See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/348666331

Global maps of twenty-first century forest carbon fluxes

Article in Nature Climate Change · March 2021

DOI: 10.1038/s41558-020-00976-6

CITATIONS READS

560 10,916

17 authors, including:

Nancy L Harris David Gibbs

Winrock International World Resources Institute
24 PUBLICATIONS 6,062 CITATIONS 21 PUBLICATIONS 1,199 CITATIONS

SEE PROFILE SEE PROFILE

A. Baccini Sytze de Bruin

Boston University Wageningen University & Research
99 PUBLICATIONS 13,594 CITATIONS 197 PUBLICATIONS 5,996 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Rosa Maria Roman-Cuesta on 06 February 2022.

The user has requested enhancement of the downloaded file.

 Articles
https://doi.org/10.1038/s41558-020-00976-6

Global maps of twenty-first century forest carbon

fluxes
Nancy L. Harris 1 ✉, David A. Gibbs 1, Alessandro Baccini2,10, Richard A. Birdsey2, Sytze de Bruin 3
,
Mary Farina2,11, Lola Fatoyinbo4, Matthew C. Hansen 5, Martin Herold 3, Richard A. Houghton2,
Peter V. Potapov 5, Daniela Requena Suarez 3, Rosa M. Roman-Cuesta6, Sassan S. Saatchi7,8,
Christy M. Slay 9, Svetlana A. Turubanova5 and Alexandra Tyukavina5

Managing forests for climate change mitigation requires action by diverse stakeholders undertaking different activities with
overlapping objectives and spatial impacts. To date, several forest carbon monitoring systems have been developed for differ-
ent regions using various data, methods and assumptions, making it difficult to evaluate mitigation performance consistently
across scales. Here, we integrate ground and Earth observation data to map annual forest-related greenhouse gas emissions
and removals globally at a spatial resolution of 30 m over the years 2001–2019. We estimate that global forests were a net
carbon sink of −7.6 ± 49 GtCO2e yr−1, reflecting a balance between gross carbon removals (−15.6 ± 49 GtCO2e yr−1) and gross
emissions from deforestation and other disturbances (8.1 ± 2.5 GtCO2e yr−1). The geospatial monitoring framework introduced
here supports climate policy development by promoting alignment and transparency in setting priorities and tracking collective
progress towards forest-specific climate mitigation goals with both local detail and global consistency.

C
limate change must be addressed by various actors includ- non-anthropogenic sink of atmospheric carbon on land—predomi-
ing scientists, policymakers, companies, investors and civil nantly forests9—is then inferred as the residual of the other terms
society, all of whom operate under different mandates and of the global carbon budget1. Another approach compiles national
capabilities. Both IPCC reports1,2 and the Paris Agreement3 rec- GHG inventories (GHGIs), which reflect methodologies developed
ognize that climate change mitigation goals cannot be achieved by the IPCC and agreed to under the United Nations Framework
without a substantial contribution from forests but monitoring the Convention on Climate Change10,11. The quality, methodological
extent to which forests impact atmospheric greenhouse gas (GHG) complexity and sources of data used by each country vary, as do the
concentrations is challenging. Opposing fluxes (emissions from completeness and frequency of reporting. These approaches produce
sources (+) and removals by sinks (-)) occur simultaneously within dissimilar global net forest fluxes; GHGI estimates compiled from
regions on the basis of where and when disturbance and manage- country reports are 4.3 GtCO2 yr−1 lower than global estimates from
ment take place, interannual variability can be high and land-use models summarized in IPCC reports—a discrepancy larger than the
patterns are more dynamic and operate on finer spatiotemporal total annual emissions of India, the world’s third highest emitter12.
scales than reflected in most global models4. Furthermore, ability to A substantial part of this discrepancy (about 3.2 GtCO2 yr−1) can
distinguish anthropogenic from non-anthropogenic effects is lim- be explained by conceptual differences in what is counted in the
ited on the basis of direct observation2 and most estimation meth- anthropogenic forest sink. Beyond this large disparity in global esti-
ods offer few details about where, when and why forest fluxes occur. mates, data and methodological mismatches also exist across proj-
Yet understanding the magnitude, drivers and spatial distribution of ect, subnational and national forest GHG measurement systems,
carbon fluxes across the world’s forests, and how they can be man- leading to complications around integrating smaller-scale activities
aged both to reduce emissions and enhance removals, is increas- into larger national or subnational monitoring programmes13 and
ingly important for climate policy and the various actors developing around the potential international transfer of forest-related emis-
nature-based solutions5. sion reductions versus those achieved as part of a country’s own
Current estimates of terrestrial GHG fluxes vary with respect nationally determined contribution14. In sum, the complexity and
to scope, definitions, assumptions and level of transparency and lack of spatial detail in GHG measurement systems contributes to
completeness. At the global scale, the net annual carbon diox- confusion about the role forests play in climate mitigation targets
ide (CO2) flux from anthropogenic land-use and land-cover and discourages the transformational action and ambition needed
change—driven mainly by tropical deforestation—is estimated in in the forest sector to achieve global climate goals.
IPCC reports1,2 and the Global Carbon Project6 by a bookkeeping Here, we introduce a transparent, independent and spatially
model7,8 or by dynamic global vegetation models6. The remaining explicit global system for monitoring the collective impact of

World Resources Institute, Washington DC, USA. 2Woodwell Climate Research Center, Falmouth, MA, USA. 3Laboratory of Geo-Information Science
1

and Remote Sensing, Wageningen University and Research, Wageningen, the Netherlands. 4Biospheric Sciences Laboratory, NASA Goddard Space
Flight Center, Greenbelt, MD, USA. 5Department of Geographical Sciences, University of Maryland, College Park, MD, USA. 6Center for International
Forestry Research (CIFOR), Bogor, Indonesia. 7Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA. 8Institute of Environment,
University of California, Los Angeles, CA, USA. 9The Sustainability Consortium, University of Arkansas, Fayetteville, AR, USA. 10Present address:
Department of Earth and Environment, Boston University, Boston, MA, USA. 11Present address: Department of Land Resources and Environmental
Sciences, Montana State University, Bozeman, MT, USA. ✉e-mail: [email protected]

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

forest-related climate policies implemented by diverse actors across forests, 12% in tropical primary forests, 10% in plantations, 3.5%
multiple scales. We complement existing global forest carbon flux in young (<20 yr) forest regrowth, 1.3% in mangroves and 0.34% in
estimation approaches of large area vegetation models and aggrega- boreal and temperate intact forest landscapes (Table 1).
tion of national inventories with a third approach that capitalizes
on recent advances in Earth observation. Using recently revised Fluxes for specific localities and drivers of forest change
IPCC guidelines as a methodological framework10,11, we separately Our analysis enables consistent evaluation of forest GHG dynam-
map GHG emissions (sources) and carbon dioxide removals (sinks) ics across scales and in custom geographies beyond national or cli-
from global forest lands at 30-m resolution between 2001 and 2019 mate domain boundaries (Fig. 1). For example, ~27% of the global
(Methods). Areas of forest extent, loss and gain from the Global net forest GHG sink occurred within protected areas16. Forests in
Forest Change product of Hansen et al.15 form the basis of the activ- the Brazilian Amazon were a net carbon source of 0.22 GtCO2e yr−1
ity data. By co-locating activity data with spatially explicit emission between 2001 and 2019, whereas forests across the larger Amazon
and removal factors developed from integrating ground and Earth River basin—encompassing 514 Mha of forests across nine coun-
observation monitoring data on land use and management type, tries—were a net carbon sink of −0.10 GtCO2e yr−1. Although
forest type, forest age class, fire history and biomass and soil carbon smaller in extent than the Amazon, the net sink in forests of Africa’s
stocks, we separately map gross annual carbon removals occurring Congo River basin (298 Mha) was approximately six times stron-
within natural, seminatural and planted forests and gross annual ger (−0.61 GtCO2e yr−1), reflecting nearly identical gross removals
emissions arising from five dominant drivers of forest disturbance. (−1.1 versus −1.2 GtCO2e yr−1) but gross emissions that were half
We then map the difference between gross emissions (+) and gross those of the Amazon basin (0.53 versus 1.1 GtCO2e yr−1).
removals (−) as the net annual forest-related GHG flux, which From overlaying forest GHG flux maps in Fig. 1 with a global
may be positive or negative in an area depending on the balance map of dominant drivers of forest disturbance17, we estimate that
of gross fluxes. Tracking gross emissions and removals separately, commodity-driven deforestation was the largest source of gross
rather than solely the net balance between the two, underscores the forest-related emissions between 2001 and 2019 (2.8 GtCO2e yr−1)
dual role of forests as sources and sinks in the global carbon cycle and occurred primarily in the rainforests of South America and
and facilitates more complete and transparent accounting of the Southeast Asia. Forests in shifting agriculture landscapes, a domi-
individual pathways involved in forest-based mitigation (reducing nant land use in the tropics characterized by cycles of small-scale
emissions and increasing removals). forest clearing of both primary and secondary forests followed by
secondary regrowth, contributed another 2.1 GtCO2e yr−1 to gross
Global distribution of forest emissions and removals emissions and −3.3 GtCO2 yr−1 to gross removals, leading to a
Between 2001 and 2019, deforestation and other satellite-observed net sink in these areas of −1.2 GtCO2e yr−1. Gross emissions from
forest disturbances resulted in global gross GHG emissions of stand-replacing forest fires, occurring primarily in temperate and
8.1 ± 2.5 GtCO2e yr−1 (mean ± s.d.). Carbon dioxide (CO2) was the boreal forests, averaged 0.69 GtCO2e yr−1. Forestry-dominated land-
dominant GHG; methane (CH4) and nitrous oxide (N2O) emis- scapes, comprised of both plantations and natural and seminatural
sions from stand-replacing forest fires and drainage of organic forests, were a net sink of −3.3 GtCO2e yr−1 between 2001 and 2019.
soils in deforested areas accounted for 1.1% of gross emissions This reflects 2.4 GtCO2 yr−1 of gross emissions from harvest offset
(0.088 GtCO2e yr−1). Over the same period, gross carbon removals by by −5.5 GtCO2 yr−1 of gross removals from forest management and
forest ecosystems were −15.6 ± 49 GtCO2e yr−1. Taken together, the regeneration and −0.16 GtCO2e yr−1 of increased carbon storage in
balance of these opposing fluxes (gross emissions and gross remov- harvested wood products.
als) yields a global net GHG forest sink of −7.6 ± 49 GtCO2e yr−1
(Table 1 and Fig. 1). The large uncertainties in global gross remov- A flexible data integration framework
als and net flux are almost entirely due to extremely high uncer- The IPCC Guidelines used as the overarching methodological
tainty in removal factors from the IPCC Guidelines11 applied to old framework in this analysis10,11 provide three tiers of methods, param-
secondary temperate forests outside the United States and Europe eters and data sources for GHG flux estimation, where progression
(Supplementary Table 1). from Tier 1 to Tier 3 generally results in more accurate and precise
Tropical and subtropical forests contributed the most to estimates at the expense of more analytical complexity and larger
global gross forest fluxes, accounting for 78% of gross emis- data requirements. For forests, Tier 3 estimates are characterized by
sions (6.3 ± 2.4 GtCO2e yr−1) and 55% of gross removals the incorporation of repeated, country-specific measurements over
(−8.6 ± 7.6 GtCO2e yr−1) (Table 1). While these forests removed time but the land-use definitions and the spatial scale of data sources
more atmospheric carbon than temperate and boreal forests on a chosen can impact the resulting estimates. Therefore, in addition
gross basis (−8.6 versus −4.4 and −2.5 GtCO2e yr−1, respectively), to estimating uncertainty in GHG estimates within geographies for
tropical and subtropical forests contributed just 30% to the global which information was available to do so (climate domains), we also
net carbon sink; about two-thirds of the global net sink was in tem- conducted sensitivity analyses to demonstrate how estimates change
perate (47%) and boreal (21%) forests, resulting from substantially as data inputs and model assumptions are varied within our spa-
lower gross emissions there than in the subtropics and tropics (0.87 tial data integration framework (Supplementary Information). At
and 0.88 versus 6.3 GtCO2e yr−1, respectively). the global scale, GHG flux estimates were relatively insensitive to
Just six large forested countries (Brazil, Canada, China, changes in model assumptions; estimates for most pixels changed
Democratic Republic of the Congo, Russia and the United States) less than 15% in either direction and sources stayed sources while
accounted for 51% of global gross emissions, 56% of global gross sinks stayed sinks.
removals and 60% of net flux. Forests in nearly all countries were However, estimates were more affected by changes in data
net carbon negative, that is, gross carbon removals from established sources, particularly at local scales. For example, replacing the
and regrowing forests exceeded gross emissions from land-use global 30-m biomass map developed in this study as the basis of
change and other forest disturbances. The main exceptions were emission factors (Extended Data Fig. 1) with a coarser (1-km)
in Indonesia, Malaysia, Cambodia and Laos, where annual gross resolution biomass map produced by Saatchi et al.18 for the trop-
emissions across these countries (1.36 GtCO2e yr−1), including peat ics produced 12% lower gross GHG emissions there than our
drainage and burning (0.14 GtCO2e yr−1), exceeded gross removals original estimate. Replacing the 30-m annual tree cover loss data
(−0.83 GtCO2e yr−1) (Fig. 2). Globally, 72% of gross removals were from Hansen et al.15 in the Brazilian Amazon with annual forest
concentrated in older (>20 yr) secondary natural and seminatural loss data from Brazil’s national forest monitoring system19, which

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles
Table 1 | Forest-related GHG fluxes by climate domain and forest type
Climate Forest type Forest GtCO2e yr−1, 2001–2019
domain extent
Gross Percentage Gross removals Percentage Net GHG flux Percentage
2000
emissions of global of global of global
(Mha)
total total totald
Boreal Primarya 38 0.26 3.2 −0.044 0.28 0.22
Old secondary (>20 yr) 1,030 0.60 7.4 −2.4 15 −1.8
Young secondary (≤20 yr) 22 0.015 0.19 −0.037 0.24 −0.022
Plantations/tree cropsb 0.21 0.000056 0.00070 −0.0027 0.017 −0.0027
Total boreal 1,090 0.88 ± 0.42 11 −2.5 ± 0.96 16 −1.6 ± 1.1 21
Temperate Primarya 2.3 0.036 0.45 −0.0092 0.059 0.027
Old secondary (>20 yr) 560 0.71 8.8 −4.2 27 −3.5
Young secondary (≤20 yr) 16 0.049 0.60 −0.039 0.25 0.0092
Plantations/tree cropsb 12 0.071 0.88 −0.14 0.92 −0.073
Total temperate 590 0.87 ± 0.60 11 −4.4 ± 48 28 −3.6 ± 48 47
Subtropical Primary a
3.6 0.0062 0.076 −0.0058 0.037 0.00035
Old secondary (>20 yr) 270 0.46 5.7 −0.84 5.4 −0.38
Young secondary (≤20 yr) 13 0.11 1.3 −0.067 0.43 0.040
Plantations/tree crops c
54 0.40 5.0 −0.71 4.6 −0.31
Mangroves 0.070 0.000066 0.00082 −0.0040 0.026 −0.0040
Total subtropical 340 1.0 ± 0.59 12 −1.6 ± 0.56 10 −0.65 ± 0.81 8.6
Tropical Primary a
1,010 1.8 22 −1.9 12 −0.12
Old secondary (>20 yr) 880 1.9 23 −3.8 24 −1.9
Young secondary (≤20 yr) 47 0.76 9.5 −0.40 2.5 0.37
Plantations/tree crops c
47 0.89 11 −0.73 4.7 0.16
Mangroves 7.2 0.010 0.12 −0.16 1.0 −0.15
Total tropical 1,990 5.3 ± 2.4 66 −7.0 ± 7.6 45 −1.7 ± 8.0 22
Global Primary 1,060 2.1 26 −2.0 13 0.13
Old secondary (>20 yr) 2,750 3.7 45 −11 72 −7.7
Young secondary (≤20 yr) 99 0.9 12 −0.54 3.5 0.39
Plantations/tree crops 113 1.4 17 −1.6 10 −0.23
Mangroves 8.7 0.012 0.14 −0.20 1.3 −0.19
Total global 4,029 8.1 ± 2.5 100 −16 ± 49 100 −7.6 ± 49 100
Average annual gross GHG emissions, gross GHG removals and net GHG fluxes across global forest lands between 2001 and 2019. Estimates reflect forest ecosystem fluxes only; harvested wood
products are excluded. Uncertainties are expressed as s.d. Large uncertainties in net flux estimates should be interpreted with caution; s.d. are very large relative to the estimates in part because net flux
estimates reflect the sum of negative (removals) and positive (emissions) terms, complicating the combination of their error terms. aThe extent of primary forests was delineated differently for tropical and
extratropical regions (Methods). bFluxes occurring within seminatural managed forests are reported in the relevant secondary forest category (old or young). cFluxes reported in the plantation/tree crop
category include those associated with conversion of natural forests to plantations or tree crops (for example, oil palm) over the 2001–2019 analysis period. dCalculating percentages of net flux by forest
type is complicated by the mixture of sources and sinks among forest types, and is thus omitted.

excludes deforestation events smaller than 6.25 ha, reduced aver- systematic, structured, transparent and verifiable system for explor-
age gross emissions there from 1.1 to 0.74 GtCO2e yr−1. This differ- ing differences in data, assumptions and resulting estimates than
ence arises from increased detection of emissions from small forest what has been available previously.
clearings. Both examples highlight the value of our spatially detailed
approach in capturing more changes and larger fluxes occurring Forest fluxes in the global carbon budget
at small scales where many human-induced forest changes are Our results are not directly comparable to other global estimates
occurring. In the United States, replacing Tier 3 removal factors esti- because other estimates typically reflect all terrestrial fluxes (versus
mated specifically for US forest types and age classes from repeated forests only), report only net fluxes (versus gross and net fluxes),
inventory measurements with generalized Tier 1 defaults from the include only CO2 (versus all relevant GHGs) and make assump-
updated IPCC Guidelines11 led to a 38% stronger net carbon sink tions to partition between anthropogenic and non-anthropogenic
there than the original estimate. (See Supplementary Table 2 and net fluxes2,12. While the spatial, observation-based framework intro-
Extended Data Figs. 2–8 for additional examples.) These analyses duced here permits estimation of fluxes for any forest definition and
quantitatively and spatially demonstrate tradeoffs between globally the inclusion (or exclusion) of any geographic area of interest, it can-
consistent analyses and locally derived values that are difficult to not distinguish between anthropogenic versus non-anthropogenic
aggregate globally and may not be available or comparable across effects or between managed versus unmanaged land until the req-
regions. The flexible spatial data integration framework introduced uisite spatial data become available to differentiate them20. When
here enhances science-policy coordination by providing a more considering only CO2 fluxes to improve comparability with the

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

Gross forest
GHG emissions
MtCO2e yr–1 (2001–2019)
0.21

Gross forest
GHG removals
MtCO2e yr–1 (2001–2019)
0

–0.089

Net forest
GHG flux
MtCO2e yr–1 (2001–2019)
0.17

–0.087

Fig. 1 | Forest-related GHG fluxes (annual average, 2001–2019). a, Gross annual GHG emissions. b, Gross annual GHG removals. c, Net annual GHG flux.
For display purposes, maps have been resampled from the 30-m observation scale to a 0.04° geographic grid. Values in the legend reflect the average
annual GHG flux from all forest dynamics occurring within a grid cell, including emissions from all observed disturbances and removals from both forest
regrowth after disturbance as well as removals occurring in undisturbed forests.

Global Carbon Budget, we estimate a larger net CO2 sink by forest fires accounting for 53, 30 and 17% of the total, respectively21.
ecosystems (−7.8 GtCO2 yr−1) than its estimate of −5.2 GtCO2 yr−1 Adding this (non-spatial) estimate of gross degradation emissions
for all terrestrial fluxes over the same time period6. One potential to our satellite-based gross carbon emission and removal estimates
reason for this difference is that our model underestimates gross occurring within forest ecosystems, as well as −0.16 GtCO2 yr−1
forest-related emissions due to the exclusion of forest disturbances of net removals in harvested wood products, yields a revised net
that go undetected and unquantified in the medium resolution sat- forest-atmosphere CO2 flux of −5.8 GtCO2 yr−1 (Table 2). Taken
ellite observations that underpin our analysis. Gross emissions from together, these estimates of gross removals (−15.6 GtCO2 yr−1)
tropical forest degradation have been estimated as 2.1 GtCO2e yr−1, and gross emissions related to forests (including degradation:
with selective logging, fuelwood harvest and non-stand-replacing 10 GtCO2 yr−1) appear to nearly balance the global carbon budget

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles
Africa Asia

Democratic Republic Indonesia

of the Congo

Madagascar Russia

Ivory Coast Malaysia

Mozambique China

Angola Myanmar

Other countries Other countries

Europe North America/Central America/Caribbean

Sweden United States

Finland Canada

France Mexico

Spain Nicaragua

Poland Guatemala

Other countries Other countries

Oceania South America

Australia Brazil

Papua New Guinea Bolivia

New Zealand Colombia

Solomon Islands Peru

Annual gross removals

Brunei Paraguay
Annual gross emissions
Annual net flux Other countries Other countries

−2,000 −1,000 0 1,000 2,000 −2,000 −1,000 0 1,000 2,000

Average annual flux (MtCO2e yr–1)

Fig. 2 | Gross and net GHG fluxes from forests by region (annual average, 2001–2019). Net forest-related fluxes (grey bars) are shown with their two
component gross fluxes: gross emissions from land-use change and other forest disturbances (purple) and gross removals occurring in undisturbed
forests as well as removals from forest regrowth after disturbance (green). The top five countries per region are ranked high to low on the basis of gross
emissions, with all other countries in the region grouped into ‘other countries’.

(Table 2) but other important fluxes are omitted from our analysis against the potential benefits of publicly accessible, operational and
such as those occurring within grasslands, semi-arid savannas and fit-for-purpose systems that provide enough spatial detail to incen-
shrublands22 (due to the 30% per 5 m of tree cover definition used tivize real, near-term and sustained investment in nature-based cli-
in our analysis), non-stand-replacing fires23, degradation outside mate solutions on the ground. In this study, we combined publicly
the tropics and other terrestrial fluxes not previously included in available data into a global monitoring framework that generates
any global budget to date24. We include Table 2 to highlight how consistent information on forest carbon fluxes cost-effectively over
our gross estimates of forest-related fluxes fit within the context of large spatial scales. However, this approach encounters limitations
the global carbon budget but our research is geared towards high- that should be addressed as research progresses.
lighting forest emission and removal hotspots for policy-relevant First, the global forest change data used as the basis of activ-
applications and stakeholders (Fig. 1), not towards producing ity data in our analysis are spatially detailed but contain temporal
a comprehensive and precise accounting of the full terrestrial inconsistencies. While the forest loss product is updated annually
carbon budget. through 2019, gain has not been updated past 2012 and represents a
cumulative total (2000–2012). Therefore, although gross emissions
Limitations and future improvements can be estimated annually (Extended Data Fig. 9), estimating annual
All forest monitoring systems reflect a balance between data avail- trends in gross removals and net flux is limited by a lack of a con-
ability, scale of applicability, measurement costs, reducing uncertain- sistent time series on forest regrowth. Globally, GHG flux estimates
ties and other constraints. Given the urgency of addressing climate were relatively insensitive to this limitation; we estimate that expan-
change, the time and costs required to develop monitoring systems sion of forest extent observed after 2000 accounted for less than 5%
that reduce uncertainties as far as practicable25 must be balanced of global gross carbon removals, with the vast majority occurring

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

ground and Earth observation data, several biomass and soil carbon
Table 2 | Comparison of results from this study to the Global
maps have been developed that inform spatially explicit emission
Carbon Project, 2001–2018
factors. However, accurate and precise estimation of forest carbon
Global carbon budget, 2001–2018 (GtCO2 yr−1) removal factors requires information derived from long-term forest
Global Carbon Project This study
inventories applied consistently and repeatedly through time across
different forest types and age classes. For many of the world’s for-
Sources ests, this information does not exist28. Many developing countries
Fossil fuel and cement 32.0 Fossil fuel and cement 32.0 have not completed their first forest inventory, let alone repeated
Land-use change (net, 5.3 Forests (gross, all observed 7.9 inventories. Efforts to combine georeferenced plot networks with
anthropogenic)a disturbances)b other spatially explicit data inputs to create maps over large scales
of forest carbon accumulation rates over time, similar to what has
Forests (gross, unobserved 2.1
emission sources)c
been done to develop biomass density maps at a single point in
time, have begun but are still in their infancy29. We therefore applied
Total sources 37.3 42.0 removal factors using a stratification approach, where each forest
Sinks pixel is assigned a removal factor on the basis of its geographic
Atmosphere 16.9 Atmosphere 16.9 region, forest type and age class (Methods). Removal factors reflect
both ecological forest dynamics (tree growth, mortality and recruit-
Ocean 8.7 Ocean 8.7
ment through natural regeneration) and indirect effects (long-term
Terrestrial (net, 10.5 Forests (gross, all forests) e
15.6 increases in atmospheric CO2 concentrations and temperature,
non-anthropogenic)d nutrient fertilization). Going forward, new satellite missions such
Harvested wood products 0.16 as GEDI, ICESAT-2 and BIOMASS will provide repeated measure-
Total sinks 36.1 41.4 ments of forest height and biomass over time that should improve
understanding of spatial variation in rates of carbon removal across
Land (net, all land) −5.2 Forests (net, all forests)f −5.8
heterogeneous forest landscapes.
Budget imbalance g
1.2 0.6 The global forest carbon monitoring framework introduced here,
Estimates from the Global Carbon Project (GCP)6 and this study are not directly comparable and the main improvements identified above, allow for efficient pri-
due to differences in scope (all land versus forests, respectively), data, methodologies and oritization and evaluation of how data updates and improvements
reporting structure. In GCP reporting, land-use change emissions (sources) reflect the net
balance between anthropogenic emissions (+) and removals (–), thus the net emission estimate
influence GHG flux estimates and their uncertainties. As satellite-
is lower than gross emissions reported in this study. Similarly, gross removals reported in this and ground-based forest monitoring improve, so too will the associ-
study reflect removals across all forest lands, including removals implicit (but unreported) in the ated forest GHG flux estimates.
net land-use change estimate of GCP. aEstimates only net direct anthropogenic effects, including
deforestation, afforestation/reforestation and wood harvest. Gross fluxes higher but not reported.
b
Gross emissions from all forest disturbances (anthropogenic and non-anthropogenic) observed Conclusions
from Landsat data. Estimate includes CO2 only for comparability with GCP; non-CO2 emissions Our analysis reinforces the need to reduce gross emissions from
are 0.086 GtCO2e yr−1. cGross emissions from forest degradation in 74 developing countries
tropical deforestation as a climate change mitigation strategy,
covering 2.2 billion hectares of forest, from Pearson et al.21. dIn IPCC’s Fifth Assessment Report1,
calculated as the residual of all other terms in the carbon budget. eGross removals from all forest while also highlighting the substantial but often underappreciated
processes (direct, indirect and natural). fCalculated as the net balance between gross forest contribution of intact primary and older secondary forests to car-
ecosystem emissions and removals (7.9 + 2.1–15.6 GtCO2 yr−1) plus an additional net removal of bon dioxide removals. Quantifying gross emissions and removals
−0.16 GtCO2 yr−1 in harvested wood products. gBudget imbalance is the difference between total
sources and total sinks. separately and consistently across all forest lands—and producing
maps in addition to tabular statistics—improves transparency in
the accounting of factors and geographies contributing to the global
net forest GHG flux. It also provides a framework to integrate new
instead in forests established before 2000. However, accurate moni- and improved data sources over time. Governments interested in
toring of the timing of recent regrowth becomes more important in spatially prioritizing implementation and tracking of national and
local contexts where rapid forest loss/gain dynamics are occurring, subnational forest mitigation targets can increasingly make use of
such as in plantations with short rotation cycles and other dynamic such data. Non-government actors, such as companies aiming to
areas dominated by intensive forestry or short-fallow shifting cul- reduce emissions from deforestation associated with commodity
tivation systems (Extended Data Fig. 5). Temporal inconsistencies supply chains and emerging market mechanisms considering the
are also present within the global loss product; one algorithm cov- inclusion of forests for carbon offset programs, could benefit from
ers years 2001–2010 and another covers 2011–2019, with later years a globally consistent and spatially explicit forest monitoring sys-
of loss likely to be more sensitive to changes related to small-scale tem developed using the same internationally accepted methods as
agriculture, fires and other forms of forest degradation. For these national governments use but based on independent observations
reasons, we report only long-term averages and not annual trends in and with GHG estimates that can be linked to individual actions
forest GHG fluxes. A forthcoming ‘version 2’ global tree cover loss and generated at scales relevant to diverse climate-related policies,
product and an improved global gain product, already piloted for programmes and stakeholders.
the lower Mekong region of Southeast Asia26, will improve temporal The goals of the Paris Agreement—primarily, net zero anthro-
consistency. Incorporating these improvements into the forest GHG pogenic emissions in the second half of this century—create an
flux model will more accurately capture interannual variability in imperative to track forest-related emissions and removals trans-
emissions and removals over time and will thus provide a consistent parently and at scales that link more closely to mitigation activities
basis for more temporally detailed monitoring of the long-term net on the ground. As the capacity of national governments to collect,
impact of forests on atmospheric GHGs27. process and analyse data continues to improve, the global forest
Second, information is currently lacking to develop globally con- carbon monitoring framework introduced here can help to enhance
sistent and spatially detailed maps of forest carbon removals. In our transparency, inform forest-related climate policy and implemen-
analysis, uncertainty in gross removals is substantially higher than tation initiatives, underpin independent technical assessments,
uncertainty in gross emissions, driven primarily by high uncer- reconcile differences between national reports and scientific
tainty in removal factors for established forests in temperate regions studies, and provide a more consistent and comparable basis for
(Table 1 and Supplementary Table 1). Through the integration of tracking progress at local scales and for assessing atmospheric

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles
impacts of global forest change under the Paris Agreement’s forth- 13. Lee, D., Llopis, P., Waterworth, R., Roberts, G. & Pearson, T. Approaches to
coming Global Stocktake30. REDD+ Nesting: Lessons Learned from Country Experiences (World Bank, 2018).
14. Streck, C. et al. Options for Enhancing REDD+ Collaboration in the Context of
Article 6 of the Paris Agreement (Meridian Institute, 2017).
Online content 15. Hansen, M. C. et al. High-resolution global maps of 21st-century forest cover
Any methods, additional references, Nature Research report- change. Science 342, 850–853 (2013).
ing summaries, source data, extended data, supplementary infor- 16. World Database on Protected Areas User Manual (UNEP, 2016); https://www.
protectedplanet.net/en/resources/wdpa-manual
mation, acknowledgements, peer review information; details of
17. Curtis, P. G., Slay, C. M., Harris, N. L., Tyukavina, A. & Hansen, M. C.
author contributions and competing interests; and statements of Classifying drivers of global forest loss. Science 361, 1108–1111 (2018).
data and code availability are available at https://doi.org/10.1038/ 18. Saatchi, S. S. et al. Benchmark map of forest carbon stocks in tropical regions
s41558-020-00976-6. across three continents. Proc. Natl Acad. Sci. USA 108, 9899–9904 (2011).
19. PRODES Deforestation (INPE, 2019); http://www.obt.inpe.br/OBT/assuntos/
Received: 15 May 2020; Accepted: 3 December 2020; programas/amazonia/prodes
20. Ogle, S. M.et al. Delineating managed land for reporting national greenhouse
Published: xx xx xxxx
gas emissions and removals to the United Nations framework convention on
climate change. Carbon Balance Manag. 13, 9 (2018).
References 21. Pearson, T. R., Brown, S., Murray, L. & Sidman, G. Greenhouse gas emissions
1. IPCC Climate Change 2014: Synthesis Report (eds Core Writing Team, from tropical forest degradation: an underestimated source. Carbon Balance
Pachauri, R. K. & Meyer L. A.) (IPCC, 2014). Manag. 12, 3 (2017).
2. IPCC Special Report on Climate Change, Desertification, Land Degradation, 22. Ahlström, A. et al. The dominant role of semi-arid ecosystems in the trend
Sustainable Land Management, Food Security, and Greenhouse Gas Fluxes in and variability of the land CO2 sink. Science 348, 895–899 (2015).
Terrestrial Ecosystems (IPCC, 2019). 23. Van Der Werf, G. R. et al. Global fire emissions estimates during 1997–2016.
3. Adoption of the Paris Agreement FCCC/CP/2015/10/Add.1 (UNFCCC, 2015). Earth Syst. Sci. Data 9, 697–720 (2017).
4. Klein Goldewijk, K., Beusen, A., Doelman, J. & Stehfest, E. New 24. Kirschbaum, M. U., Zeng, G., Ximenes, F., Giltrap, D. L. & Zeldis, J. R.
anthropogenic land use estimates for the Holocene: HYDE 3.2. Earth Syst. Towards a more complete quantification of the global carbon cycle.
Sci. Data 9, 927–953 (2017). Biogeosciences 16, 831–846 (2019).
5. Griscom, B. W. et al. National mitigation potential from natural climate 25. Global Forest Observations Initiative. Integration of Remote-sensing and
solutions in the tropics. Philos. Trans. R. Soc. B 375, 20190126 (2020). Ground-based Observations for Estimation of Emissions and Removals of
6. Friedlingstein, P. et al. Global carbon budget 2019. Earth Syst. Sci. Data 11, Greenhouse Gases in Forests 2nd edn (FAO, 2016).
1783–1838 (2019). 26. Potapov, P. et al. Annual continuous fields of woody vegetation structure in
7. Houghton, R. A. & Nassikas, A. A. Global and regional fluxes of carbon from the Lower Mekong region from 2000–2017 Landsat time-series. Remote Sens.
land use and land cover change 1850–2015. Glob. Biogeochem. Cycles 31, Environ. 232, 111278 (2019).
456–472 (2017). 27. Federici, S., Lee, D. & Herold, M. Forest Mitigation: A Permanent
8. Hansis, E., Davis, S. J. & Pongratz, J. Relevance of methodological choices for Contribution to the Paris Agreement? (Climate and Land Use Alliance, 2017).
accounting of land use change carbon fluxes. Glob. Biogeochem. Cycles 29, 28. Romijn, E. et al. Assessing change in national forest monitoring capacities of
1230–1246 (2015). 99 tropical countries. Ecol. Manag. 352, 109–123 (2015).
9. Pan, Y. et al. A large and persistent carbon sink in the world’s forests. Science 29. Cook-Patton, S. Mapping potential carbon capture from global natural forest
333, 988–993 (2011). regrowth. Nature 585, 545–550 (2020).
10. IPCC. 2006 IPCC Guidelines for National Greenhouse Gas Inventories Vol. 4 30. The Global Stocktake (UNFCCC, 2015); https://unfccc.int/topics/science/
(eds Eggleston, S. et al.) (IGES, 2006). workstreams/global-stocktake-referred-to-in-article-14-of-the-paris-agreement
11. IPCC. 2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
Gas Inventories Vol. 4 (eds Buendia, E. C. et al.) (IPCC, 2019).
published maps and institutional affiliations.
12. Grassi, G. et al. Reconciling global-model estimates and country reporting of
anthropogenic forest CO2 sinks. Nat. Clim. Change 8, 914–920 (2018). © The Author(s), under exclusive licence to Springer Nature Limited 2021

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

Methods to match the common 30-m resolution of our analysis. For mangrove forests, we
Study design and scope. We mapped gross and net GHG emissions by sources and used a 30-m soil carbon map developed specifically for mangroves43. We delineated
removals by sinks from global forest lands by synthesizing information collected locations of organic (peat) soils using maps summarized in Supplementary Table 3.
from more than 637,000 ground plots, 707,561 waveform lidar observations We used these five forest carbon pool maps as the basis for estimating emission
and other satellite data into a spatial forest carbon monitoring framework. The factors associated with various forest disturbances (see below).
analysis covers 2001 to 2019 but can be extended to include later years as data are
updated. To the extent possible, we adhered to IPCC Guidelines developed for Activity data. Activity data were defined using the global forest change product of
the agriculture, forestry and other land use (AFOLU) sector10,11. In the context of Hansen et al.15 with loss updated annually on Global Forest Watch. In the model,
IPCC land-use categories, our analysis covers only forest-related transitions (forest all pixels defined as forest were classified into one of four categories: (1) loss only;
to non-forest, non-forest to forest and forest remaining forest). We applied the (2) gain only; (3) both loss and gain; or (4) no change over the period 2001–2019.
IPCC gain-loss method (versus the stock-difference method10), in which forest Loss is defined by Hansen et al. as a stand-replacement disturbance and includes
carbon (C) stocks in five ecosystem pools were estimated for a base year (2000) all disturbances (natural and anthropogenic) observable in Landsat imagery. Gain
after which changes in C stocks were estimated by considering both annual C is defined as a non-forest to forest change, which includes tree cover gain observed
losses from land-use change and disturbance (conventionally represented by a after harvest and other disturbance. The loss product is annual, while the gain
+ sign) as well as annual C gains from forest regrowth (represented by a – sign). product represents a cumulative total (2000–2012). Loss and gain can co-occur
We included harvested wood products as a sixth (human-created) carbon pool. on pixels undergoing forest management or other forms of disturbance and
We also included methane (CH4) and nitrous oxide (N2O) emissions from regrowth. Lack of annually updated gain data is addressed through the sensitivity
stand-replacement forest fires and drainage of organic soils associated with a loss analysis (Extended Data Fig. 5). Due to a lack of information about tree cover gain
of tree cover. We summarized GHG fluxes across all relevant gases and reported in after 2012, we assumed no additional areas of gain from 2012 to 2019. Areas of
units of CO2 equivalents (CO2e) using 100-yr Global Warming Potentials (without no change reflect forest areas established before 2000 that showed no observable
climate feedbacks) from the IPCC Fifth Assessment Report1. disturbance in Landsat imagery between 2000 and 2019.
We set all data inputs to a common resolution of 0.00025° × 0.00025° to
match the resolution of Landsat-based tree cover change data of Hansen et al.15. Emission factors. We assigned emission factors to tree cover loss pixels following
Gross emissions and removals were modelled at this common resolution across an IPCC land-use classification framework, on the basis of whether each pixel
approximately 90 billion individual pixels of global forest cover (defined below). maintained its land use or was converted to a new use over the analysis period.
We resampled all input layers to this resolution so that outputs can be flexibly Since forest may remain in the same use despite a temporary loss of tree cover, we
aggregated to larger scales. Extended Data Fig. 10 summarizes the overall used the global 10-km map of Curtis et al.17 (updated through 2019) to attribute
conceptual approach and Supplementary Table 3 provides a list of data inputs. tree cover loss to one of five dominant drivers; these influence the C pools affected
(Supplementary Table 5) and thus the emission factors assigned to each individual
Forest definition and extent. Initially, we defined forest extent in the year 2000 loss pixel. Supplementary Table 6 summarizes emission factors by forest type
similarly to Hansen et al.15, that is, any 30-m Landsat pixel that met a tree canopy within each climate domain.
threshold of at least 30% with trees taller than 5 m in height. This initial definition
included natural and seminatural forests, plantations and agricultural tree crops Commodity-driven deforestation and shifting agriculture. The initial change in C
such as oil palm and agroforestry systems where minimum height and cover stocks was estimated as a full loss of C in aboveground, belowground, dead wood
thresholds were met. On the basis of available data, we made four modifications to and litter pools. In addition to CO2 emissions resulting from a loss of C stocks, we
the original tree cover map to refine our global map of forest extent: used IPCC equation 2.27 (ref. 10) and a 1-km global burned area map44 to calculate
CH4 and N2O emissions in loss pixels that overlapped with areas that burned the
1. We included pixels of tree cover gain since 2000 in addition to tree cover same year or the year before (to account for lag effects between fire occurrence
already present in the year 2000. and observed tree cover loss). For deforestation on mineral soils, soil C loss was
2. We included only tree cover pixels that also had a corresponding value in estimated using IPCC equation 2.25 (ref. 10); default soil stock change factors vary
the aboveground biomass density map (0.031% of tree cover pixels lacked a by ecological zone and were assigned spatially using ecozone boundaries45. Per
biomass value). IPCC guidelines, 1/20th of the total soil C stock change was apportioned annually
3. We excluded all areas of tree cover falling within oil palm plantation bounda- from the year of loss through the last year of the analysis period (2019) but
ries mapped for the year 2000 in Indonesia and Malaysia31–34. assigned to the year of observed tree cover loss. Due to lack of information in the
4. We replaced tree cover extent from Hansen et al.15 with mangrove forest driver attribution map17 about the specific land use established after forest clearing,
extent using data from Giri et al.35; in areas of geographic overlap, mangroves we assumed for the purposes of soil emission accounting that all deforested land
had priority. on mineral soils for commodity-driven deforestation was converted to annual
cropland with full tillage and medium inputs. A different factor was used to
Forest aboveground live biomass density in 2000. We created a year 2000 map estimate loss of soil C on mineral soils (Table 5.10 in the IPCC Guidelines10) in
of aboveground live biomass density (AGB, in Mg ha−1) at 30-m resolution by areas of shifting agriculture, which were assumed to represent transient land-use
combining two maps: one developed specifically for mangroves36 and the other conversions to cropland under shortened fallow, where vegetation recovery is
developed to cover all woody vegetation globally (Supplementary Data 1). In areas not attained before re-clearing. Soil emissions were not estimated for areas of
of geographic overlap, the mangrove biomass map had priority. The basic approach loss on mineral soils that overlapped with forest and wood fibre plantations,
is the same as that used to map tropical biomass at 500-m (ref. 37) and 30-m (ref. 38) even if they fell within the broader commodity-driven deforestation or shifting
resolution; published height–biomass equations were applied to estimate biomass agriculture classes, consistent with the assumption that loss of tree cover within
over specific regions and forest types around the world (Extended Data Fig. 1a). tree plantations follows the forestry assumptions listed in Supplementary Table 5
These equations, developed by linking observations from airborne or spaceborne (see emissions from Forestry below). For loss on organic soils that overlapped with
lidar to 20,347 ground-measured biomass plots, were applied to estimate tropical plantations and tree crops planted since 2000, GHG emissions associated
aboveground biomass density from spaceborne lidar observations across 707,561 with drainage were estimated using CO2 and CH4 emission factors provided in the
locations globally. To create a continuous biomass map (Extended Data Fig. 1b), IPCC Wetlands Supplement40. Like emissions from mineral soils, emissions from
separate random forest models were trained for each of six biogeographic realms peat drainage were assumed to continue in each year after loss up through the last
using predictor variables of Landsat imagery (bands 3, 4, 5 and 7), normalized year of the analysis period (2019) but were assigned to the year of observed tree
difference vegetation index (NDVI), normalized difference infrared index (NDII), cover loss. Emissions (CO2, CH4 and N2O) from peat burning were also calculated
mean percentage tree cover, mean elevation, mean slope and monthly mean on the basis of methods provided in the IPCC Wetlands Supplement40 where a loss
precipitation, temperature and bioclimatic data. Additional details are provided in pixel overlapped with areas burned the same year, or the year before, the loss event
Supplementary Data 1. (on the basis of global burned area data).

Forest ecosystem carbon pools in 2000. From the 30-m global AGB map, we Urbanization. The same assumptions and calculations were used for calculating
mapped belowground live biomass density (BGB) using a forest root-to-shoot gross emissions from urbanization as for commodity-driven deforestation and
ratio39 with mangrove-specific ratios based on defaults provided in Table 4.5 of shifting agriculture, except a different factor was used to estimate the loss of soil C
the 2013 IPCC Wetlands Supplement40. AGB and BGB values were converted to on mineral soils. We assumed that forest land converted to settlement was paved
C density values using a biomass-to-carbon ratio of 0.45 for mangroves40 and 0.47 over and applied the IPCC default assumption11 that 20% of the soil C relative to
for all other forest types10,11. From the final 30-m AGB map we estimated dead the previous land use was lost as a result of disturbance, removal or relocation.
wood and litter biomass densities per pixel as constant fractions of AGB using a
lookup table based on global ecological zone, elevation and precipitation regime41 Forestry. Emission factors for loss attributed to forestry were estimated as the loss
(Supplementary Table 4). Dead wood and litter biomass densities were converted of C in live biomass only, following assumptions outlined in Supplementary Table 5
to C densities using IPCC conversion factors10. that there is no net change to the dead organic matter or soil C pools in the case of
Soil organic carbon density in the top 30 cm of mineral soils was mapped using mineral soils. Emissions from peat drainage and burning associated with forestry
SoilGrids250 (v.2.0)42 after resampling from its original spatial resolution of 250 m activities, as well as non-CO2 emissions in the case of forest fires, were included in

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles
the same way as for deforestation and shifting agriculture above. Emission factors criterion applied but represents the best available information by which to spatially
for loss pixels within the ‘zero or minor loss’ category of the driver attribution map delineate primary from old secondary forests in boreal and temperate regions.
also followed assumptions for forestry (Supplementary Table 5).
Old secondary forests. We assigned removal factors from IPCC Table 4.9 (>20 yr)
Wildfire. Within 10-km grid cells of the drivers map labelled wildfire, wildfire to all forest areas that fell outside the types identified above. Given no observed
emission factors were applied only for 30-m pixels where loss occurred in the year disturbance occurred in these areas since the year 2000, we assumed they were
of, or year after, a fire event in the 1-km burned area map. In these cases, we used secondary natural forests at least 20 years old.
IPCC equation 2.27 (ref. 10) to estimate both CO2 and non-CO2 emissions from
forest fire. The AGB map determined the mass of fuel available for combustion and Harvested wood products. We used statistics reported in FAOSTAT and methods
a lookup table (Table 2.6 of the IPCC Guidelines10) provided default combustion outlined in the 2019 Refinement11 to estimate emissions and/or removals arising
and emission factors that were applied on the basis of forest type (primary versus from harvested wood products. Losses of harvested wood products in use
secondary). For boreal and temperate forests, combustion factors were applied on were assumed to result in CO2 emissions to the atmosphere, with no explicit
the basis of the assumption of a land-clearing fire, given that forest loss is defined representation of the subsequent retention of disposed wood in solid waste
in Hansen et al.15 as a stand-replacement disturbance. In cases where organic soils disposal sites (SWDS) and eventual CO2 emissions from SWDS. Calculations rely
overlapped with burned areas, emissions from peat burning (CO2, CH4 and N2O) on statistics reported by countries on production, import and export volumes
were estimated following guidance in the IPCC Wetlands Supplement40. Forestry for three aggregate semifinished wood product commodity classes: sawnwood,
emission factors, rather than wildfire factors, were applied where loss did not wood-based panels and paper and paperboard.
overlap with a fire event in the 1-km burned area map.
Uncertainty analysis. We estimated uncertainty in GHG flux estimates globally
Removal factors. We developed removal factors spatially by linking information and at the scale of climate domains by combining uncertainties in the activity
about each pixel’s geographic region, ecological zone, forest type and age data and emission/removal factors following a Taylor series statistical approach
class to corresponding growth rates on the basis of best available information. as in Roman-Cuesta et al.55 and Carter et al.56. This approach underlies the
Supplementary Table 6 summarizes removal factors by forest type in each climate IPCC Approach 1 (simple error propagation)10 and produces similar results but
domain. In areas of geographic overlap, the priority of assigning removal factors to reflect exact calculations of variances and s.d., whereas IPCC Approach 1 is an
a given pixel reflects the order of data sources listed below. Removal factors include approximated approach that yields 95% confidence intervals.
accumulation in live biomass only and reflect the net increase, accounting for both Uncertainties of all major components of the flux model were included
productivity and mortality. We assumed no change to the dead organic matter and (activity data, affected C pools of the emission/removal factors, combustion and
soil organic carbon pools, consistent with the IPCC Tier 1 assumption of no net emission factor uncertainties for fire-related emissions). Errors were assumed to
change to non-biomass pools in forest land remaining forest land. The number be statistically independent (uncorrelated), normally distributed and without bias.
of years of carbon accumulation was assigned as 19 yr for undisturbed forest, 6 yr Supplementary Table 1 shows the contribution of each uncertainty component
for areas of new tree cover gain and one less than the year in which tree cover loss for domain and global gross emissions, removals and net flux, reported as the
occurred for loss-only forest. percentage reduction in output variances as each of the uncertainty components
were assumed to have no variance. Variance of the net GHG flux was reduced
Mangroves. We applied mangrove-specific growth rates and root-to-shoot ratios the most when removing variance of the removal factor for temperate forests
from IPCC Tables 4.4 and 4.5 of the Wetlands Supplement40, respectively. older than 20 yr. Variances are likely to be lower when estimated across smaller
geographic regions. Estimation of uncertainty is currently limited to the global
Europe. We assigned removal factors spatially according to a map of dominant and biome scales based on available data for estimating uncertainty in the
tree species developed from 260,000 national inventory plot locations46. For each activity data.
species, we estimated mean annual increment (MAI) values from Table 4.11 of
the updated IPCC Guidelines11, FAO Planted Forest Assessment47 and national Reporting Summary. Further information on research design is available in the
inventories48 (Supplementary Table 3). These were converted to aboveground Nature Research Reporting Summary linked to this article.
biomass growth rates using species-specific biomass conversion and expansion
factors and belowground biomass increment was added on the basis of a Data availability
root-to-shoot ratio39. Geospatial data generated from the current study are publicly available on Global
Forest Watch’s Open Data Portal (http://data.globalforestwatch.org/) and from the
Plantations and tree crops. Outside Europe, we assigned removal factors for corresponding author upon request. Summary geospatial statistics are available
plantations and tree crops using a variety of published data sources49. For from the corresponding author upon request. All data inputs used in the current
common plantation species, we used MAI and biomass conversion and expansion study are publicly available or were obtained by the corresponding author.
factors summarized in the updated IPCC Guidelines11 to estimate aboveground
biomass increment and added belowground biomass increment on the basis of a
root-to-shoot ratio39. Rates in plantations were assigned on the basis of mapped Code availability
species when known or, when unknown, the most common mix of plantation To ensure full reproducibility and transparency of our research, we provide all of
species grown in the region. Removal factors for tree crops such as oil palm the scripts used in our analysis. Codes used for this study are permanently and
and rubber as well as various types of agroforestry systems were estimated for publicly available on GitHub (https://github.com/wri/carbon-budget).
areas mapped as such on the basis of regionally specific values derived from the
published literature and from Tables 5.1 and 5.3 of the updated IPCC Guidelines11. References
All removal factors used for plantations and tree crops, along with data sources and 31. Austin, K. et al. Shifting patterns of oil palm driven deforestation in
assumptions applied, are provided in the companion spatial attribute file associated Indonesia and implications for zero-deforestation commitments. Land Use
with the global compilation of planted tree maps used in this analysis49. Policy 69, 41–48 (2017).
32. Gaveau, D. L. et al. Four decades of forest persistence, clearance and logging
United States. We developed removal factors for three age classes (0–20, 20–100 on Borneo. PLoS ONE 9, e101654 (2014).
and >100 yr) for forest types across 11 geographic regions using methods 33. Miettinen, J., Shi, C. & Liew, S. C. Land cover distribution in the peatlands of
broadly similar to those of Smith et al.50, except that we included more forest Peninsular Malaysia, Sumatra and Borneo in 2015 with changes since 1990.
types in each region, as well as more recent and comprehensive data from the US Glob. Ecol. Conserv. 6, 67–78 (2016).
Forest Inventory and Analysis database. Removal factors were developed from 34. Gunarso, P., Hartoyo, M., Agus, F. & Killeen, T. in Reports from the Technical
approximately 130,000 inventory plot locations. Pixels were assigned removal Panels of the 2nd Greenhouse Gas Working Group of the Roundtable on
factors on the basis of dominant forest type51, age class52 and geographic Sustainable Palm Oil (eds Killeen, T. J. & Goon, J.) 29–64 (RSPO, 2013).
inventory region. 35. Giri, C. et al. Status and distribution of mangrove forests of the
world using earth observation satellite data. Glob. Ecol. Biogeogr. 20,
Young secondary forests. Outside the United States and Europe, areas of tree cover 154–159 (2011).
gain that fell outside boundaries of mangroves35 and planted trees49 were assumed 36. Simard, M. et al. Mangrove canopy height globally related to precipitation,
to be secondary natural forest regrowth <20 years old. We assigned natural forest temperature and cyclone frequency. Nat. Geosci. 12, 40–45 (2019).
regrowth removal factors to these areas using the 1-km map of Cook-Patton et al.29. 37. Baccini, A. et al. Estimated carbon dioxide emissions from tropical
deforestation improved by carbon-density maps. Nat. Clim. Change 2,
Primary forests. We used removal factors by ecological zone and continent from 182–185 (2012).
IPCC Table 4.9 of the 2019 IPCC Refinement11 and assigned them spatially 38. Zarin, D. J. et al. Can carbon emissions from tropical deforestation drop by
between 30° N and 30° S within a tropical primary humid forest map53. Outside 50% in 5 years? Glob. Change Biol. 22, 1336–1347 (2016).
30° N and 30° S, we used a map of intact forest landscapes54 as a proxy for primary 39. Mokany, K., Raison, R. J. & Prokushkin, A. S. Critical analysis of root: shoot
forests, which is likely to be highly conservative due to the relatively large extent ratios in terrestrial biomes. Glob. Change Biol. 12, 84–96 (2006).

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change
40. IPCC Supplement to the 2006 IPCC Guidelines for National Greenhouse Gas Acknowledgements
Inventories: Wetlands (eds Hiraishi, T. et al.) (IPCC, 2014). We thank S. Gibbes for her work on preliminary model development and T. Maschler
41. Methodological Tool: Estimation of Carbon Stocks and Change in Carbon for his contributions to workflows enabling efficient data processing and generation
Stocks in Dead Wood and Litter in A/R CDM Project Activities (UNFCCC, of summary statistics. Support for this research was funded in part by the Norwegian
2013); https://cdm.unfccc.int/methodologies/ARmethodologies/tools/ Ministry of Foreign Affairs (18/2721 Global Forest Watch Achieving Sustainability and
ar-am-tool-12-v3.0.pdf Scaling Impact), the UK Department for International Development (DFID FGMC
42. Hengl, T. et al. SoilGrids250m: global gridded soil information based on grant no. FGMC2018-21-WRI) and the US Agency for International Development
machine learning. PLoS ONE 12, e0169748 (2017). (cooperative agreement no. 7200AA19CA00027 Global Forest Watch 3.0) in support
43. Sanderman, J. et al. A global map of mangrove forest soil carbon at 30 m of the Global Forest Watch Partnership convened by the World Resources Institute, by
spatial resolution. Environ. Res. Lett. 13, 055002 (2018). National Aeronautics and Space Administration Earth Science Division NNH12ZDA001
44. Giglio, L., Boschetti, L., Roy, D. P., Humber, M. L. & Justice, C. O. The NICESAT2: studies with ICESAT and CryoSat-2 grant no. 12-ICESAT212-0022 to the
Collection 6 MODIS burned area mapping algorithm and product. Remote Woods Hole Research Center and by the NASA Carbon Monitoring System Program
Sens. Environ. 217, 72–85 (2018). Project ‘Estimating Total Ecosystem Carbon in Blue Carbon and Tropical Peatland
45. Global Ecological Zones for FAO Forest Reporting: 2010 Update Ecosystems’ (16- 30 CMS16-0073) to NASA Goddard. The contribution of M.H., S.deB.
(FAO, 2012). and D.R.S. was supported by CIFOR’s global comparative study on REDD+ (funded
46. Brus, D. et al. Statistical mapping of tree species over Europe. Eur. J. Res. 131, by NORAD), the European Space Agency CCI-Biomass project and the European
145–157 (2012). Commission Horizon 2020 projects VERIFY (grant no. 776810) and REDD-Copernicus
47. Del Lungo, A., Ball, J. & Carle, J. Global Planted Forests Thematic Study: (grant no. 821880). Data used in part of this publication were made possible, in part, by
Results and Analysis (FAO, 2006); http://www.fao.org/forestry/12139-03441d0 an agreement from the United States Department of Agriculture’s Forest Service. This
93f070ea7d7c4e3ec3f306507.pdf publication may not necessarily express the views or opinions of the Forest Service.
48. Portugal National Greenhouse Gas Inventory submitted to the UNFCCC,
1990–2018 (UNFCCC, 2020).
49. Harris, N. L., Goldman, E. D. & Gibbes, S. Spatial Database on Planted Trees Author contributions
Version 1.0 https://www.wri.org/publication/spatialdatabase-planted-trees N.L.H. was involved in conceptualization, data curation, formal analysis, funding
(WRI, 2019). acquisition, investigation, methodology, project administration, visualization and
50. Smith, J. E., Heath, L. S., Skog, K. E. & Birdsey, R. A. Methods for Calculating writing. D.A.G. contributed to data curation, formal analysis, investigation, methodology,
Forest Ecosystem and Harvested Carbon with Standard Estimates for Forest software, validation, visualization and writing. A.B., R.A.B., R.R.C., M.F., L.F., M.C.H.,
Types of the United States General Technical Report (USDA, Forest Service, R.A.H., P.V.P., C.M.S., D.R.S., S.S.S. and S.A.T. contributed to data curation, formal
2006); https://doi.org/10.2737/NE-GTR-343 analysis, methodology and writing. M.H. contributed to data curation, visualization and
51. Ruefenacht, B. et al. Conterminous US and Alaska forest type mapping using writing. S.deB. and A.T. contributed to formal analysis and methodology.
forest inventory and analysis data. Photogramm. Eng. Remote Sensing 74,
1379–1388 (2008). Competing interests
52. Pan, Y. et al. Age structure and disturbance legacy of North American forests. The authors declare no competing interests.
Biogeosciences 8, 715–732 (2011) .
53. Turubanova, S., Potapov, P. V., Tyukavina, A. & Hansen, M. C. Ongoing
primary forest loss in Brazil, Democratic Republic of the Congo, and Additional information
Indonesia. Environ. Res. Lett. 13, 074028 (2018). Extended data is available for this paper at https://doi.org/10.1038/s41558-020-00976-6.
54. Potapov, P. et al. The last frontiers of wilderness: tracking loss of intact forest Supplementary information is available for this paper at https://doi.org/10.1038/
landscapes from 2000 to 2013. Sci. Adv. 3, e1600821 (2017). s41558-020-00976-6.
55. Roman-Cuesta, R. M. et al. Hotspots of gross emissions from the land use Correspondence and requests for materials should be addressed to N.L.H.
sector: patterns, uncertainties, and leading emission sources for the period
2000–2005 in the tropics. Biogeosciences 13, 4253–4269 (2016). Peer review information Nature Climate Change thanks Gert-Jan Nabuurs, Seth Spawn
56. Carter, S. et al. Agriculture-driven deforestation in the tropics from and the other, anonymous, reviewer(s) for their contribution to the peer review of this
1990–2015: emissions, trends and uncertainties. Environ. Res. Lett. 13, work.
014002 (2017). Reprints and permissions information is available at www.nature.com/reprints.

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles

Extended Data Fig. 1 | Aboveground live woody biomass density in the year 2000. a, Subsets of ecoregions over which different height–biomass
equations were applied. Patterned shading indicates equations that were only applied to conifer GLAS shots within the specified ecoregion. b, Global 30-m
map of aboveground live woody biomass density in the year 2000.

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

Extended Data Fig. 2 | Results of sensitivity analysis when the source of tree cover loss data used in the forest GHG flux model is changed from the
30-m tree cover loss product of Hansen et al.15 in the standard model to PRODES, Brazil’s 250-m forest loss monitoring product for the Brazilian
Amazon19, in the alternative model. Top panel: Percent change in net GHG flux between standard model and sensitivity analysis model; Bottom panel:
Delineation of areas that remain a net GHG source or sink in the sensitivity analysis model vs. those that switch from being a net source or sink to a net
sink or source as a result of the changes applied. For display purposes, maps have been resampled from the 30-m observation scale to a 0.04-degree
geographic grid.

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles

Extended Data Fig. 3 | Results of sensitivity analysis when the source of biomass data used in the forest GHG flux model is changed from a 30-m global
AGB map in the standard model to a 1-km tropical AGB map in the alternative model. Top panel: Percent change in net GHG flux between standard model
and sensitivity analysis model; Bottom panel: Delineation of areas that remain a net GHG source or sink in the sensitivity analysis model vs. those that
switch from being a net source or sink to a net sink or source as a result of the changes applied. For display purposes, maps have been resampled from the
30-m observation scale to a 0.04-degree geographic grid.

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

Extended Data Fig. 4 | Results of sensitivity analysis when rates of AGB accumulation derived from inventory data for different forest types of the
United States in the standard model are replaced by IPCC Tier 1 default rates in the alternative model. Top panel: change in net GHG flux between
standard model and sensitivity analysis model; Bottom panel: Delineation of areas that remain a net GHG source or sink in the sensitivity analysis model
vs. those that switch from being a net source or sink to a net sink or source as a result of the changes applied. For display purposes, maps have been
resampled from the 30-m observation scale to a 0.04- degree geographic grid.

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles

Extended Data Fig. 5 | Results of sensitivity analysis when the number of years of growth in the GHG flux model is assumed to be 19 in the alternative
model vs. 6 in the standard model for pixels of tree cover gain since the year 2000. Top panel: Percent change in net GHG flux between standard model
and sensitivity analysis model; Bottom panel: Delineation of areas that remain a net GHG source or sink in the sensitivity analysis model vs. those that
switch from being a net source or sink to a net sink or source as a result of the changes applied. For display purposes, maps have been resampled from the
30-m observation scale to a 0.04-degree geographic grid.

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

Extended Data Fig. 6 | Results of sensitivity analysis when tree cover loss in the GHG flux model is attributed to commodity-driven deforestation in
the alternative model vs. shifting agriculture in the standard model. Top panel: Percent change in net GHG flux between standard model and sensitivity
analysis model; Bottom panel: Delineation of areas that remain a net GHG source or sink in the sensitivity analysis model vs. those that switch from being
a net source or sink to a net sink or source as a result of the changes applied. For display purposes, maps have been resampled from the 30-m observation
scale to a 0.04-degree geographic grid.

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles

Extended Data Fig. 7 | Results of sensitivity analysis when the post- deforestation land-use assumption in the GHG flux model is changed from
cropland in the standard model to grassland in the alternative model. Top panel: Percent change in net GHG flux between standard model and sensitivity
analysis model; Bottom panel: Delineation of areas that remain a net GHG source or sink in the sensitivity analysis model vs. those that switch from being
a net source or sink to a net sink or source as a result of the changes applied. For display purposes, maps have been resampled from the 30-m observation
scale to a 0.04-degree geographic grid.

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

Extended Data Fig. 8 | Results of sensitivity analysis when assumptions about carbon uptake in primary forests and intact forest landscapesare
changed to zero carbon uptake in the alternative model vs. positive carbon uptake in the standard model. Top panel: Percent change in net GHG flux
between standard model and sensitivity analysis model; Bottom panel: Delineation of areas that remain a net GHG source or sink in the sensitivity analysis
model vs. those that switch from being a net source or sink to a net sink or source as a result of the changes applied. For display purposes, maps have been
resampled from the 30-m observation scale to a 0.04-degree geographic grid.

Nature Climate Change | www.nature.com/natureclimatechange

Nature Climate Change Articles

Extended Data Fig. 9 | Gross forest-related emissions, 2001–2019. Emissions reflect all stand-replacement disturbances (natural and anthropogenic)
observable in Landsat imagery.

Nature Climate Change | www.nature.com/natureclimatechange

Articles Nature Climate Change

Extended Data Fig. 10 | Conceptual framework for modelling forest- related GHG fluxes. For each 30-m pixel included in the model, gross forest-related
emissions and removals are estimated as the product of activity data and emission/removal factors. Net forest GHG flux is the sum of gross fluxes. Text
and arrows in orange are portions of the removals methodology that are passed into the emissions methodology.

Nature Climate Change | www.nature.com/natureclimatechange

 nature research | reporting summary
Corresponding author(s): Nancy Harris, [email protected]
Last updated by author(s): November 23, 2020

Reporting Summary
Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency
in reporting. For further information on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement
A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly
The statistical test(s) used AND whether they are one- or two-sided
Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested

A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons
A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient)
AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted
Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes
Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated
Our web collection on statistics for biologists contains articles on many of the points above.

Software and code

Policy information about availability of computer code
Data collection No software was used to collect data in this study.

Data analysis Code for developing the model and summarizing geospatial statistics in this study are available on GitHub at https://github.com/wri/carbon-
budget and https://github.com/wri/gfw_forest_loss_geotrellis, respectively. Figure 2 was generated using R v3.5.0 and package ggplot2 v3.3.
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and
reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data
All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:
- Accession codes, unique identifiers, or web links for publicly available datasets
- A list of figures that have associated raw data
April 2020

- A description of any restrictions on data availability

Geospatial data generated from the current study will be made publicly available on Global Forest Watch's Open Data Portal (http://data.globalforestwatch.org/)
and from the corresponding author upon request. Summary geospatial statistics will be available from the Global Forest Watch dashboards and a publicly accessible
API, as well as from the corresponding author upon request. All data analyzed as part of the current study are either publicly available or were obtained by the
corresponding author upon request.

1
 nature research | reporting summary
Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.
Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Ecological, evolutionary & environmental sciences study design

All studies must disclose on these points even when the disclosure is negative.
Study description The study design is structured as a geographic information system (GIS) model and is based on a synthesis of spatially explicit data
sources to map gross and net greenhouse gas emissions and removals from global forest lands at 30-m resolution between 2001 and
2019.

Research sample Results of this study are not based on a statistical sampling design but on published wall-to-wall maps of global tree cover and tree
cover change generated by Hansen et al. (2013) as well as other published geospatial data sets. The study area includes all global
land except Antarctica and some small oceanic and Arctic islands. Table S3 documents the dozens of data sources we combined for
this analysis. The model includes greenhouse gases CO2, CH4, and N2O, and six carbon pools.

Sampling strategy Results of this study are not based on a statistical sampling design, although quantification of uncertainty in this study is based in part
on results from a probability-based stratified random sample of 1,500 blocks (120x120 m) to validate map-based estimates of global
forest change. Three sub-strata were created per biome: no change, loss and gain. The sample allocation for each biome was 150
blocks for no change, 90 for change and 60 for gain (1,500 blocks total). For more information, refer to Hansen et al. (2013).

Data collection Results of this study are based on a synthesis of published datasets, not on original data collection. The datasets are combined using
existing frameworks for calculating GHG emissions and removals (IPCC guidelines for national GHG inventory reporting).

Timing and spatial scale The time period of analysis covers 2001 to 2019. The starting and ending years were determined by the ranges of key data sources
available during model development. The spatial scale of this study's results is 30-m resolution with full coverage of all continental
land.

Data exclusions No data were excluded from this analysis.

Reproducibility The reproducibility of this study's results is facilitated by providing all model code used in a public repository (GitHub). Various data
sources used as inputs into the study are publicly acccessible and freely available.

Randomization This is not relevant to our study because results are not based on a statistical sampling design.

Blinding This is not relevant to our study because results are not generated based on a statistical sampling design.

Did the study involve field work? Yes No

Reporting for specific materials, systems and methods

We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material,
system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Materials & experimental systems Methods

n/a Involved in the study n/a Involved in the study
Antibodies ChIP-seq
Eukaryotic cell lines Flow cytometry
Palaeontology and archaeology MRI-based neuroimaging
Animals and other organisms
Human research participants
Clinical data
April 2020

Dual use research of concern

View publication stats