LETTER • OPEN ACCESS You may also like

- Turnover rates of regenerated forests

Mapping global forest regeneration–an untapped challenge restoration efforts in the
Brazilian Atlantic forest
potential to mitigate climate change and Pedro Ribeiro Piffer, Marcos Reis Rosa,
Leandro Reverberi Tambosi et al.

biodiversity loss - Influence of logging on the effects of

wildfire in Siberia
E A Kukavskaya, L V Buryak, G A Ivanova
To cite this article: Pui-Yu Ling et al 2023 Environ. Res. Lett. 18 054025 et al.

- Expanding wildland-urban interface alters

forest structure and landscape context in
the northern United States
Nancy F Sonti, Rachel Riemann, Miranda
View the article online for updates and enhancements. H Mockrin et al.

This content was downloaded from IP address 71.47.239.39 on 26/09/2023 at 18:30

 Environ. Res. Lett. 18 (2023) 054025 https://doi.org/10.1088/1748-9326/accfaf

LETTER

Mapping global forest regeneration–an untapped potential to

OPEN ACCESS
mitigate climate change and biodiversity loss
RECEIVED
28 June 2022 Pui-Yu Ling1,∗, Naikoa Aguilar-Amuchastegui1, William Baldwin-Cantello2, Tim Rayden3,
REVISED James Gordon2, Stuart Dainton2, April L Bagwill3 and Pablo Pacheco1
15 March 2023
1
ACCEPTED FOR PUBLICATION
World Wildlife Fund, United States, 1250 24th Street NW, Washington, DC 20037, United States of America
2
24 April 2023 World Wide Fund for Nature, United Kingdom, The Living Planet Centre, Rufford House, Brewery Road, Woking, Surrey GU21 4LL,
United Kingdom
PUBLISHED 3
5 May 2023 Wildlife Conservation Society, 2300 Southern Boulevard, Bronx, NY 10460, United States of America
∗
Author to whom any correspondence should be addressed.

Original Content from E-mail: [email protected]

this work may be used
under the terms of the Keywords: forest regeneration, remote sensing, global forest restoration, nature-based solutions, Bonn Challenge
Creative Commons
Attribution 4.0 licence. Supplementary material for this article is available online
Any further distribution
of this work must
maintain attribution to
the author(s) and the title Abstract
of the work, journal Forest regeneration can be a low-cost solution to mitigate climate change, and mapping its extent
citation and DOI.
can support global goals such as the Bonn Challenge, which set a goal to put 350 million hectares
of degraded forests and landscapes into restoration by 2030. Our study combined multiple remote
sensing datasets and expert surveys, identifying 55.7 ± 6.2 million hectares of likely regenerated
forests between 2000 and 2015 across areas that were not forested before 2000 and have remained
forested from 2015 to 2018. The identified forest regeneration could potentially represent
22–25 billion young trees and a total biomass of about 3.2 billion tonnes. Forest regeneration took
place in sites with less opportunity cost for agriculture for every country, but in more developed
regions, forest regeneration took place in sites with higher suitability for cultivation. Expert
feedback associated agricultural land use transitions and the establishment of protected areas,
coupled with effective management and local support, as the key factors leading to successful forest
regeneration. The results, publicly available, can facilitate discussions and help identify strategic
locations to foster forest regeneration to achieve the global goals of mitigating climate change and
restoring biodiversity.

1. Introduction not represent the full picture of forest regeneration:

The dataset only flags a pixel when its forest cover is
Mapping and estimating forest regeneration at the over 50% regardless of the previous land cover [9],
global level can help quantify progress towards the or whether the forest remains [10], or whether the
global goals of sequestering carbon for climate change gain came from commercial forest plantation [11].
mitigation and improving habitat for biodiversity, as In addition, the spatial resolution at 30 m likely only
forest regeneration is one of the most cost-effective represents a few trees in tropical regions. As a result,
ways to achieve both of those goals [1–4]. While there Global Forest Watch, a widely used open-source tool
are local and regional estimates of regenerated forest used for monitoring forest change that uses the same
areas [5–7], global estimates are mostly derived from dataset, uses the term ‘tree cover’ rather than ‘forest
the Food and Agriculture Organization (FAO) of cover’ to describe the change detected [12].
the United Nations’ Global Forest Resources Assess- Remotely sensed data represents a viable and
ment (FRA) which quantifies forest area gained. practical first proxy for identifying areas where forest
Based on national land use reports, the FRA estim- regeneration may have occurred [13]. However, its
ated 140 million hectares of forest gained between use for this purpose remains more complex than the
2000 and 2015 [8]. However, the estimate is not spa- detection of forest loss [14], as forest regeneration
tially explicit. The currently available spatially expli- is a long and slow process. Forest regeneration can
cit global forest gain data [9] has limitations and does take many forms: 1. tree planting through industrial

© 2023 WWF-UK. Published by IOP Publishing Ltd

Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

plantation (herein, referred to as ‘plantation’), 2. act- based on the discussions with forest experts around
ive forest regeneration (including agroforestry, seed- the world (appendix section 1).
ling, site preparation, etc), and 3. natural forest regen-
eration (NFR) [12, 15]. NFR delivers great benefits for 2.1. Remote sensing analysis
biodiversity and other environmental services, such We used multiple sources of remote sensing datasets
as regulating the local hydrological cycle, mitigating and products to create a global forest regeneration
heat waves, and preventing landslides [16, 17]. How- map. The advantage of using multiple sources of data
ever, using remote sensing alone is almost impossible is particularly relevant at the global scale with diverse
to distinguish NFR from the other two forms of forest forest types and locations such that no single data-
regeneration, let alone other phenomena that can res- set or classification method works well enough. For
ult in an increase in biomass or greenness. instance, optical sensors such as Landsat have diffi-
Given these challenges, we used multiple remote culties in monitoring tropical areas with persistent
sensing datasets and products coupled with relevant cloud cover, but they work well in boreal regions [20,
local ancillary data and context information to cre- 21]; whereas active sensor data such as synthetic aper-
ate a global forest regeneration map. Using multiple ture radar data can help overcome the issues related to
remote sensing datasets and products from various cloud cover [19], but have difficulties with noise res-
satellite sources can help improve the mapping and ulting from snow cover in boreal regions [22].
monitoring of land cover change, as each of them can We used 23 land use and land cover remote sens-
complement the shortcomings of one another [18, ing datasets and products to identify areas of poten-
19]. Ancillary information provided by local experts tial forest regeneration using the following criteria
and practitioners on the ground around the world can (table 1): 1. classified as non-forest for 10 years before
help calibrate remote sensing results, identify the type 2000; 2. the presence of forest stand age younger than
of forest regeneration, exclude plantation areas that 4 years in 2000 or a statistically significant mono-
are only known locally, and provide local socioeco- tonically increasing trend in the biomass/greenness
nomic or policy context for the drivers of observed estimates. Such trend was described by various yearly
forest change. time series of vegetation indices between 2000 and
To gain an understanding of the biophysical 2015 using the Mann–Kendall test (p < 0.05) [37, 38]
and socioeconomic contexts of forest regeneration and the Sen’s non-parametric estimator [39] of slope
and where it has occurred, this study 1. developed greater than 0.01 on the normalised data; and 3. clas-
a method that combined available forest cover sified as forest from 2015 to 2018. We flagged areas
products, remote sensing data, and survey informa- where at least two remote sensing datasets supported
tion to map global forest regeneration; 2. collected each of the three criteria as potential forest regener-
local expert feedback via a survey to help validate the ation (figure 1). Threshold values of p-value, slope,
map and distinguish between different forms of forest and the number of datasets used for each criterion,
regeneration; and 3. acquired a preliminary under- were selected based on discussions with global forest
standing of why the regeneration has occurred based experts (appendix sections 2 and 3). Given that this
on the feedback. We aimed to provide a global view of study had a global scope, each remote sensing dataset
where forest regeneration has likely occurred and its and product was resampled to 250 m resolution using
contributing factors. The definition of forest regener- the nearest neighbour resampling method.
ation adopted in this study is defined specifically as
forest cover gain between 2000 and 2015 in areas that 2.2. Expert feedback
were not forested for 10 years prior and the regrown The intermediate map was calibrated with the
forest persisted until 2018. Such definition is intended feedback from local experts. A web platform was
to help exclude regrowth due to plantations and tem- developed to allow local experts to provide feed-
porary disturbance events such as fire or windthrows. back on the intermediate map generated by remote
sensing. Local experts were able to draw polygons
of areas they were familiar with. A list of ques-
2. Materials and methods tions associated with the polygon allows the local
experts to verify the map and provide contextual local
Multiple remote sensing datasets and products were information that enabled forest regeneration to occur
used to create an intermediate global forest regener- (appendix section 4). Qualitative analyses were used
ation map, which was sent to the forest experts for to 1. determine for each biome [40] if sufficient feed-
feedback through a standardised survey. The feed- back responses were received and whether they were
back was used to validate the intermediate map and representative; 2. detect the patterns of error in the
understand the local contextual factors that enabled intermediate map; and 3. correct the remote sensing
forest regeneration, and was incorporated to create estimates (appendix section 4.1). Regeneration areas
the final global forest regeneration map. A 25% can- labelled by the experts as due to plantations or natural
opy cover threshold was used for defining a forest disturbances were removed, and missed forest regen-
in the remote sensing analyses, which was concluded eration areas were added after confirmation with high

2
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

Table 1. Remote sensing data used in creating the global forest regeneration map. The ‘criterion’ column shows the criterion that a
remote sensing dataset was used to assess: ‘No forest’ corresponds to an area that was not forested between 1990 and 1999; ‘Growing’ to
forest growth between 2000 and 2015; ‘Persistent’ to the persistent forest cover between 2016 and 2018. ‘Res.’ column shows the
resolution of the original dataset. For the datasets, ‘ESA’ stands for ‘European Space Agency’; ‘CCI’ for ‘Climate Change Initiative’;
‘NDVI’ for ‘Normalised Difference Vegetation Index’; ‘VCF’ for ‘Vegetation Continuous Fields’; ‘MODIS’ for ‘Moderate Resolution
Imaging Spectroradiometer’; ‘CORINE’ for ‘Coordination of Information on the Environment’; ‘LAI’ for ‘Leaf Area Index’; ‘ALOS’ for
‘Advanced Land Observing Satellite’. ‘Stat. sig. mono. inc.’ stands for ‘statistically significant monotonically increasing with p-value
<0.5’. Complete list of dataset assessed and the justification of whether a dataset was used and how it was processed can be found in
appendix section 2.

Criterion Data Period Area Res. Selection

No forest ESA CCI Land 1992–1999 Global 300 m Non-forest land cover
Cover [23]
No forest Landsat NDVI 1990–1999 Global 20 m Annual average
[24] NDVI ⩾ 0.6 and its
standard deviation ⩾ 0.1
No forest Landsat VCF tree 2000 Global 30 m Tree canopy cover less than
cover [25] 25%
No forest Brazil Land Cover 1990–1999 Brazil 30 m Non-forest land cover
(MapBiomas) [26]
No forest CORINE Land 1990, 2000 Europe 100 m Non-forest land cover
Cover [27]
Growing MODIS VCF 2000–2015 Global 250 m Stat. sig. mono. inc. trend
percent tree cover of percent tree canopy
[28] cover
Growing MODIS NDVI 2000–2015 Non-Boreal 250 m Stat. sig. mon. inc. trend of
[29] annualised NDVI between
25th and 75th percentile
Growing MODIS EVI [29] 2000–2015 Non-Boreal 250 m Stat. sig. mono. inc. trend
of annualised EVI between
25th and 75th percentile
Growing MODIS LAI [30] 2002–2015 Boreal 500 m Stat. sig. mono. inc. trend
of annualised LAI between
25th and 75th percentile
Growing Landsat VCF tree 2000–2015 Global 30 m Tree canopy cover ⩾ 25%
cover [25] in 2015 and experienced
growth since 2000
Growing Tropical Biomass 2003–2014 Tropics 500 m Biomass gain ⩾ 50t/ha
[31]
Growing Russia Young Tree 2012 Russia 500 m Tree age < 5 years old in
[32] 2000
Growing Canada Young 2004 Canada 1 km Tree age < 5 years old in
Tree [33] 2000
Growing Canada Tree Age 2001, 2010 Canada 250 m Stand age difference
[34] between 2010 and 2001
< 15 years
Growing Alaska Young Tree 2006 Alaska 1 km Tree age < 5 years old in
[33] 2000
Growing Mekong Tree Gain 2001–2005 Mekong 30 m Tree gain
[35]
Persistent ALOS 2015–2017 Non-Boreal 25 m Forest land cover
Forest/Non-forest
[22]
Persistent ESA CCI Land 2015–2018 Global 300 m Forest land cover
Cover [23]
Persistent Global Forest 2015–2018 Global 30 m Forest land cover
Change [9]
Persistent Mekong Forest 2015–2016 Mekong 30 m Forest land cover
Extent [35]
Persistent Brazil Land Cover 2015–2017 Brazil 30 m Forest land cover
(MapBiomas) [26]
Persistent ESA Africa Land 2016 Africa 30 m Forest land cover
Cover [36]
Persistent CORINE Land 2018 Europe 100 m Forest land cover
Cover [27]

3
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

Figure 1. Maps of criteria used to determine potential forest regeneration areas. Each map shows where at least two remote
sensing datasets and products agree with (a) no forest between 1990 and 1999; (b) growing forest between 2000 and 2015;
(c) persistent forest between 2015 and 2018.

resolution remote sensing time series images. Based regeneration areas with their 1 km buffer. For the
on the conclusion from the qualitative analyses of the global level, 2141 out of 2500 sample points were
feedback, plantation areas were further masked out used. Because the proportion of the total mapped
based on the local land use information, the Spatial forest regeneration area is only 0.1% of the land area,
Database of Planted Trees [41], and the Global Oil to gain control over the sample points allocated to
Palm Plantation Map [42] to create the final forest the mapped regeneration area, we created 10 000 km2
regeneration map. hexagonal grid cells covering all land areas. Hexagons
with 5000 or more pixels of mapped forest regen-
2.3. Accuracy assessment eration were selected, which is equivalent to at least
Accuracy assessments were conducted for the final 3.1% of the area of a hexagon. We used 2145 out of
regeneration map at three levels: 1. global, 2. selec- the 2700 stratified random sampling points distrib-
ted hexagons with each consists of at least 3.1% of uted across those selected hexagons. The buffer-level
mapped forest regeneration area, and 3. the mapped assessment was used to calculate the 95% confidence

4
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

Figure 2. To make it visible at the global level, the forest regeneration map was aggregated by hexagons of 100 000 km2 each. Each
hexagon indicated the number of pixels that indicated forest regeneration within. Projection of the map is Eckert IV, an equal area
projection. Interactive version of the map that is not aggregated can be accessed at this website.

intervals of the area estimates [43], in which, 2020 out 2010 [48] by ESRI [49] and the Agricultural Suitab-
of the 2200 stratified random sampling points were ility of Global Soil [50] data for the major regions
used. For all three levels, the sampling points were dis- in the world. The Agricultural Suitability of Global
tributed across the land areas where Antarctica was Soil data was developed based on an index derived
excluded. We visually assessed whether the 250 m × from climate and soil data that represents the suitab-
250 m surrounding area of each sample point satisfied ility level for cultivation in a particular location [51]
all three criteria of forest regeneration. High spatial and provides the fraction of each grid cell suitable
and temporal resolution images from 1990 to 2018 for cultivation [50] (appendix section 9). The ana-
available on Google Earth Pro were used as reference lysis was done for areas that were identified by forest
land cover. Sample points were eliminated due to the experts as NFR, active regeneration, and without any
inaccessibility of high spatial resolution time series identification. The map of annual gross economic
images necessary for validation. Accuracy assessment rents of the world’s crop and grazing lands of 2005
of each level was summarised in a confusion matrix [52] was used to compare the opportunity cost of
[44] (appendix section 5). agriculture where forest regeneration had taken place
and where it had not (appendix section 11).
2.4. Additional analyses
The final global forest regeneration map was com-
bined with tree density predictions using the ecore- 3. Results
gion and the biome-level models [45] to estimate the
total number of trees regenerated. The resolution of 3.1. Global forest regeneration map and accuracy
the tree density maps is 897.3 m. Each pixel was par- assessments
titioned to ensure the correct number of trees were Our results show an estimated 55.7 ± 6.2 million ha
counted in each resampled 250 m resolution pixel where forests appear to have regrown between 2000
(appendix section 8). We estimated the regenerated and 2015 in areas that were not forested from 1990 to
forest biomass by combining the map and the bio- 1999 and remained forested until 2018. These regen-
mass density estimates of forests with age younger erated forests could represent, based on the spatial
than 20 [46]. Calculations were performed for a hypo- model result of tree density at maturity [45], between
thetical scenario where 100% of the regenerated forest 22.0 and 25.1 billion trees. Total aboveground bio-
were natural (appendix section 8). A map with a mass (dry mass) recovered is 3.2 billion tonnes, or 5.9
finer definition of ecoregions produced by the Nature Gt CO2 e, if all were naturally regenerated. About 1%
Conservancy [47] was used to estimate the encroach- of the total mapped areas were located in the tundra
ment of forest into non-forest ecoregions (refer to ecoregion and about 3.6% in grassland/savanna eco-
appendix section 10 and dataset S5 for the classific- regions (appendix section 10).
ation of the forest and non-forest ecoregions). Remote sensing results showed clustering of
To understand the dynamics of different types of apparent regeneration in the boreal region (includ-
forest regeneration with agriculture, the final map ing Russia, Canada, and northern Europe, espe-
was overlaid with global slope values calculated from cially Sweden), northern Mongolia, central China, the
the Global Multi-resolution Terrain Elevation Data lower Mekong region, Europe around the Caspian

5
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

Figure 3. Zoom-in views of the global forest regeneration map and the expert feedback in: (a) Mongolia; (b) Greece; (c) Brazil;
(d) Gabon. Green area indicates where forest regeneration occurred between 2000 and 2015 and remained as forest until 2018 in
areas that were not forested between 1990 and 1999. The red boundaries were drawn by local experts where they could indicate
the type of forest regeneration (blue text), the current status of the forest, and the policy and socioeconomic context of the area.

Sea and the Mediterranean Sea, west Africa, north- Table 2. Top ten countries with the highest mapped forest
regeneration area (in million hectares) and the percentage share
ern Congo Basin, southeastern Brazil, Cuba, Cent- of the total mapped area. Full table is available in the supporting
ral America, and southeastern United States (figure 2, dataset S1.
figure 3). Large countries such as Russia, China, and Country Area (M ha) Share (%)
Canada showed the largest areas of mapped regen-
erating forest (table 2). The top ten countries with Russia 15.85 28.42
the highest area mapped accounted for 78% of the China 7.80 13.98
total. When accounting the mapped regeneration area Canada 7.46 13.38
Brazil 4.27 7.66
relative to the land area of a country, smaller coun-
United States 2.46 4.40
tries in the Caribbean, the lower Mekong, and eastern Central African Republic 1.45 2.60
Europe topped the list, with the top ten of these coun- Mongolia 1.24 2.21
tries accounting for 23.1% of the total (table 3) (total Thailand 1.18 2.12
and relative regeneration areas for all countries are Mexico 0.91 1.63
available in dataset S1). The area estimates of forest Vietnam 0.71 1.27
regeneration in the tropical regions (between 25◦ N
and 25◦ S) of our study were compared with those
from Fagan et al (2022) for a similar time period [7] The accuracy levels for nearly all biomes are above
(table 4). 84%, except for ‘Tropical and Subtropical Grasslands,
Accuracy assessment at the global level showed Savannas, and Shrublands’, a non-forest biome, with
an overall accuracy of 86.8 ± 1.5% (the uncertainty accuracy of about 77.5% (appendix section 6).
is expressed as the estimate ±95% confidence inter-
val, henceforth). Producer’s accuracy was estimated 3.2. Expert-local feedback
at 94.8% while the user’s accuracy was estimated at Expert feedback was collected from 123 sites in 29
72.5% for the regeneration class. At the hexagon level, countries, 13 biomes, and accounted for 141.8 mil-
the overall accuracy was 90.6 ± 1.4%. Producer’s and lion ha. Qualitative analyses of the feedback showed
the user’s accuracies of the regeneration class are that additional feedback is needed for the Tropical
94.6% and 76.3% respectively. See appendix section and Subtropical Coniferous Forests biome, because
5 for the confusion matrices and detailed explana- the percentage of areas received feedback accounted
tions of validation procedures. Global-level valida- for less than that of the forest regeneration mapped
tion sample points were also partitioned by biomes. using remote sensing analyses (table 5). Additionally,

6
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

Table 3. Top ten countries with the highest proportion (Prop.) of mapped forest regeneration per total land area and total mapped forest
regeneration area (in thousand hectares). Full table is available in the supporting dataset S1.

Country Area (k ha) Prop. (%)

Cuba 391.56 3.57

Guinea-Bissau 99.87 2.98
Anguilla 0.25 2.72
Central African Republic 1448.54 2.33
Thailand 1183.54 2.30
Vietnam 709.66 2.18
North Macedonia 53.19 2.09
Saint Martin 0.11 2.06
Côte d’Ivoire 561.44 1.74
Albania 45.29 1.58

Table 4. Comparison with Fagan et al [7] for estimated area of forest gained due to natural regrowth and the corresponding 95%
confidence interval in various regions located within 25◦ N and 25◦ S. Note that our study period is from 1990 to 2018, while that for
Fagan et al [7] is from 2000 to 2015.

Tropical Region Our study (Mha) Fagan et al [7] (Mha)

Africa 5.35 ± 1.19 7.2 ± 8.5

Latin America 6.41 ± 1.92 14.8 ± 7.2
Asia+Oceania 5.01 ± 1.81 9.5 ± 4.1
All tropics 16.78 ± 2.79 31.6 ± 11.9

Table 5. Proportions (Prop.) of the intermediate mapped forest regenerated areas and expert feedback (FB) over the land area of each
biome. The mapped regeneration area was derived from remote sensing datasets, before incorporating any additional feedback and
further masking of plantation data.

Biome Regen Prop. (%) FB Prop. (%)

Boreal Forests/Taiga 1.22 1.59

Deserts and Xeric Shrublands 0.01 0.14
Flooded Grasslands and Savannas 0.37 0.00
Mangroves 0.38 0.37
Mediterranean Forests, Woodlands and Scrub 0.24 2.99
Montane Grasslands and Shrublands 0.10 0.24
Rock and Ice 0 0
Temperate Broadleaf and Mixed Forests 0.82 1.90
Temperate Conifer Forests 0.85 0.92
Temperate Grasslands, Savannas and Shrublands 0.18 0.20
Tropical and Subtropical Coniferous Forests 0.52 0
Tropical and Subtropical Dry Broadleaf Forests 0.77 1.21
Tropical and Subtropical Grasslands, Savannas and Shrublands 0.32 1.89
Tropical and Subtropical Moist Broadleaf Forests 0.56 2.30
Tundra 0.05 0.06

most of the errors were contributed by falsely identi- regeneration was reported to have occurred on newly
fying plantation as regeneration (appendix section available lands due to naturally occurring phenom-
4.1). Based on the local information provided, addi- ena such as meandering rivers. The impact of pro-
tional plantation and naturally disturbed areas were tected areas was interpreted as fostering NFR by the
further masked to create the final global forest regen- reduction in grazing, illegal logging, and fires, which
eration map. ensure the survival of the seedlings. Lastly, NFR was
Regarding the underlying factors thought to also reported to occur in abandoned or fallow agri-
enable forest regeneration, 42% of the responses cultural and pastoral lands. The abandonment was
reported NFR and 19% reported active regenera- reportedly due to increases in labour cost, migration,
tion such as assisted regeneration and agroforestry and mechanisation.
(dataset S2). Enabling factors for NFR could be cat-
egorised into three main types: 1. natural dynam- 3.3. Interaction with agriculture
ics (17%); 2. establishment and impact of protec- The forest regeneration identified by local experts as
ted areas (33%); and 3. agriculture or pasture land NFR tended to occur in areas with a steeper slope,
abandonment (42%). For natural dynamics, forest compared with those identified as active regeneration

7
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

or those unidentified (figure S1). The only excep- our study having a longer time frame, a higher canopy
tion is eastern Asia, where the mean of the slope cover threshold in defining forest (Fagan et al 2022
for NFR (mean = 8.65◦ ) was significantly less than used 10% as the threshold [7]), and more restric-
that of both unidentified (mean = 10.21◦ ) and active tions on the prior land cover. Our estimate could be
(mean = 9.60◦ ) regeneration areas, according to the considered as conservative because for an area to be
unpaired two-samples Wilcoxon test (p < 0.0001). In identified, at least two remote sensing datasets were
general, regeneration occurred in areas with a slope needed to support each and all of the criteria stated
less than 10◦ , except for eastern and southern Asia. above. Such an approach was used to balance the
Forest regeneration occurred over a wide range of need of minimising false positives, while providing
levels of soil suitability for cultivation (figure S2). The information that can be used for large-scale planning.
forest regeneration map showed that most regenera- Estimating the proportion of NFR within the total
tion in the boreal region took place at where the soil area of identified forest regeneration remains highly
was the least suitable for agriculture. The locations uncertain, as the total area of feedback received was
include northern Europe, northern America (Canada limited. Collecting feedback from around the world
and the United States), and Asiatic Russia, where most remains challenging. If we assume the survey feed-
regeneration occurred in areas with less than 14% soil back represented a comprehensive and an unbiased
suitability for cultivation. sample of the total mapped regeneration area, then
For some non-boreal regions, regeneration also out of the 55.7 million ha of potentially regenerated
tended to occur in regions with low soil cultivation forests, 42% could be attributed to NFR. However,
suitability, such as areas in southeastern Asia, South neither remote sensing nor experts can accurately dif-
America, and Sub-Saharan Africa. Regeneration in ferentiate between active regeneration and NFR at the
developed regions such as Europe and Australia/New global scale, because in reality, spontaneous forest
Zealand occurred in places with a wide range of soil regeneration can happen in between intensely man-
suitability levels and the mean and median values aged forests [53]. Additionally, NFR could have been
were on the higher ends of suitability (over 53% of missed in the boreal region due to the slow growing
the areas were suitable for cultivation). In contrast, nature of secondary forests or dry forest regions due
in central America (including Mexico) and the Carib- to low vegetation signal-to-noise ratios [54].
bean, the majority of regeneration occurred in areas
with greater than 80% soil suitability for cultivation.
4.2. Forest encroachment to other ecoregions
But areas reported as NFR were located in lands with
Forest encroachment to other non-forest ecoregions
soil less suitable for cultivation than those reported as
creates more negative impacts on the environment
agroforestry/active regeneration, except for Europe.
than positive. About 4.6% of the mapped forest
For each country, the agriculture opportunity
regrowth occurred in non-forest biomes. Such forest
cost of the sites where forest regeneration happened
encroachment in non-forest biomes, such as tundra
was less than where it did not occur. The average dif-
[55, 56], grassland/savanna [55], or other previ-
ference was about $126/ha. Distribution of the dif-
ously non-forested areas at high altitude [57] could
ference in the agriculture opportunity cost for all the
be part of the complex interactions and dynamics
countries is available in appendix section 11.
between the changing climate and biomes [55]. Forest
encroachment in those non-forest biomes not only
4. Discussion causes a positive feedback that exacerbates global
warming due to the decrease in albedo as darker
4.1. On the scale of forest regeneration and carbon
tree canopy covers replace lighter surfaces such as
sequestration
snow [58] and grass. It also results in the loss of
In comparison with the FRA estimates, where
biodiversity [59]. In areas vulnerable to fire, burning
150 million ha of forest were reported as recovered
of the encroached forests releases more carbon than
between 2000 and 2015 [8], our estimate of
grasses [60].
55.7 million ha is a much lower figure. However, our
study results are complementary to the estimate of the
FAO and not suitable for direct comparison, as the 4.3. Conditions driving forest regeneration
FAO estimate was based on national land use reports The conditions driving forest regeneration are quite
[8] while our estimate was based on land cover data. different when comparing boreal and non-boreal
The recovered forest that countries reported could regions. For the boreal region, forest regeneration is
be too young to be detected by remote sensing and mostly associated with events caused by global cli-
the exclusion of the areas that were forested from mate change, such as post-fire or post-disturbance
1990 to 1999 likely resulted in the lower estimate in regrowth [61, 62]. Global climate change also disturbs
our study. Similarly, our estimates were smaller than the fire and hydrological cycles, which could lead to
those of Fagan et al [7], but they fell within the range forest loss [63, 64] and species composition change
of uncertainties. The smaller values are likely due to [63] in the boreal region. For example, local experts

8
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

from Mongolia confirmed changes in species com- where the protected area successfully enhanced NFR,
position in the northern part of the country. Spe- it was part of larger land use planning that benefit-
cies composition change can be detected from remote ted the communities, such as mitigating landslides,
sensing as greening [65]. increasing agricultural yield as the result of improv-
In contrast, forest regeneration in non-boreal ing habitats, or generating economic benefits from
region was more closely related to agricultural land tourism. For long term sustainability of a forest res-
use dynamics, as depicted in the forest transition toration project, the interests of the local stakeholders
theory [66]: it is expected that forest loss during the must be taken into account.
early stages of economic development will be fol- We further invite additional feedback from
lowed by recovery at later stages in areas where the experts about the socioeconomic, policy, and his-
dependence on agriculture declines, or when coun- torical context of the mapped regenerated areas4 .
tries depend more on agricultural imports [15, 66]. The goal is to form a baseline understanding of the
The pattern of the distribution of forest regeneration current conditions (2000–2018) that resulted in suc-
across soil suitability supports such theory (figure cessful forest regeneration so that it can be promoted
S2): it was more evenly distributed in the more in places where there appear to be low barriers to
economically developed geographical regions, where success. Remotely sensed data, when coupled with
most of the countries belong to the post-transition periodic local surveys, can provide information for
status (e.g. Europe, Australia/New Zealand), than in land managers to meet the different needs of vari-
regions where most countries are dependent on the ous stakeholders [13]. Additional expert feedback
agricultural sector [67]. Another evenly distributed would still be needed in order for the feedback to be
region was eastern Asia, where the majority of the representative of the distribution of the regenerated
mapped regeneration came from China. Its govern- forest, particularly for the following biomes: Boreal
ment policies for nation-wide afforestation/reforest- Forests/Taiga, Temperate Coniferous Forests, Tem-
ation programmes were the primary factor contrib- perate Grasslands Savannas and Shrublands, Tropical
uting to forest regeneration [68]. For regions that and Subtropical Coniferous Forests, and Tropical and
are highly dependent on the agricultural sector, such Subtropical Dry Broadleaf Forests (see dataset S3
as South America and sub-Saharan Africa, forest for details). The results of this study show NFR has
regeneration took place in areas with lower levels been occurring at significant scale and the enabling
of soil suitability, suggesting that agricultural aban- conditions appear to differ substantially across con-
donment occurred in marginal lands. In southeast- tinents, biomes and countries. Therefore, to better
ern Asia, the detected forest regeneration was likely achieve global goals in carbon reduction and improv-
due to regrowth after land was left fallow under shift- ing habitats, such as the Bonn Challenge, we suggest
ing agriculture [69–71], an observation confirmed by prioritising areas for in-depth analyses based on local
feedback from various experts. Similar feedback was relevance and demand.
received from experts for areas in southern Mexico.
Data availability statement
4.4. Implication to forest restoration policies and
looking forward All data that support the findings of this study are
A sustainable project would be located in an area included within the article (and any supplementary
where biophysical or socioeconomic risk is minim- files).
ised. Forests that regenerated in areas highly suitable
for agriculture tend to be more productive and resili- Acknowledgments
ent to biophysical risks such as fires or drought [5].
Since many of the regenerated forests identified in This research was supported by Trillion Trees, a
this study were found on land with poorer soil suit- coalition of BirdLife International, WCS, and WWF-
ability, they might be more vulnerable to biophysical UK, as well as the WWF-Forest Practice Innova-
risks. Socioeconomic risk cannot be generalised at the tions funds. We thank Bart Gajc, Aurelie Shapiro,
global level. However, a successful restoration project Helga Kuechly, Espin Bowder, Emily Mills, and Nas-
has to be locally based, inclusive, socially desirable, ser Olwero in the Gobil team for helping us to upload
and satisfying to the interests of various stakeholders the map onto the Gobil server for collecting feed-
at both the local and national level [72]. Establishing back from experts and solving other technical issues,
protected areas, such as national parks or indigenous as well as the two anonymous reviewers for help-
territories, if they are well managed, allows for better ful comments and suggestions. We would also like
fire management and decreases pressure from graz- to thank the experts who provided feedback on the
ing or other land uses, which could ensure the sur- regeneration area, both verbally and also by drawing
vival of young trees [73]. An important note is that the polygons.
the effectiveness of protected areas depends on the
interests of local stakeholders. In all reported cases 4 The global forest regeneration map is available at this website.

9
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

ORCID iDs [28] DiMiceli C, Carroll M, Sohlberg R, Kim D H, Kelly M and

Townshend J 2015 MOD44B MODIS/Terra Vegetation
Continuous Fields Yearly L3 Global 250m SIN Grid V006
Pui-Yu Ling  https://orcid.org/0000-0002-3540-
NASA EOSDIS Land Processes DAAC (https://doi.org/
609X 10.5067/MODIS/MOD44B.006)
William Baldwin-Cantello  https://orcid.org/0000- [29] Didan K 2015 MOD13Q1 MODIS/Terra Vegetation Indices
0003-3427-4379 16-Day L3 Global 250m SIN Grid V006 NASA EOSDIS Land
Processes DAAC (https://doi.org/10.5067/MODIS/
April L Bagwill  https://orcid.org/0000-0003-0046-
MOD13Q1.006)
1491 [30] Myneni R, Knyazikhin Y and Park T 2015 MCD15A2H
Pablo Pacheco  https://orcid.org/0000-0001-7192- MODIS/Terra+Aqua Leaf Area Index/FPAR 8-day L4
6861 Global 500m SIN Grid V006 NASA EOSDIS Land
Processes DAAC (https://doi.org/10.5067/MODIS/
MCD15A2H.006)
References [31] Baccini A et al 2012 Nat. Clim. Change 2 182–5
[32] Loboda T V and Chen D 2016 Distribution of Young Forests
[1] Gilroy J J, Woodcock P, Edwards F A, Wheeler C, and Estimated Stand Age across Russia, 2012 ORNL DAAC,
Baptiste B L, Uribe C A M, Haugaasen T and Edwards D P Oak Ridge, TN (https://doi.org/10.3334/ORNLDAAC/1330)
2014 Nat. Clim. Change 4 503–7 [33] Pan Y, Chen J M, Birdsey R, McCullough K, He L and Deng F
[2] Griscom B W et al 2017 Proc. Natl Acad. Sci. 114 11645–50 2012 NACP Forest Age Maps at 1-km Resolution for Canada
[3] Busch J, Engelmann J, Cook-Patton S C, Griscom B W, (2004) and the U.S.A. (2006) ORNL DAAC, Oak Ridge, TN
Kroeger T, Possingham H and Shyamsundar P 2019 Nat. (https://doi.org/10.3334/ORNLDAAC/1096)
Clim. Change 9 463–6 [34] Beaudoin A, Bernier P Y, Villemaire P, Guindon L and
[4] Strassburg B B et al 2019 Nat. Ecol. Evol. 3 62–70 Guo X J 2017 Species composition, forest properties and land
[5] Poorter L et al 2016 Nature 530 211–4 cover types across Canada’s forests at 250m resolution for
[6] Baccini A, Walker W, Carvalho L, Farina M, 2001 and 2011 Natural Resources Canada, Canadian Forest
Sulla-Menashe D and Houghton R 2017 Science 358 230–4 Service, Laurentian Forestry Centre, Quebec, Canada (https://
[7] Fagan M E et al 2022 Nat. Sustain. 5 681–8 doi.org/10.23687/ec9e2659-1c29-4ddb-87a2-6aced147a990)
[8] FAO 2020 Global Forest Resources Assessment 2020—Key [35] Potapov P et al 2019 Remote Sens. Environ. 232 111278
Findings (available at: https://doi.org/10.4060/ca8753en) [36] European Space Agency 2018 S2 prototype lc map at 20m of
(Accessed 14 July 2020) africa 2016 (available at: www.esa.int/ESAMultimedia/
[9] Hansen M C et al 2013 Science 342 850–3 Images/2017/10/Africanlandcover) (Accessed 25 December
[10] Baccini A, Walker W, Carvalho L, Farina M and 2019)
Houghton R A 2019 Science 363 eaat1205 [37] Mann H B 1945 Econometrica 13 245
[11] Hansen M, Potapov P, Margono B, Stehman S, Turubanova S [38] Kendall M G 1948 p 160 (available at: http://hdl.handle.net/
and Tyukavina A 2014 Science 344 981 123456789/11758)
[12] Chazdon R L, Brancalion P H S, Laestadius L, [39] Sen P K 1968 J. Am. Stat. Assoc. 63 1379–89
Bennett-Curry A, Buckingham K, Kumar C, Moll-Rocek J, [40] Olson D M et al 2001 BioScience 51 933
Vieira I C G and Wilson S J 2016 Ambio 45 538–50 [41] Harris N, Goldman E D and Gibbes S 2019 Spatial Database
[13] Cordell S, Questad E J, Asner G P, Kinney K M, Thaxton J M, of Planted Trees (SDPT Version 1.0) Technical Report
Uowolo A, Brooks S and Chynoweth M W 2017 Restor. Ecol. (Washington, DC: World Resources Institute) (available at:
25 S147–54 www.wri.org/publication/spatial-database-planted-trees)
[14] Chazdon R L and Guariguata M R 2016 Biotropica 48 716–30 [42] Descals A, Wich S, Meijaard E, Gaveau D L A, Peedell S and
[15] Wilson S J, Schelhas J, Grau R, Nanni A S and Sloan S 2017 Szantoi Z 2021 Earth System Science Data 13 1211–31
Ecol. Soc. 22 38 [43] Olofsson P, Foody G M, Stehman S V and Woodcock C E
[16] Locatelli B, Catterall C P, Imbach P, Kumar C, Lasco R, 2013 Remote Sens. Environ. 129 122–31
Marín-Spiotta E, Mercer B, Powers J S, Schwartz N and [44] Olofsson P, Foody G M, Herold M, Stehman S V,
Uriarte M 2015 Restor. Ecol. 23 337–43 Woodcock C E and Wulder M A 2014 Remote Sens. Environ.
[17] Crouzeilles R, Ferreira M S, Chazdon R L, Lindenmayer D B, 148 42–57
Sansevero J B B, Monteiro L, Iribarrem A, Latawiec A E and [45] Crowther T W et al 2015 Nature 525 201–5
Strassburg B B N 2017 Sci. Adv. 3 e1701345 [46] Domke G et al 2019 Forest land 2019 Refinement to the 2006
[18] Joshi N et al 2016 Remote Sens. 8 70 IPCC Guidelines for National Greenhouse Gas Inventories
[19] Reiche J et al 2016 Nat. Clim. Change 6 120–2 (Agriculture, Forestry and Other Land Use) vol 4 ed
[20] Nolin A W 2010 J. Glaciol. 56 1141–50 E Calvo Buendia, K Tanabe, A Kranjc, J Baasansuren,
[21] Thiel C and Schmullius C 2016 Remote Sens. Environ. M Fukuda, S Ngarize, A Osako, Y Pyrozhenko, P Shermanau
173 258–73 and S Federici (Switzerland: Intergovernmental Panel on
[22] Shimada M, Itoh T, Motooka T, Watanabe M, Shiraishi T, Climate Change (IPCC)) pp 4.1–4.71 (available at: www.
Thapa R and Lucas R 2014 Remote Sens. Environ. 155 13–31 ipcc-nggip.iges.or.jp/public/2019rf/pdf/4_Volume4/19R_
[23] European Space Agency 2019 ESA Climate Change Initiative V4_Ch04_Forest%20Land.pdf)
Land Cover project (available at: http://maps.elie.ucl.ac.be/ [47] The Nature Conservancy 2009 TNC—Terrestrial Ecoregions
CCI/viewer/download.php) (Accessed 25 December 2019) (available at: http://maps.tnc.org/files/metadata/TerrEcos.
[24] Masek J, Vermote E, Saleous N, Wolfe R, Hall F, xml) (Accessed 24 April 2021)
Huemmrich K, Gao F, Kutler J and Lim T K 2006 IEEE [48] Danielson J J and Gesch D B 2011 Global Multi-resolution
Geosci. Remote Sens. Lett. 3 68–72 Terrain Elevation Data 2010 (GMTED2010) US Department
[25] Sexton J O et al 2013 Int. J. Digital Earth 6 427–48 of the Interior, US Geological Survey (https://doi.org/
[26] MapBiomas Project 2019 Collection 4 of the annual land use 10.5066/F7J38R2N)
land cover maps (available at: https://mapbiomas.org/en/ [49] ESRI 2009 World Slope GMTED (available at: https://
download?cama_set_language=en) (Accessed 2 December landscape6.arcgis.com/arcgis/rest/services/World_Slope_
2019) GMTED/ImageServer) (Accessed 2 July 2020)
[27] European Environmental Agency 2019 Copernicus [50] Ramankutty N, Foley J, Norman J and McSweeney K 2011
land monitoring service—corine land cover (available Agricultural suitability of global soils (availability at: https://
at: www.eea.europa.eu/ds_resolveuid/ecb838dabf48498 databasin.org/datasets/fdfcc35510ba44ac8c174221f1762e2d)
38ba5f3dc81ca6b0e) (Accessed 10 January 2020) (Accessed 2 July 2020)

10
Environ. Res. Lett. 18 (2023) 054025 P-Y Ling et al

[51] Ramankutty N, Foley J A, Norman J and McSweeney K 2002 [62] Sulla-Menashe D, Woodcock C E and Friedl M A 2018
Glob. Ecol. Biogeogr. 11 377–92 Environ. Res. Lett. 13 014007
[52] Naidoo R and Iwamura T 2007 Biol. Conserv. 140 40–49 [63] Buermann W, Parida B, Jung M, MacDonald G M, Tucker C J
[53] Reid J L, Fagan M E and Zahawi R A 2018 Sci. Adv. and Reichstein M 2014 Geophys. Res. Lett. 41 1995–2002
4 eaas9143 [64] Carpino O A, Berg A A, Quinton W L and Adams J R 2018
[54] Smith W K et al 2019 Remote Sens. Environ. 233 111401 Environ. Res. Lett. 13 084018
[55] Criado M G, Myers-Smith I H, Bjorkman A D, [65] Myers-Smith I H et al 2020 Nat. Clim. Change 10 106–17
Lehmann C E R and Stevens N 2020 Glob. Ecol. Biogeogr. [66] Rudel T K 1998 Rural Sociol. 63 533–52
29 925–43 [67] Meyfroidt P and Lambin E F 2011 Annu. Rev. Environ.
[56] Terskaia A, Dial R J and Sullivan P F 2020 Ecography Resour. 36 343–71
43 769–78 [68] Chen C et al 2019 Nat. Sustain. 2 122–9
[57] Malfasi F and Cannone N 2020 Ecosystems 23 1657–75 [69] Cochard R, Ngo D T, Waeber P O and Kull C A 2017
[58] Zhang W, Miller P A, Jansson C, Samuelsson P, Mao J and Environ. Rev. 25 199–217
Smith B 2018 Geophys. Res. Lett. 45 7102–11 [70] Mukul S A, Herbohn J and Firn J 2016 Sci. Rep. 6 22483
[59] Abreu R C, Hoffmann W A, Vasconcelos H L, Pilon N A, [71] Kim S B and Alounsavath O 2015 Forest Sci. Technol.
Rossatto D R and Durigan G 2017 Sci. Adv. 3 e1701284 11 166–71
[60] Bond W J, Stevens N, Midgley G F and Lehmann C E 2019 [72] Ashton M S, Goodale U M, Bawa K S, Ashton P S and
Trends Ecology Evol. 34 963–5 Neidel J D 2014 Forest Ecol. Manag. 329 335–9
[61] Helbig M, Pappas C and Sonnentag O 2016 Geophys. Res. [73] Augustine D J, Frelich L E and Jordan P A 1998 Ecol. Appl.
Lett. 43 1598–606 8 1260–9

11