
converted to arable land with some planted trees were considered
potential forest restoration areas. This bias led, for example, to predict-
ing forest restoration across the Carpathian Basin in Eastern-Central
Europe, home to the largest stretch of intact steppe grassland within
the European Union. Moreover, Bastin et al.9, Strassburg et al.10 and a
recent tree-centric model by Mo et al.28 considered total carbon stocks,
although reaching those values by 2100 seems unlikely in certain eco-
system types, especially if restoration is not performed instantaneously
but with a realistic schedule over the twenty-first century.
At the low end of the predicted carbon sequestration potential,
Cook-Patton et al.5 listed ecosystem restoration as the least effective of
the main NCSs, with protecting intact ecosystems and improving land
management ranked as best alternatives. This aligns with Mo et al.28,
who also focused on increasing carbon stocks of existing forests. Nolan
et al.6 then highlighted the uncertainty of restoration-mediated CO 2
removal, given the tenfold difference between current highest and
lowest predictions. The low predicted values of some models are due
to potential difficulties in realizing large-scale ecosystem restoration
due to social, economic and governance constraints. Recently, some
of these constraints have started to be alleviated via top-down mecha-
nisms such as the EU Nature Restoration Law. Williams et al.29 provide a
partial solution for economic and governance constraints via effective
ongoing unassisted large-scale forest regeneration, requiring minimal
intervention but potentially contributing >2.15 million km 2 of forest
gain within 30 years.
Relying on previous modelling approaches and via addressing
the criticisms they received, we present results of a global ecosystem
restoration model. In our approach (Fig. 1), we applied machine learn-
ing to predict the potential cover percentages of native ecosystem types
Nature Geoscience | Volume 18 | August 2025 | 761–768
763
Article https://doi.org/10.1038/s41561-025-01742-z
Predicted 2061–2080 potential forest extent is 35.9–37.4 million km2,
depending on the climate scenario (SSP1–2.6, SSP2–4.5, SSP3–7.0 or
SSP5–8.5), with forest expansion at high latitudes but widespread loss
in tropical zones (Extended Data Figs. 1–4a). The latter is especially
prevalent in the Amazon Basin, as forecast by other studies33–35. Taking
such modelled state transitions into account includes both expansion
and loss compared to the present, with a net outcome that is negative,
indicating 2.3–3.4 million km 2 forest loss globally (Extended Data
Figs. 1–4e).
Compared to current predictions, future potential shrubland
area doubles, mostly due to potential forest loss, and reaches 11.2–
11.7 million km2 (Extended Data Figs. 1–4b) with a net available amount
for restoration and state transitions of 0.7–1.1 million km2 (Extended
Data Figs. 1–4f). Grasslands (and savannahs) expand their potential
area, predicted to increase to 26.9–28.8 million km 2, largely due to
savannah expansion, montane grassland cover growth in Tibet and
forest-steppe expansion over boreal forests (Extended Data Figs. 1–4c).
However, dependent upon the combination of climate scenario, resto-
ration opportunities and state transitions, there is a combined change
from a loss of 0.6 million km 2 to a gain of 1.1 million km 2 (Extended
Data Figs. 1–4g). Conversely, potential future wetland area, amounting
to 3.7–5.2 million km2 (Extended Data Figs. 1–4d), is not only higher
than current potential, but 2.6–3.8 million km 2 will be available for
restoration and state transitions (Extended Data Figs. 1–4h).
Carbon capture potential
Overall, restoration of available land (all potential land minus areas
with natural vegetation, built-up and intensive agricultural areas
and arid and polar regions) using current climate predictions would
lead to carbon capture of 1.92 Gt yr−1, summing to a total of 136.3 Gt
between 2030 and 2100 (Fig. 3). However, reclaiming all this land by
2030 and initiating target ecosystem restoration are extremely unlikely.
Using the momentum of the UN Decade on Ecosystem Restoration
combined with the targets of the EU’s Nature Restoration Law, feasibility
Model
training and
evaluation
Database compilation Prediction and rescaling
Filtering
Carbon gain estimation
Present cover of
target ecosystems
Training sites
Copernicus Global
Land Service
World Database on
Protected Areas
WorldClim
SoilGrids250m
GMTED2010
Predictors
Model training
database
Predictive
distribution
models
Global prediction
database
Predicted cover of
target ecosystems
Distribution of
target ecosystems
in degraded areas
Resolve
EcoRegions2017
database
Higher-productivity
biomes
Biomes
Predicted cover of
restorable biome-
specific ecosystems
C sequestration rates
of the biome-specific
target ecosystem types
Scientific literature
C sequestration
rates after predicted
restoration
Present
C sequestration
rates
Predicted
C gain
Present cover of
biome-specific
target ecosystems
A
B
C
D
E
F
G
H
J
I
L
M
K
Fig. 1 | Flowchart of the modelling process to predict the carbon gain
potential of global ecosystem restoration until 2100. Detailed description of
the steps marked with capital letters A to M can be found in Methods. Extended
Data Figs. 6–8 provide further details of the model. Boxes in green are external
sources of input data, orange boxes signify intermittent datasets and models
and the blue box is the final output. Horizontal tan shading separates the main
steps of modelling and prediction.
Nature Geoscience | Volume 18 | August 2025 | 761–768
764
Article
https://doi.org/10.1038/s41561-025-01742-z
calculations36 and the potential of natural regeneration29, achieving
20% of the potential area is theoretically possible for restoration
initiation by 2030. The remaining 80% could be implemented evenly
across the 2031–2100 period (Fig. 4). This more realistic timeframe
suggests sequestration of 85.2 Gt by 2100. Of this, 49.4 Gt (58.1%)
is allocated to forests, which is substantially less than either Bastin
et al.1 or Mo et al.28. Open ecosystems combined sum to 35.8 Gt (41.9%).
Including the three main open ecosystems in global agendas can thus
Forest
Latitude
Latitude
Latitude
Latitude
Shrubland
Grassland
Wetland
Available for restoration
Potential total cover
180°
180°
120° E
120° E
60° E
60° E
0°
0°
60° W
60° W
120° W
120° W
180°
180°
30° N
30° N
0°
0°
30° S
30° S
60° N
60° N
60° S
60° S
100%
0%
a
e
b
f
c
g
d
h
180°
180°
120° E
120° E
60° E
60° E
0°
0°
60° W
60° W
120° W
120° W
180°
180°
30° N
30° N
0°
0°
30° S
30° S
60° N
60° N
60° S
60° S
100%
0%
180°
180°
120° E
120° E
60° E
60° E
0°
0°
60° W
60° W
120° W
120° W
180°
180°
30° N
30° N
0°
0°
30° S
30° S
60° N
60° N
60° S
60° S
100%
0%
180°
180°
120° E
120° E
60° E
60° E
0°
0°
60° W
60° W
120° W
120° W
180°
180°
30° N
30° N
0°
0°
30° S
30° S
60° N
60° N
60° S
60° S
100%
0%
180°
180°
120° E
120° E
60° E
60° E
0°
0°
60° W
60° W
120° W
120° W
180°
180°
30° N
30° N
0°
0°
30° S
30° S
60° N
60° N
60° S
60° S
100%
0%
180°
180°
120° E
120° E
60° E
60° E
0°
0°
60° W
60° W
120° W
120° W
180°
180°
30° N
30° N
0°
0°
30° S
30° S
60° N
60° N
60° S
60° S
100%
0%
180°
180°
120° E
120° E
60° E
60° E
0°
0°
60° W
60° W
120° W
Longitude
Longitude
120° W
180°
180°
30° N
30° N
0°
0°
30° S
30° S
60° N
60° N
60° S
60° S
100%
0%
180°
180°
120° E
120° E
60° E
60° E
0°
0°
60° W
60° W
120° W
120° W
180°
180°
30° N
30° N
0°
0°
30° S
30° S
60° N
60° N
60° S
60° S
100%
0%
Fig. 2 | Potential distribution of modelled ecosystems and the available area for
restoration using current climates for predictions. a–d, Potential distribution
of modelled forest (a), shrubland (b), grassland (c) and wetland (d) ecosystems
using current climates for predictions. e–h, The available area for forest (e),
shrubland (f), grassland (g) and wetland (h) restoration. Colour coding indicates
the percentage of each ecosystem type (predicted and restorable) within a
1 × 1 km grid. Thus, ecosystem combinations (for example, forest steppes and
savannah-forest mosaics) are also allowed in our grid-level restoration planning,
although the proportion of each constituting ecosystem type appears on different
maps. For example, a savannah-forest mosaic landscape can contain forested,
shrubby and grassy parts within a grid cell, and their proportional values are shown
on each of the three corresponding maps. Available area is the potential area
minus (1) intact areas not requiring restoration, (2) intensive agricultural areas,
(3) built-up areas and (4) biomes with low productivity (polar and arid regions).