ArticlePDF Available

High-Resolution Global Maps of 21st-Century Forest Cover Change

Authors:

Abstract and Figures

Forests in Flux Forests worldwide are in a state of flux, with accelerating losses in some regions and gains in others. Hansen et al. (p. 850 ) examined global Landsat data at a 30-meter spatial resolution to characterize forest extent, loss, and gain from 2000 to 2012. Globally, 2.3 million square kilometers of forest were lost during the 12-year study period and 0.8 million square kilometers of new forest were gained. The tropics exhibited both the greatest losses and the greatest gains (through regrowth and plantation), with losses outstripping gains.
Content may be subject to copyright.
16. T. Conrad, A. Akhtar, Nat. Rev. Genet. 13,123134
(2011).
17. A. A. Alekseyenko et al., Cell 134,599609 (2008).
18. A. A. Alekseyenko et al., Genes Dev. 27,853858 (2013).
19. I. Marín, A. Franke, G. J. Bashaw, B. S. Baker, Nature
383,160163 (1996).
20. Materials and methods are available as supplementary
materials on Science Online.
21. Q. Zhou, C. E. Ellison, V. B. Kaiser, A. A. Alekseyenko,
A. A. Gorchakov, D. Bachtrog, PLoS Biol. 11, e1001711
(2013).
22. M. Steinemann, S. Steinemann, Proc. Natl. Acad. Sci.
U.S.A. 89, 75917595 (1992).
23. J. Jurka, Repbase Reports 12, 1376 (2012).
24. V. V. Kapitonov, J. Jurka, Trends Genet. 23,521529
(2007).
25. J. R. Bateman, A. M. Lee, C. T. Wu, Genetics 173,
769777 (2006).
26. Q. Zhou, D. Bachtrog, Science 337,341345 (2012).
27. C. E. Grant, T. L. Bailey, W. S. Noble, Bioinformatics 27,
10171018 (2011).
Acknowledgments: This work was funded by NIH grants
(R01GM076007 and R01GM093182) and a Packard
Fellowship to D.B. and a NIH postdoctoral fellowship to
C.E.E. All DNA-sequencing reads generated in this study
are deposited at the National Center for Biotechnology
Information Short Reads Archive (www.ncbi.nlm.nih.gov/sra)
under the accession no. SRS402821. The genome assemblies
are available at the National Center for Biotechnology
Information under BioProject PRJNA77213. We thank
Z. Walton and A. Gorchakov for technical assistance.
Supplementary Materials
www.sciencemag.org/content/342/6160/846/suppl/DC1
Materials and Methods
Supplementary Text
Figs. S1 to S20
Tables S1 to S3
References (2854)
23 April 2013; accepted 30 September 2013
10.1126/science.1239552
High-Resolution Global Maps of
21st-Century Forest Cover Change
M. C. Hansen,
1
*P. V. Potapov,
1
R. Moore,
2
M. Hancher,
2
S. A. Turubanova,
1
A. Tyukavina,
1
D. Thau,
2
S. V. Stehman,
3
S. J. Goetz,
4
T. R. Loveland,
5
A. Kommareddy,
6
A. Egorov,
6
L. Chini,
1
C. O. Justice,
1
J. R. G. Townshend
1
Quantification of global forest change has been lacking despite the recognized importance of
forest ecosystem services. In this study, Earth observation satellite data were used to map global
forest loss (2.3 million square kilometers) and gain (0.8 million square kilometers) from
2000 to 2012 at a spatial resolution of 30 meters. The tropics were the only climate domain to
exhibit a trend, with forest loss increasing by 2101 square kilometers per year. Brazils
well-documented reduction in deforestation was offset by increasing forest loss in Indonesia,
Malaysia, Paraguay, Bolivia, Zambia, Angola, and elsewhere. Intensive forestry practiced within
subtropical forests resulted in the highest rates of forest change globally. Boreal forest loss
due largely to fire and forestry was second to that in the tropics in absolute and proportional terms.
These results depict a globally consistent and locally relevant record of forest change.
Changes in forest cover affect the delivery
of important ecosystem services, including
biodiversity richness, climate regulation,
carbon storage, and water supplies (1). However,
spatially and temporally detailed information on
global-scale forest change does not exist; pre-
vious efforts have been either sample-based or
employed coarse spatial resolution data (24).
We mapped global tree cover extent, loss, and
gain for the period from 2000 to 2012 at a spatial
resolution of 30 m, with loss allocated annually.
Our global analysis, based on Landsat data, im-
proves on existing knowledge of global forest
extent and change by (i) being spatially explicit;
(ii) quantifying gross forest loss and gain; (iii)
providing annual loss information and quantify-
ing trends in forest loss; and (iv) being derived
through an internally consistent approach that is
exempt from the vagaries of different definitions,
methods, and data inputs. Forest loss was defined
as a stand-replacement disturbance or the com-
plete removal of tree cover canopy at the Landsat
pixel scale. Forest gain was defined as the inverse
of loss, or the establishment of tree canopy from
a nonforest state. A total of 2.3 million km
2
of
forest were lost due to disturbance over the study
period and 0.8 million km
2
of new forest es-
tablished. Of the total area of combined loss
and gain (2.3 million km
2
+ 0.8 million km
2
),
0.2 million km
2
of land experienced both loss
and subsequent gain in forest cover during the
study period. Global forest loss and gain were
related to tree cover density for global climate
domains, ecozones, and countries (refer to tables
S1 to S3 for all data references and comparisons).
Results are depicted in Fig. 1 and are viewable
at full resolution at http://earthenginepartners.
appspot.com/science-2013-global-forest.
The tropical domain experienced the greatest
total forest loss and gain of the four climate
domains (tropical, subtropical, temperate, and
boreal), as well as the highest ratio of loss to
gain (3.6 for >50% of tree cover), indicating
the prevalence of deforestation dynamics. The
tropics were the only domain to exhibit a statis-
tically significant trend in annual forest loss, with
an estimated increase in loss of 2101 km
2
/year.
Tropical rainforest ecozones totaled 32% of
global forest cover loss, nearly half of which oc-
curred in South American rainforests. The trop-
ical dry forests of South America had the highest
rate of tropical forest loss, due to deforestation
dynamics in the Chaco woodlands of Argentina,
Paraguay (Fig. 2A), and Bolivia. Eurasian rain-
forests (Fig. 2B) and dense tropical dry forests
of Africa and Eurasia also had high rates of
loss.
Recently reported reductions in Brazilian
rainforest clearing over the past decade (5)were
confirmed, as annual forest loss decreased on
average 1318 km
2
/year. However, increased an-
nual loss of Eurasian tropical rainforest (1392
km
2
/year), African tropical moist deciduous forest
(536 km
2
/year), South American dry tropical for-
est (459 km
2
/year), and Eurasian tropical moist
deciduous (221 km
2
/year) and dry (123 km
2
/year)
forests more than offset the slowing of Brazilian
deforestation. Of all countries globally, Brazil
exhibited the largest decline in annual forest loss,
with a high of over 40,000 km
2
/year in 2003 to
2004 and a low of under 20,000 km
2
/year in
2010 to 2011. Of all countries globally, Indonesia
exhibited the largest increase in forest loss
(1021 km
2
/year), with a low of under 10,000 km
2
/year
from 2000 through 2003 and a high of over
20,000 km
2
/year in 2011 to 2012. The converging
rates of forest disturbance of Indonesia and Brazil
are shown in Fig. 3. Although the short-term
decline of Brazilian deforestation is well docu-
mented, changing legal frameworks governing
Brazilian forests could reverse this trend (6). The
effectiveness of Indonesias recently instituted
moratorium on new licensing of concessions in
primary natural forest and peatlands (7), initiated
in 2011, is to be determined.
Subtropical forests experience extensive for-
estry land uses where forests are often treated as a
crop and the presence of long-lived natural for-
ests is comparatively rare (8). As a result, the
highest proportional losses of forest cover and the
lowest ratio of loss to gain (1.2 for >50% of tree
cover) occurred in the subtropical climate do-
main. Aggregate forest change, or the proportion
of total forest loss and gain relative to year-2000
forest area [(loss+gain)/2000 forest], equaled 16%,
or more than 1% per year across all forests within
the domain. Of the 10 subtropical humid and dry
forest ecozones, 5 have aggregate forest change
>20%, three >10%, and two >5%. North Amer-
ican subtropical forests of the southeastern United
States are unique in terms of change dynamics
because of short-cycle tree planting and harvest-
ing (Fig. 2C). The disturbance rate of this eco-
zone was four times that of South American
1
Department of Geographical Sciences, University of Mary land,
College Park, MD 20742, USA.
2
Google, Mountain View, CA,
USA.
3
Department of Forest and Natural Resources Manage-
ment, State University of New York, Syracuse, N Y, USA.
4
Woods
Hole Research Center, 149 Woods Hole Road, Falmouth, MA
02540, USA.
5
Earth Resources Observation and Science, United
States Geological Survey, 47914 252nd Street, Sioux Falls, SD
57198, USA.
6
Geographic Information Science Center of Ex-
cellence, South Dakota State University, Brookings, SD, USA.
*Corresponding author. E-mail: mhansen@umd.edu
15 NOVEMBER 2013 VOL 342 SCIENCE www.sciencemag.org850
REPORTS
Fig. 1. (A) Tree cover, (B) forest loss, and (C) forest gain. Acolorcom-
posite of tree cover in green, forest loss in red, forest gain in blue, and
forest loss and gain in magenta is shown in (D), with loss and gain en-
hanced for improved visualization. All map layers have been resampled
for display purposes from the 30-m observation scale to a 0.05° geo-
graphic grid.
Fig. 2. Regional subsets of 2000 tree cover and 2000 to 2012 forest loss and gain. (A)Paraguay,centeredat21.9°S,59.8°W;(B) Indonesia,
centered at 0.4°S, 101.5°E; (C) the United States, centered at 33.8°N, 93.3°W; and (D)Russia,centeredat62.1°N,123.4°E.
www.sciencemag.org SCIENCE VOL 342 15 NOVEMBER 2013 851
REPORTS
rainforests during the study period; over 31% of
its forest cover was either lost or regrown. Areas
of colocated loss and gain (magenta tones in Fig.
1D), indicating intensive forestry practices, are
found on all continents within the subtropical
climate domain, including South Africa, central
Chile, southeastern Brazil, Uruguay, southern
China, Australia, and New Zealand.
The temperate climatic domain has a forestry-
dominant change dynamic and a relatively low
ratio of loss to gain (1.6 for >50% of tree cover).
Oceanic ecozones, in particular, are similar to the
subtropics in the intensity of indicated forest land
use. The northwest United States is an area of
intensive forestry, as is the entire range of tem-
perate Canada. The intermountain West of North
America exhibits a loss dynamic, largely due to
fire, logging, and disease [for example, large-scale
tree mortality due to mountain pine bark beetle
infestation, most evident in British Colombia,
Canada (9)]. Temperate Europe has a forestry
dynamic with Estonia and Latvia exhibiting a
high ratio of loss to gain. Portugal, which strad-
dles the temperate and subtropical domains, has a
complicated dynamic of forestry and forest loss
due to fire; the resulting aggregate change dy-
namic is fourth in intensity globally. Elevated
loss due to storm damage is indicated for a few
areas. For example, a 2005 extratropical cyclone
led to a historic blowdown of southern Sweden
temperate forests, and a 2009 windstorm lev-
eled extensive forest areas in southwestern
France (10).
Fire is the most significant cause of forest loss
in boreal forests (11), and it occurred across a
range of tree canopy densities. Given slower
regrowth dynamics, the ratio of boreal forest loss
to gain is high over the study period (2.1 for >50%
of tree cover). Boreal coniferous and mountain
ecozones are similar in terms of forest loss rates,
with North America having a higher overall rate
and Eurasia a higher absolute area of loss. Forest
gain is substantial in the boreal zone, with Eur-
asian coniferous forests having the largest area
of gain of all global ecozones during the study
period, due to forestry, agricultural abandonment
(12), and forest recovery after fire [as in Euro-
pean Russia and the Siberia region of Russia
(Fig. 2D)]. Russia has the most forest loss glob-
ally. Co-located gain and loss are nearly absent in
the high-latitude forests of the boreal domain,
reflecting a slower regrowth dynamic in this cli-
matic domain. Areas with loss and gain in close
proximity, indicating forestry land uses, are found
within nearly the entirety of Sweden and Finland,
the boreal/temperate transition zone in eastern
Canada, parts of European Russia, and along the
Angara River in central Siberia, Russia.
Agoaloflarge-arealandcovermappingisto
produce globally consistent characterizations that
have local relevance and utility; that is, reliable
information across scales. Figure S1 reflects this
capability at the national scale. Two measures of
change, (i) proportion of total aggregate forest
change relative to year-2000 forest area [(loss +
gain)/2000 forest], shown in column q of table
S3; and (ii) proportion of total change that is loss
[loss/(loss + gain)], calculated from columns b
and c in table S3, are displayed. The proportion
of total aggregate forest change emphasizes coun-
tries with likely forestry practices by including
both loss and gain in its calculation, whereas the
proportion of loss to gain measure differentiates
countries experiencing deforestation or another
loss dynamic without a corresponding forest re-
covery signal. The two ratio measures normal-
ize the forest dynamic in order to directly compare
national-scale change regardless of country size
or absolute area of change dynamic. In fig. S1,
countries that have lost forests without gain are
high on the yaxis (Paraguay, Mongolia, and
Zambia). Countries with a large fraction of forest
area disturbed and/or reforested/afforested are
high on the xaxis (Swaziland, South Africa, and
Uruguay). Thirty-one countries have an aggre-
gate dynamic >1% per year, 11 have annual loss
rates >1%, and 5 have annual gain rates of >1%.
Figure S2 compares forest change dynamics dis-
aggregated by ecozone (http://foris.fao.org/static/
data/fra2010/ecozones2010.jpg).
Brazil is a global exception in terms of forest
change, with a dramatic policy-driven reduction
in Amazon Basin deforestation. Although Bra-
zilian gross forest loss is the second highest glob-
ally, other countries, including Malaysia, Cambodia,
Cote dIvoire, Tanzania, Argentina, and Paraguay,
experienced a greater percentage of loss of forest
cover. Given consensus on the value of natural
forests to the Earth system, Brazilspolicyinter-
vention is an example of how awareness of forest
valuation can reverse decades of previous wide-
spread deforestation. International policy ini-
tiatives, such as the United Natons Framework
Convention of Climate Change Reducing Emis-
sions from Deforestation and forest Degradation
(REDD) program (13), often lack the institutional
investment and scientific capacity to begin im-
plementation of a program that can make use of
the global observational record; in other words,
the policy is far ahead of operational capabilities
(14). BrazilsuseofLandsatdataindocumenting
trends in deforestation was crucial to its policy
formulation and implementation. To date, only
Brazil produces and shares spatially explicit
information on annual forest extent and change.
The maps and statistics we present can be used as
an initial reference for a number of countries
lacking such data, as a spur to capacity building
in the establishment of national-scale forest ex-
tent and change maps, and as a basis of com-
parison in evolving national monitoring methods.
Global-scale studies require systematic global
image acquisitions available at low or no direct
Fig. 3. Annual forest loss totals for Brazil and Indonesia from 2000 to 2012. The forest loss annual increment is the slope of the estimated
trend line of change in annual forest loss.
15 NOVEMBER 2013 VOL 342 SCIENCE www.sciencemag.org
852
REPORTS
cost and the preprocessing of geometric and ra-
diometric corrections of satellite imagery, exempli-
fied by the Landsat program. Given such progressive
data policies and image processing capabilities, it
is now possible to use advanced computing sys-
tems, such as the Google cloud, to efficiently
process and characterize global-scale time-series
data sets in quantifying land change. There are
several satellite systems in place or planned for
collecting data with similar capabilities to Land-
sat. Similar free and open data policies would
enable greater use of these data for public good
and foster greater transparency of the development,
implementation, and reactions to policy initia-
tives that affect the worldsforests.
The information content of the presented data
sets, which are publicly available, provides a
transparent, sound, and consistent basis on which
to quantify critical environmental issues, includ-
ing (i) the proximate causes of the mapped forest
disturbances (15); (ii) the carbon stocks and asso-
ciated emissions of disturbed forest areas (1618);
(iii) the rates of growth and associated carbon
stock gains for both managed and unmanaged
forests (19); (iv) the status of remaining intact
natural forests of the world and threats to bio-
diversity (20,21); (v) the effectiveness of existing
protected-area networks (22); (vi) the economic
drivers of natural forest conversion to more in-
tensive land uses (23); (vii) the relationships be-
tween forest dynamics and social welfare, health,
and other relevant human dimensions data; (viii)
forest dynamics associated with governance and
policy actionsand many other regional-to-global
scale applications.
References and Notes
1. J. A. Foley et al., Science 309,570574 (2005).
2. M. C. Hansen, S. V. Stehman, P. V. Potapov, Proc. Natl.
Acad. Sci. U.S.A. 107, 86508655 (2010).
3. Food and Agricultural Organization of the United
Nations, Global Forest Land-Use Change 1990-2005,
FAO Forestry Paper No. 169 (Food and Agricultural
Organization of the United Nations, Rome, 2012).
4. M. Hansen, R. DeFries, Ecosystems 7,695716 (2004).
5. Instituto Nacional de Pesquisas Especias, Monitoring of
the Brazilian Amazonian Forest by Satellite, 2000-2012
(Instituto Nacional de Pesquisas Especias, San Jose dos
Campos, Brazil, 2013).
6. G. Sparovek, G. Berndes, A. G. O. P. Barretto, I. L. F. Klug,
Environ. Sci. Policy 16,6572 (2012).
7. D. P. Edwards, W. F. Laurance, Nature 477, 33 (2011).
8. M. Drummond, T. Loveland, Bioscience 60,286298
(2010).
9. W. A. Kurz et al., Nature 452,987990 (2008).
10. B. Gardiner et al., Destructive Storms in European
Forests: Past and Forthcoming Impacts (European Forest
Institute, Freiburg, Germany, 2010).
11. P. Potapov, M. Hansen, S. Stehman, T. Loveland,
K. Pittman, Remote Sens. Environ. 112,37083719 (2008).
12. A. Prishchepov, D. Muller, M. Dubinin, M. Baumann,
V. Radeloff, Land Use Policy 30,873884 (2013).
13. United Nations Framework Convention on Climate
Change, Reducing Emissions from Deforestation in
Developing Countries: Approaches to Stimulate
Action Draft Conclusions Proposed by the President
(United Nations Framework Convention on Climate
Change Secretariat, Bonn, Germany, 2005).
14. R. Houghton et al., Carbon Manage.1,253259.
15. H. Geist, E. Lambin, Bioscience 52,143150 (2002).
16. S. S. Saatchi et al., Proc. Natl. Acad. Sci. U.S.A. 108,
98999904 (2011).
17. A. Baccini et al., Nature Clim. Change 2,182185 (2012).
18. N. L. Harris et al., Science 336, 15731576 (2012).
19. R. Waterworth, G. Richards, C. Brack, D. Evans,
For. Ecol. Manage. 238,231243 (2007).
20. P. Potapov et al., Ecol. Soc. 13, 51 (2008).
21. T. M. Brooks et al., Science 313,5861 (2006).
22. A. S. Rodrigues et al., Nature 428,640643 (2004).
23. T. Rudel, Rural Sociol. 63,533552 (1998).
Acknowledgments: Support for Landsat data analysis and
characterization was provided by the Gordon and Betty
Moore Foundation, the United States Geological Survey, and
Google, Inc. GLAS data analysis was supported by the
David and Lucile Packard Foundation. Development of all
methods was supported by NASA through its Land Cover and
Land Use Change, Terrestrial Ecology, Applied Sciences, and
MEaSUREs programs (grants NNH05ZDA001N, NNH07ZDA001N,
NNX12AB43G, NNX12AC78G, NNX08AP33A, and NNG06GD95G)
and by the U.S. Agency for International Development
through its CARPE program. Any use of trade, firm, or product
names is for descriptive purposes only and does not imply
endorsement by the U.S. government. Results are depicted and
viewable online at full resolution: http://earthenginepartners.
appspot.com/science-2013-global-forest.
Supplementary Materials
www.sciencemag.org/content/342/6160/850/suppl/DC1
Materials and Methods
Supplementary Text
Figs. S1 to S8
Tables S1 to S5
References (2440)
14 August 2013; accepted 15 October 2013
10.1126/science.1244693
Changes in Cytoplasmic Volume Are
Sufficient to Drive Spindle Scaling
James Hazel,
1
Kaspars Krutkramelis,
2
Paul Mooney,
1
Miroslav Tomschik,
1
Ken Gerow,
3
John Oakey,
2
J. C. Gatlin
1
*
The mitotic spindle must function in cell types that vary greatly in size, and its dimensions
scale with the rapid, reductive cell divisions that accompany early stages of development. The
mechanism responsible for this scaling is unclear, because uncoupling cell size from a developmental
or cellular context has proven experimentally challenging. We combined microfluidic technology
with Xenopus egg extracts to characterize spindle assembly within discrete, geometrically defined
volumes of cytoplasm. Reductions in cytoplasmic volume, rather than developmental cues or
changes in cell shape, were sufficient to recapitulate spindle scaling observed in Xenopus
embryos. Thus, mechanisms extrinsic to the spindle, specifically a limiting pool of cytoplasmic
component(s), play a major role in determining spindle size.
Organelles and other intracellular struc-
tures must scale with cell size in order to
function properly. Maintenance of these
dimensional relationships is challenged by the
rapid and reductive cell divisions that character-
ize early embryogenesis in many organisms. The
cellular machine that drives these divisions, the
mitotic spindle, functions to segregate chromo-
somes in cells that vary greatly in size while also
adapting to rapid changes in cell size. The issue
of scale is epitomized during Xenopus embryo-
genesis, where a rapid series of divisions reduces
cell size 100-fold: from the 1.2-mm-diameter fer-
tilized egg to ~12-mm-diameter cells in the adult
frog (1). In large blastomeres, spindle length reaches
an upper limit that is uncoupled from changes in
cell size. However, as cell size decreases, a strong
correlation emerges between spindle length and
cell size (2). Although this scaling relationship
has been characterized in vivo for several differ-
ent org anisms, little is known about the direct
regulation of spindle size by cell size or the under-
lying mechanism(s) (24). Spindle size may be
directly dictated by the physical dimensions of a
cell, perhaps through microtubule-mediated in-
teraction with the cell cortex [i.e., boundary sensing
(57)]. Alternatively, cell size could constrain spin-
dle size by providing a fixed and finite cytoplasmic
volume and, therefore, a limiting pool of resources
such as cytoplasmic spindle assembly or length-
determining components [i.e., component limi-
tation (8,9)]. Last, mechanisms intrinsic to the
spindle could be actively tuned in response to
systematic changes in cytoplasmic composition
occurring during development [i.e., developmen-
tal cues (10,11)].
To elucidate the responsible scaling mechanism(s),
we developed a microfluidic-based platform to con-
fine spindle assembly in geometrically defined vol-
umes of Xenopus egg extract (12). Interphase extract
containing Xenopus sperm nuclei was induced
to enter mitosis and immediately pumped into a
microfluidic droplet-generating device before
nuclear envelope breakdown and the onset of
spindle assembly. At the same time, a fluorinated
oil/surfactant mixture was pumped into the de-
vice through a second inlet. These two discrete,
immiscible phases merged at a T-shaped junction
within the device to produce stable emulsions of
extract droplets in a continuous oil phase (Fig. 1,
AandC).ChangingtheT-junctionchanneldi-
mensions and relative flow rates of the two phases
1
Department of Molecular Biology, University of Wyoming,
Laramie, WY 82071, USA.
2
Department of Chemical and Pe-
troleum Engineering, University of Wyoming, Laramie, WY
82071, USA.
3
Department of Statistics,UniversityofWyoming,
Laramie, WY 82071, USA.
*Corresponding author. E-mail: jgatlin@uwyo.edu
www.sciencemag.org SCIENCE VOL 342 15 NOVEMBER 2013 853
REPORTS
www.sciencemag.org/content/342/6160/850/suppl/DC1
Supplementary Materials for
High-Resolution Global Maps of 21st-Century Forest Cover Change
M. C. Hansen,* P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina,
D. Thau, S. V. Stehman, S. J. Goetz, T. R. Loveland, A. Kommareddy, A. Egorov, L.
Chini, C. O. Justice, J. R. G. Townshend
*Corresponding author. E-mail: mhansen@umd.edu
Published 15 November 2013, Science 342, 850 (2013)
DOI: 10.1126/science.1244693
This PDF file includes:
Materials and Methods
Supplementary Text
Figs. S1 to S8
Tables S1 to S5
References (2440)
2
Materials and Methods
The study area included all global land except for Antarctica and a number of Arctic
islands, totaling 128.8Mkm2, or the equivalent of 143 billion 30m Landsat pixels. For
this study, trees were defined as all vegetation taller than 5m in height. Forest loss was
defined as a stand-replacement disturbance. Results were disaggregated by reference
percent tree cover stratum (e.g. >50% crown cover to ~0% crown cover) and by year.
Forest degradation (24), for example selective removals from within forested stands that
do not lead to a non-forest state, was not included in the change characterization. Gain
was defined as the inverse of loss, or a non-forest to forest change; longer-lived
regrowing stands of tree cover that did not begin as non-forest within the study period
were not mapped as forest gain. Gain was related to percent tree crown cover densities
>50% and reported as a twelve year total. In this study, the term “forest” refers to tree
cover and not land use unless explicitly stated, e.g. “forest land use”.
The global Landsat analysis was performed using Google Earth Engine, a cloud
platform for earth observation data analysis that combines a public data catalog with a
large-scale computational facility optimized for parallel processing of geospatial data.
Google Earth Engine contains a nearly complete set of imagery from the Landsat 4, 5, 7,
and 8 satellites downloaded from the USGS Earth Resources Observation and Science
archive (25). For this study, we analyzed 654,178 growing season Landsat 7 Enhanced
Thematic Mapper Plus (ETM+) scenes from a total of 1.3 million available at the time of
the study. Growing season data are more appropriate for land cover mapping than
imagery captured during senescence or dormant seasonal periods (26). Automated
Landsat pre-processing steps included: (i) image resampling, (ii) conversion of raw
digital values (DN) to top of atmosphere (TOA) reflectance, (iii) cloud/shadow/water
screening and quality assessment (QA), and (iv) image normalization. All pre-processing
steps were tested at national scales around the globe using a method prototyped for the
Democratic Republic of Congo (27). The stack of QA layers was used to create a per-
pixel set of cloud-free image observations which in turn was employed to calculate time-
series spectral metrics. Metrics represent a generic feature space that facilitates regional-
scale mapping and have been used extensively with MODIS and AVHRR data (2,4) and
more recently with Landsat data in characterizing forest cover loss (27,28). Three groups
of per-band metrics were employed over the study interval: (i) reflectance values
representing maximum, minimum and selected percentile values (10, 25, 50, 75 and 90%
percentiles); (ii) mean reflectance values for observations between selected percentiles
(for the max-10%, 10-25%, 25-50%, 50-75%, 75-90%, 90%-max, min-max, 10-90%, and
25-75% intervals); and (iii) slope of linear regression of band reflectance value versus
image date. Training data to relate to the Landsat metrics were derived from image
interpretation methods, including mapping of crown/no crown categories using very high
spatial resolution data such as Quickbird imagery, existing percent tree cover layers
derived from Landsat data (29), and global MODIS percent tree cover (30), rescaled using
the higher spatial resolution percent tree cover data sets. Image interpretation on-screen
was used to delineate change and no change training data for forest cover loss and gain.
Percent tree cover, forest loss and forest gain training data were related to the time-
series metrics using a decision tree. Decision trees are hierarchical classifiers that predict
class membership by recursively partitioning a data set into more homogeneous or less
3
varying subsets, referred to as nodes (31). For the tree cover and change products, a
bagged decision tree methodology was employed. Forest loss was disaggregated to
annual time scales using a set of heuristics derived from the maximum annual decline in
percent tree cover and the maximum annual decline in minimum growing season
Normalized Vegetation Difference Index (NDVI). Trends in annual forest loss were
derived using an ordinary least squares slope of the regression of y=annual loss versus
x=year. Outputs per pixel include annual percent tree cover, annual forest loss from 2000
to 2012, and forest gain from 2000 to 2012. To facilitate processing, each continent was
characterized individually: North America, South America, Eurasia, Africa, and
Australia.
Earth Engine uses a lazy computation model in which a sequence of operations may
be executed either interactively on-the-fly or in bulk over a complete data set. We used
the former mode during development and debugging, and the latter mode during the
computation of the final data products. In both cases all image processing operations
were performed in parallel across a large number of computers, and the platform
automatically handled data management tasks such as data format conversion,
reprojection and resampling, and associating image metadata with pixel data. Large-
scale computations were managed using the FlumeJava framework (32). A total of 20
terapixels of data were processed using one million CPU-core hours on 10,000 computers
in order to characterize year 2000 percent tree cover and subsequent tree cover loss and
gain through 2012.
Supplementary Text
Comparison with FAO data
The standard reference for global scale forest resource information is the UNFAO’s
Forest Resource Assessment (FRA) (33), produced at decadal intervals. There are several
limitations of the FRA reports that diminish their utility for global change assessments,
including (i) inconsistent methods between countries; (ii) defining “forest” based on land
use instead of land cover thereby obscuring the biophysical reality of whether tree cover
is present; (iii) forest area changes reported only as net values; and (iv) forest definitions
used in successive reports have changed over time (34).
Several discrepancies exist between FAO and earth observation-derived forest area
change data. For example, the large amount of tree cover change observed in satellite
imagery in Canada and the USA does not conform to the land use definitions applied in
the FRA for these countries. While there is significant forest change from a biophysical
perspective (i.e., forest cover), there is little or no land use change, the main criterion
used in the FRA report. Additionally, China, and to a lesser extent India, report
significant forest gains that are not readily observable in time-series satellite imagery,
including this analysis (Fig. S3). Large country change area discrepancies such as these
preclude a significant correlation between FAO and Landsat-based country data at the
global scale. However, regional differences in strength of agreement exist, and examples
are illustrated in Fig. S3 and Tab. S4. The region with the highest correlation between
FAO and Landsat net change is Latin America. Deforestation is the dominant dynamic,
and a number of countries, including Brazil, employ earth observation data in estimating
forest area change for official reporting. There is much less agreement for African
countries, though the correlation improves when lowering the tree cover threshold to
4
include more change. The lack of agreement in Africa reflects the difficult nature of
mapping change in environments with a range of tree cover as well as the lack of
systematic forest inventories and mapping capabilities for many African countries.
Southeast Asian countries exhibit changes primarily in dense canopy forests. However,
there is little correlation between Landsat-based change estimates and FAO data. The
forestry dynamics and differing governance and development contexts within this region
may lead to inconsistencies between countries. European data have the least correlation
of the regions examined, with comparatively little net area change reported in either our
Landsat analysis or in the FAO FRA.
The importance of forest definition and its impact on change area estimation is seen
for countries located in boreal and dry tropical climates. Our estimate of Canada’s net
change from the Landsat-based study doubled when including forest loss across all tree
cover strata, largely due to extensive burning in open boreal woodlands. Countries such
as Australia, Paraguay and Mozambique have similar outcomes related to disturbances
occurring within a range of tropical forest, woodland and parkland environments.
Gross forest area gain and loss for >50% tree cover were also compared to FAO
roundwood production data summed by country from 2000 to 2011 (Fig. S4). FAO data
are available at http://faostat.fao.org/. The national coniferous and non-coniferous
"total roundwood" production data (in cubic meters) were multiplied by 0.225MgC/m^3
and 0.325 MgC/m^2 respectively, and then added together to give national total
roundwood production in Megatons of carbon. Tab. S4 illustrates the strength of the
relationship between FRA roundwood production and Landsat-derived gross forest area
gain and loss for selected regions. While Africa and Southeast Asia have extremely poor
correlations, Landsat-derived forest area gain for Latin America and both forest gain and
loss for Europe exhibit strong correlations. The FRA roundwood production data
correlate well with satellite-based tree cover change area estimation for forestry land use-
dominated countries.
The FAO comparison reflects the confusion that results when comparing tabular
data that apply differing criteria in defining forest change. Deforestation is the
conversion of natural forests to non-forest land uses; the clearing of the same natural
forests followed by natural recovery or managed forestry is not deforestation and often
goes undocumented, whether in the tropical or boreal domains. Understanding where
such changes occur is impossible given the current state of knowledge, i.e. the FAO FRA.
While countries such as Canada and Indonesia both clear natural forests without
conversion to non-forest land uses, Indonesia reports over 5,000km2 per year of forest
area loss in the FRA while Canada reports no change. Consistent, transparent and spatio-
temporally explicit quantification of natural and managed forest change is required to
fully understand forest change from a biophysical and not solely forest land use
perspective.
Recent global forest mapping research
The FAO and others have turned to earth observation data, specifically Landsat
imagery, to provide a more consistent depiction of global forest change. Sample-based
methods have enabled national to global scale estimation of forest extent and change
(35,2,3). Such methods result in tabular aggregated estimates for areas having sufficient
sampling densities, but do not allow for local-scale area estimation or spatially explicit
representation of extent and change. While exhaustive land cover mapping using Landsat
5
data has been prototyped using single best-date image methods (36,37) based on the
National Aeronautics and Space Administration (NASA)-United States Geological
Survey (USGS) Global Land Survey data set (38), data mining of the Landsat archive to
quantify global forest cover change has not been implemented until this study.
Validation
The validation exercise was performed independently of the mapping exercise. Areas
of forest loss and gain were validated using a probability-based stratified random sample
of 120m blocks per biome. Boreal forest, temperate forest, humid tropical forest and dry
tropical forest biomes and other land constituted the five major strata, and were taken
from our previous study on global forest cover loss (2). The map product was used to
create three sub-strata 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). Each 120m sample block was interpreted into quartiles of reference change as gain
or loss (i.e., the proportion of gain or loss was interpreted as 0, 0.25, 0.50, 0.75, or 1),
where reference change was obtained as follows. Image interpretation of time-series
Landsat, MODIS and very high spatial imagery from GoogleEarth, where available, was
performed in estimating reference change for each sample block. Forest loss estimated
from the validation reference data set totaled 2.2Mkm2 (SE of 0.3Mkm2) compared to the
map total of 2.3Mkm2. Forest gain estimated from the validation sample totaled 0.9Mkm2
(SE of 0.2 Mkm2) compared to the map total of 0.8Mkm2. Fig. S5 shows the results as
mean map and validation change per block for the globe and per FAO climate domain.
Fig. S6 illustrates the mean per block difference of the map and reference loss and gain
estimates. Comparable map and reference loss and gain results were achieved at the
global and climate domain scales.
Estimated error matrices and accuracy summary statistics are shown in Tab. S5. For
loss, user’s and producer’s accuracies are balanced and greater than 80% per climate
domain and the globe as a whole. Results for forest gain indicate a possible
underestimate of tropical forest gain with a user’s accuracy of 82% and a producer’s
accuracy of 48%. However, the 95% confidence interval for the bias of tropical forest
gain (expressed as a % of land area) is 0.01% to 0.35%, indicating high uncertainty in the
validation estimate. A possible overestimate of boreal forest gain is also indicated.
Overall, the comparison of individually interpreted sample sites with the algorithm output
illustrates a robust product at the 120m pixel scale.
The annual allocation of change was validated using annual growing season NDVI
imagery from the MODIS sensor. All validation sample blocks were interpreted and if a
single, unambiguous drop in NDVI was observed in the MODIS NDVI time series, a year
of disturbance was assigned. Only 56% of the validation sample blocks were thus
assigned. The sample blocks interpreted represented 46% of the total forest loss mapped
with the Landsat imagery, a fraction similar to the 50% ratio of MODIS to Landsat-
detected change in a previous global forest cover loss study (39). For the interpreted
blocks, the mean deviation of the loss date was 0.06 years and the mean absolute
deviation was 0.29 years. The year of disturbance matched for 75.2% of the forest loss
events and 96.7% of the loss events occurred within one year before or after the estimated
year of disturbance.
A second evaluation of forest change was made using LiDAR (light detection and
ranging) data from NASA’s GLAS (Geoscience Laser Altimetry System) instrument
onboard the IceSat-1 satellite. Global GLAS release 28 (L1A Global Altimetry Data and
6
the L2 Global Land Surface Altimetry Data) data were screened for quality and viable
GLAS shots used to calculate canopy height (40). For forest loss, GLAS shots co-located
with Landsat forest loss by pixel were identified. The Landsat-estimated year of
disturbance was subtracted from the year of the GLAS shots and populations of ‘year
since disturbance’ created. Significant differences in height before and after Landsat-
derived forest loss indicate both a reasonable approximation of forest loss and year of
disturbance. Fig. S7 shows the results by ecozone, all of which passed Wilcoxon-Mann-
Whitney significance tests (non-parametric alternative of t-test) for pairs of +1\-1 and
+2\-2 years.
Forest gain was not allocated annually, but over the entire study period. To compare
GLAS-derived change in height with Landsat-derived gain, gain-identified pixels with no
tree cover for year 2000 and co-located with GLAS data were analyzed. Additionally,
only clustered gain was analyzed, specifically sites where six out of nine pixels within a
3x3 kernel were labeled as forest gain. Fig. S8 illustrates the results. All climate domains
except for the boreal passed Wilcoxon-Mann-Whitney significance tests for 2004 and
2008, the beginning and end years for GLAS data collection. The growth-limiting
climate of the boreal domain would preclude the observation of regrowth over such a
short period.
Table S1. Climate domain tree cover extent, loss and gain summary statistics (km2), ranked by total loss.
.
Climate
Domain
a)
Total
Loss
b)
Total
Gain
c)
Treecover 2000
Loss within treecover
Total loss
/ total
land area
(excluding
water) (%)
l)
>25% tree
cover loss
/ year
2000
>25% tree
cover (%)
m)
>50% tree
cover loss
/ year
2000
>50% tree
cover (%)
n)
>75% tree
cover loss
/ year
2000
>75% tree
cover (%)
o)
Total gain
/ year
2000
>50% tree
cover (%)
p)
>50% loss
+ total
gain /
2000
>50% tree
cover (%)
q)
Previous
column less
double
counting
pixels with
both loss
and gain (%)
r)
<25%
d)
26-50%
e)
51-75%
f)
76-100%
g)
<25%
h)
26-50%
i)
51-75%
j)
76-100%
k)
Tropical
1105786
247233
35866276
4175597
3524230
13241470
85479
125921
180106
714278
1.9
4.9
5.3
5.4
1.5
6.8
6.3
Boreal
606841
207100
11066215
2988449
3782283
4360312
51285
114990
147631
292935
2.7
5.0
5.4
6.7
2.5
8.0
7.9
Subtropical
305835
194103
19087918
769954
829023
1830148
37924
28761
32363
206787
1.4
7.8
9.0
11.3
7.3
16.3
13.8
Temperate
273390
155989
20938580
676500
1195036
4080868
7856
11588
25881
228064
1.0
4.5
4.8
5.6
3.0
7.8
7.5
Total
2291851
804425
86958989
8610500
9330573
23512797
182544
281261
385982
1442065
1.8
5.1
5.6
6.1
2.4
8.0
7.5
Table S2. Ecozone tree cover extent, loss and gain summary statistics (km2), ranked by total loss.
Ecozones by vegetation realm
AFR Africa
AUS Australia/Oceania
EAS –Eurasia
NAM North America
SAM South America
a)
Total
Loss
b)
Total
Gain
c)
Treecover 2000
Loss within treecover
Total loss
/ total
land area
(excluding
water) (%)
l)
>25% tree
cover loss
/ year
2000
>25% tree
cover (%)
m)
>50% tree
cover loss
/ year
2000
>50% tree
cover (%)
n)
>75% tree
cover loss
/ year
2000
>75% tree
cover (%)
o)
Total gain
/ year
2000
>50% tree
cover (%)
p)
>50% loss
+ total
gain /
2000
>50% tree
cover (%)
q)
Previous
column less
double
counting
pixels with
both loss
and gain (%)
r)
<25%
d)
26-50%
e)
51-75%
f)
76-100%
g)
<25%
h)
26-50%
i)
51-75%
j)
76-100%
k)
SAM Tropical rainforest
253435
33042
678685
100431
131817
5607324
2125
3847
10919
236544
3.9
4.3
4.3
4.2
0.6
4.9
4.7
EAS Boreal coniferous forest
229331
124488
1774827
904244
1577094
1831760
15280
29602
50285
134164
3.8
5.0
5.4
7.3
3.7
9.1
9.0
EAS Tropical rainforest
228011
104488
563048
116976
275092
1661696
2666
2856
15050
207438
8.7
11.0
11.5
12.5
5.4
16.9
14.4
SAM Tropical moist deciduous
forest
162095
33615
2456904
379932
404571
1003315
17675
25214
36530
82676
3.8
8.1
8.5
8.2
2.4
10.9
10.2
EAS Boreal mountain system
143573
36555
2266444
756678
1000381
1048086
23000
30952
38751
50869
2.8
4.3
4.4
4.9
1.8
6.2
6.1
NAM Subtropical humid forest
122915
103420
399595
55264
65639
522260
854
1582
6486
113993
11.8
19.0
20.5
21.8
17.6
38.1
31.2
NAM Boreal coniferous forest
120804
39978
314668
264683
474734
917683
3223
21444
22386
73751
6.1
7.1
6.9
8.0
2.9
9.8
9.5
AFR Tropical rainforest
96848
24575
511549
752270
742393
1941290
2272
10590
28693
55293
2.5
2.8
3.1
2.8
0.9
4.0
3.7
SAM Tropical dry forest
88784
3032
898296
382134
231661
150647
10581
30683
29104
18416
5.3
10.2
12.4
12.2
0.8
13.2
13.1
AFR Tropical moist deciduous
forest
84719
3649
2370104
1331743
845218
40070
27852
28032
25474
3361
1.8
2.6
3.3
8.4
0.4
3.7
3.5
NAM Temperate mountain system
82998
39105
791649
107937
153970
891664
2385
4274
6475
69864
4.3
7.0
7.3
7.8
3.7
11.0
10.7
NAM Boreal tundra woodland
63370
4342
924788
659389
405281
284991
3266
21744
20580
17779
2.8
4.5
5.6
6.2
0.6
6.2
6.2
EAS Temperate continental forest
63068
48939
3126680
210065
500447
1003265
1118
2023
9041
50886
1.3
3.6
4.0
5.1
3.3
7.2
7.1
NAM Temperate continental
forest
56749
26139
731687
80291
131018
1000282
216
549
2045
53939
2.9
4.7
4.9
5.4
2.3
7.3
6.9
EAS Subtropical humid forest
44693
17066
989723
193841
339619
469305
1208
1962
10323
31200
2.2
4.3
5.1
6.6
2.1
7.2
6.6
NAM Boreal mountain system
39485
1332
568948
180517
189178
233314
1145
8887
13879
15573
3.4
6.4
7.0
6.7
0.3
7.3
7.3
AFR Tropical dry forest
33259
3298
3076095
412938
125125
19520
11240
11866
7820
2332
0.9
3.9
7.0
11.9
2.3
9.3
8.2
EAS Tropical moist deciduous
forest
28166
7837
739252
91874
189904
298608
1829
1938
6724
17674
2.1
4.5
5.0
5.9
1.6
6.6
6.2
NAM Tropical moist deciduous
forest
25169
8174
284373
55814
73719
247384
737
1413
3932
19087
3.8
6.5
7.2
7.7
2.5
9.7
9.4
NAM Tropical rainforest
22777
2641
97950
23299
36653
253265
202
470
1770
20334
5.5
7.2
7.6
8.0
0.9
8.5
8.4
EAS Temperate oceanic forest
19089
13471
929909
40517
84314
216456
379
640
2583
15488
1.5
5.5
6.0
7.2
4.5
10.5
10.0
EAS Temperate mountain system
19037
8892
4100378
101904
174080
568263
995
1449
2534
14058
0.4
2.1
2.2
2.5
1.2
3.4
3.4
NAM Subtropical mountain system
18861
5631
370255
56439
45414
117601
2137
2207
2500
12018
3.2
7.6
8.9
10.2
3.5
12.4
11.5
SAM Subtropical humid forest
17149
25269
971549
38429
40670
132183
491
648
1350
14660
1.4
7.9
9.3
11.1
14.6
23.9
20.5
SAM Tropical mountain system
15624
4700
1154706
64730
90321
559185
561
799
1303
12961
0.8
2.1
2.2
2.3
0.7
2.9
2.7
EAS Tropical mountain system
14459
5388
301136
34772
93240
350957
304
495
2473
11189
1.9
3.0
3.1
3.2
1.2
4.3
3.9
EAS Tropical dry forest
13952
1755
1181479
74787
75701
67938
1575
1586
3699
7093
1.0
5.7
7.5
10.4
1.2
8.7
8.5
AFR Tropical mountain system
12236
4813
1098346
176193
80558
107352
1040
2161
3400
5635
0.8
3.1
4.8
5.2
2.6
7.4
6.8
EAS Subtropical dry forest
10987
5120
734840
62992
48342
87437
1515
1729
2677
5066
1.2
4.8
5.7
5.8
3.8
9.5
8.7
AUS Tropical rainforest
9972
3374
48615
20176
30783
687847
65
97
345
9465
1.3
1.3
1.4
1.4
0.5
1.8
1.7
EAS Subtropical mountain system
9695
4213
3012481
116266
177782
285763
726
1155
2571
5243
0.3
1.5
1.7
1.8
0.9
2.6
2.5
EAS Boreal tundra woodland
8300
286
1081882
150336
86494
15996
4393
1951
1432
524
0.6
1.5
1.9
3.3
0.3
2.2
2.2
SAM Subtropical dry forest
8256
10797
57044
5787
6298
30265
49
84
269
7854
8.3
19.4
22.2
26.0
29.5
51.7
42.5
AUS Subtropical humid forest
6488
5983
130692
22580
30269
90635
82
206
394
5806
2.4
4.5
5.1
6.4
4.9
10.1
8.5
AUS Temperate oceanic forest
5439
5786
108801
6052
13996
82774
24
54
216
5146
2.6
5.3
5.5
6.2
6.0
11.5
9.6
AFR Subtropical mountain system
5180
5137
390362
9522
6115
5316
295
536
1898
2451
1.3
23.3
38.0
46.1
44.9
83.0
58.1
AUS Subtropical dry forest
4234
3902
81757
17375
10024
12111
417
1037
886
1894
3.5
9.7
12.6
15.6
17.6
30.2
25.4
AUS Temperate mountain system
4220
2269
94608
14214
21282
60561
35
68
257
3860
2.2
4.4
5.0
6.4
2.8
7.8
6.8
SAM Temperate oceanic forest
3292
3510
107543
8332
23900
90484
41
15
45
3192
1.4
2.6
2.8
3.5
3.1
5.9
4.9
NAM Tropical dry forest
3177
870
142817
31827
25026
22140
283
616
1151
1127
1.4
3.7
4.8
5.1
1.8
6.7
6.6
NAM Temperate oceanic forest
2854
2317
14521
1511
1900
20962
7
12
35
2801
7.3
11.7
12.4
13.4
10.1
22.5
20.1
NAM Tropical mountain system
2832
522
123244
30126
31947
72471
113
236
448
2035
1.1
2.0
2.4
2.8
0.5
2.9
2.8
AFR Subtropical dry forest
1864
880
308429
10588
7605
8179
224
341
526
772
0.6
6.2
8.2
9.4
5.6
13.8
12.9
NAM Subtropical dry forest
1717
723
67075
3801
3009
12153
128
170
238
1181
2.0
8.4
9.4
9.7
4.8
14.1
13.0
AFR Subtropical humid forest
1556
1229
58127
14101
8704
3540
74
160
517
805
1.8
5.6
10.8
22.7
10.0
20.8
15.0
AUS Tropical mountain system
751
317
7778
3977
7622
101354
4
9
46
691
0.6
0.7
0.7
0.7
0.3
1.0
0.9
AUS Tropical dry forest
707
379
414034
31011
7067
8741
312
222
82
91
0.2
0.8
1.1
1.0
2.4
3.5
2.1
AUS Tropical moist deciduous
forest
375
94
10919
7706
10154
25954
4
7
46
317
0.7
0.8
1.0
1.2
0.3
1.3
1.2
SAM Temperate mountain system
248
142
52497
1603
3662
17420
15
3
8
221
0.3
1.0
1.1
1.3
0.7
1.8
1.7
SAM Subtropical mountain system
147
119
226487
2143
1706
7185
8
3
7
129
0.1
1.3
1.5
1.8
1.3
2.9
2.6
Table S3. Country tree cover extent, loss and gain summary statistics (km2), ranked by total loss.
Country
a)
Total
loss
b)
Total
gain
c)
Treecover 2000
Loss within treecover
Total loss
/ total
land area
(excluding
water) (%)
l)
>25% tree
cover loss
/ year
2000
>25% tree
cover (%)
m)
>50% tree
cover loss
/ year
2000
>50% tree
cover (%)
n)
>75% tree
cover loss
/ year
2000
>75% tree
cover (%)
o)
Total gain
/ year
2000
>50% tree
cover (%)
p)
>50% loss
+ total
gain /
2000
>50% tree
cover (%)
q)
Previous
column less
double
counting
pixels with
both loss
and gain (%)
r)
<25%
d)
26-50%
e)
51-75%
f)
76-100%
g)
<25%
h)
26-50%
i)
51-75%
j)
76-100%
k)
Russia
365015
162292
8414687
1846554
2707260
3304608
43907
62789
88346
169972
2.2
4.1
4.3
5.1
2.7
7.0
7.0
Brazil
360277
75866
3118579
509317
464530
4292417
23699
30816
43577
262185
4.3
6.4
6.4
6.1
1.6
8.0
7.5
United States
263944
138082
6294153
424662
462709
1982786
9274
17956
30306
206408
2.9
8.9
9.7
10.4
5.6
15.3
13.4
Canada
263943
91071
4141357
1068282
1074635
2167262
6860
44780
46295
166007
3.1
6.0
6.5
7.7
2.8
9.4
9.1
Indonesia
157850
69701
252964
68334
141996
1411892
1558
1611
8887
145795
8.4
9.6
10.0
10.3
4.5
14.4
12.4
China
61130
22387
7565646
387994
694764
618901
2250
3955
17457
37469
0.7
3.5
4.2
6.1
1.7
5.9
5.5
DRCongo
58963
13926
175163
469228
453121
1190506
917
4863
12362
40821
2.6
2.7
3.2
3.4
0.8
4.1
3.8
Australia
58736
14142
7209820
209287
79484
177890
28168
17318
2335
10915
0.8
6.6
5.1
6.1
5.5
10.6
9.1
Malaysia
47278
25798
32061
5538
17263
272365
289
309
1818
44862
14.4
15.9
16.1
16.5
8.9
25.0
20.6
Argentina
46958
6430
2352414
170266
125227
105950
5039
14574
14378
12966
1.7
10.4
11.8
12.2
2.8
14.6
13.9
Paraguay
37958
510
146165
110139
64482
75473
1230
12592
11751
12385
9.6
14.7
17.2
16.4
0.4
17.6
17.6
Bolivia
29867
1736
424460
75142
96512
478387
464
1340
4295
23768
2.8
4.5
4.9
5.0
0.3
5.2
5.1
Sweden
25533
15281
130774
42494
85855
153996
95
302
1790
23346
6.2
9.0
10.5
15.2
6.4
16.9
16.6
Colombia
25193
5516
302224
41544
61240
721695
315
619
1943
22315
2.2
3.0
3.1
3.1
0.7
3.8
3.7
Mexico
23862
6333
1391412
131003
126317
294988
1919
2013
4042
15887
1.2
4.0
4.7
5.4
1.5
6.2
6.0
Mozambique
21552
1446
403453
266480
101514
3414
3078
10181
7818
476
2.8
5.0
7.9
13.9
1.4
9.3
8.8
Tanzania
19903
3041
544450
243832
84708
12320
5111
7880
5740
1173
2.2
4.3
7.1
9.5
3.1
10.3
9.5
Finland
19516
10849
85929
36353
78775
104264
74
263
1684
17496
6.4
8.9
10.5
16.8
5.9
16.4
16.3
Angola
19320
638
612241
314210
281472
37883
4423
6556
6579
1762
1.6
2.4
2.6
4.7
0.2
2.8
2.8
Peru
15288
1910
498646
14200
23555
744591
80
101
245
14863
1.2
1.9
2.0
2.0
0.2
2.2
2.1
Myanmar
14958
3149
227580
37863
123465
274434
556
896
3991
9514
2.3
3.3
3.4
3.5
0.8
4.2
3.9
Cote d'Ivoire
14889
2298
139172
116351
58468
5111
2893
3830
6953
1213
4.7
6.7
12.8
23.7
3.6
16.5
15.5
Madagascar
14659
4051
402077
64144
63773
58637
1382
2490
4900
5888
2.5
7.1
8.8
10.0
3.3
12.1
10.8
Zambia
13163
181
419962
224188
91539
429
3460
5562
3918
223
1.8
3.1
4.5
52.0
0.2
4.7
4.7
Venezuela
12958
1910
328461
33724
43779
493663
695
993
2389
8881
1.4
2.1
2.1
1.8
0.4
2.5
2.4
Cambodia
12595
1096
86064
16785
16401
58431
484
748
2478
8884
7.1
13.2
15.2
15.2
1.5
16.6
16.1
Vietnam
12289
5643
152816
22284
40869
106971
660
637
2247
8744
3.8
6.8
7.4
8.2
3.8
11.3
10.5
Laos
12084
3379
34908
11904
36244
144908
384
422
1861
9417
5.3
6.1
6.2
6.5
1.9
8.1
7.2
Thailand
12049
4992
301598
25899
69216
110454
1402
1050
2985
6612
2.4
5.2
5.3
6.0
2.8
8.1
7.7
Chile
11879
14611
546019
17728
36107
142031
112
107
331
11329
1.6
6.0
6.5
8.0
8.2
14.7
12.2
Nigeria
10239
603
772371
82016
44023
3125
5987
2670
1435
147
1.1
3.3
3.4
4.7
1.3
4.6
4.4
South Africa
9526
8313
1145626
39602
20818
10826
678
1074
3365
4411
0.8
12.4
24.6
40.7
26.3
50.8
36.2
India
8971
2549
2703530
112692
136677
167827
810
855
2117
5189
0.3
2.0
2.4
3.1
0.8
3.2
3.0
Guatemala
8883
1094
29734
8709
11952
57571
105
323
1097
7357
8.2
11.2
12.2
12.8
1.6
13.7
13.4
Nicaragua
8225
662
39402
8527
12384
58289
109
230
650
7236
6.9
10.2
11.2
12.4
0.9
12.1
12.0
France
7664
5062
373705
19842
37021
115684
209
330
1149
5977
1.4
4.3
4.7
5.2
3.3
8.0
7.6
Spain
6908
4482
386553
36472
28533
52468
676
982
1698
3552
1.4
5.3
6.5
6.8
5.5
12.0
11.4
New Zealand
6883
7102
149606
5653
14375
93838
24
22
126
6711
2.6
6.0
6.3
7.2
6.6
12.9
10.7
Papua New Guinea
6337
2308
27933
11647
23295
396660
29
35
197
6076
1.4
1.5
1.5
1.5
0.5
2.0
1.9
Philippines
6227
2726
102570
15788
33051
141108
84
90
485
5569
2.1
3.2
3.5
3.9
1.6
5.0
4.7
Poland
5829
5041
201808
11013
30655
64857
79
163
814
4773
1.9
5.4
5.8
7.4
5.3
11.1
10.7
Ukraine
5657
3529
470946
19510
28312
68474
150
234
900
4372
1.0
4.7
5.4
6.4
3.6
9.1
8.8
Ghana
5406
1345
153157
36659
40464
2074
911
1099
2863
533
2.3
5.7
8.0
25.7
3.2
11.1
10.3
Ecuador
5246
1027
62024
12832
20737
158764
58
118
373
4697
2.1
2.7
2.8
3.0
0.6
3.4
3.3
Portugal
4987
2866
64476
8726
6753
9147
459
784
1246
2497
5.6
18.4
23.5
27.3
18.0
41.6
36.1
Germany
4890
2585
226231
10219
25175
91960
61
79
457
4294
1.4
3.8
4.1
4.7
2.2
6.3
6.1
Honduras
4860
582
32713
11870
14297
52664
84
238
560
3978
4.4
6.1
6.8
7.6
0.9
7.6
7.6
Cameroon
4816
651
120385
90110
77877
174702
548
691
1595
1981
1.0
1.2
1.4
1.1
0.3
1.7
1.6
Mongolia
4779
103
1508028
26031
15552
3080
857
1514
1485
922
0.3
8.8
12.9
29.9
0.6
13.5
13.5
Central African Republic
4719
395
101812
209547
238922
68765
226
910
1869
1714
0.8
0.9
1.2
2.5
0.1
1.3
1.3
Japan
4303
2570
102902
9616
22119
233863
71
57
253
3924
1.2
1.6
1.6
1.7
1.0
2.6
2.5
Belarus
4167
3755
112240
8835
25820
57913
26
89
547
3504
2.0
4.5
4.8
6.1
4.5
9.3
9.1
Latvia
4120
1857
27713
2742
7086
26126
10
32
253
3825
6.5
11.4
12.3
14.6
5.6
17.9
17.6
Liberia
3955
1084
1221
3783
54435
36117
7
96
2305
1547
4.1
4.2
4.3
4.3
1.2
5.5
4.9
Guinea
3933
296
130230
91912
19736
1821
1251
1604
923
155
1.6
2.4
5.0
8.5
1.4
6.4
6.1
Zimbabwe
3869
486
362829
21175
2132
799
2284
1018
353
214
1.0
6.6
19.3
26.8
16.6
35.9
29.2
Uganda
3654
685
105539
67065
24230
8489
154
703
1118
1679
1.8
3.5
8.5
19.8
2.1
10.6
10.2
Norway
3520
1729
187652
21453
35282
62297
18
37
214
3252
1.1
2.9
3.6
5.2
1.8
5.3
5.3
Turkey
3426
1783
664081
24543
23040
59104
535
488
706
1697
0.4
2.7
2.9
2.9
2.2
5.1
4.7
Benin
3307
69
108992
5962
81
8
2835
405
59
8
2.9
7.8
75.3
100.0
77.5
152.8
143.8
Chad
3306
1
1257866
9549
72
0
2914
382
10
0
0.3
4.1
13.9
-
1.4
15.3
15.3
Kenya
3059
1005
530692
20815
9271
9636
732
587
612
1129
0.5
5.9
9.2
11.7
5.3
14.5
13.8
Republic of Congo
2993
467
52860
46265
25236
214202
101
272
803
1819
0.9
1.0
1.1
0.8
0.2
1.3
1.2
Ethiopia
2821
625
968731
100537
33866
20399
375
947
828
671
0.3
1.6
2.8
3.3
1.2
3.9
3.8
United Kingdom
2689
2111
203651
8793
14394
15185
84
167
478
1960
1.1
6.8
8.2
12.9
7.1
15.4
14.9
Panama
2675
323
16563
3089
4854
49687
35
69
262
2308
3.6
4.6
4.7
4.6
0.6
5.3
5.2
Romania
2307
1530
154828
7088
10939
62836
20
25
103
2158
1.0
2.8
3.1
3.4
2.1
5.1
5.0
Estonia
2179
894
16601
2052
5085
19569
5
11
126
2036
5.0
8.1
8.8
10.4
3.6
12.4
12.3
Uruguay
2027
4985
157077
3568
5359
8522
26
73
282
1646
1.2
11.5
13.9
19.3
35.9
49.8
45.2
Austria
2015
658
39299
2743
7205
33933
11
22
130
1853
2.4
4.6
4.8
5.5
1.6
6.4
6.3
Burkina Faso
1993
0
274158
11
0
0
1987
5
0
0
0.7
45.5
-
-
-
-
-
Sierra Leone
1967
451
11320
24844
33463
2424
24
240
1231
472
2.7
3.2
4.7
19.5
1.3
6.0
5.6
Dominican Republic
1929
393
21365
4064
4619
17744
72
120
276
1462
4.0
7.0
7.8
8.2
1.8
9.5
9.4
Gabon
1891
391
11898
8112
10885
230775
27
110
285
1469
0.7
0.7
0.7
0.6
0.2
0.9
0.8
Lithuania
1845
1226
40296
1889
5303
16472
9
20
160
1655
2.9
7.8
8.3
10.0
5.6
14.0
13.7
Cuba
1725
2271
68008
4982
7388
28775
90
126
295
1214
1.6
4.0
4.2
4.2
6.3
10.5
10.4
Mali
1694
0
1247103
1007
3
0
1650
40
3
0
0.1
4.3
100.0
-
0.0
100.0
100.0
Costa Rica
1653
382
11327
2752
5663
31183
28
68
200
1356
3.2
4.1
4.2
4.3
1.0
5.3
5.1
Czech Republic
1646
1331
46934
2445
6429
22264
14
25
197
1410
2.1
5.2
5.6
6.3
4.6
10.2
10.0
South Sudan
1635
38
460581
128358
39278
1773
567
629
356
83
0.3
0.6
1.1
4.7
0.1
1.2
1.1
North Korea
1605
137
67695
12808
31773
9164
96
262
837
411
1.3
2.8
3.0
4.5
0.3
3.4
3.4
Italy
1603
898
201331
13199
19020
64805
113
105
244
1142
0.5
1.5
1.7
1.8
1.1
2.7
2.5
Greece
1566
356
91341
10219
8443
21132
188
181
311
886
1.2
3.5
4.0
4.2
1.2
5.3
5.1
South Korea
1463
271
43787
9776
28496
16694
230
215
560
458
1.5
2.2
2.3
2.7
0.6
2.9
2.8
Malawi
1290
103
71949
19030
2967
163
399
540
309
43
1.4
4.0
11.2
26.4
3.3
14.5
13.8
Slovakia
1237
523
24608
1245
2814
20170
6
11
67
1153
2.5
5.1
5.3
5.7
2.3
7.6
7.4
Belize
1206
128
4073
437
675
16434
5
10
35
1155
5.6
6.8
7.0
7.0
0.7
7.7
7.5
Hungary
1107
1350
71070
2658
3705
14263
12
19
84
992
1.2
5.3
6.0
7.0
7.5
13.5
13.0
Sri Lanka
985
264
24684
4952
9285
26177
64
77
294
551
1.5
2.3
2.4
2.1
0.7
3.1
3.0
Guyana
915
114
18733
1096
1319
187681
4
6
14
890
0.4
0.5
0.5
0.5
0.1
0.5
0.5
Senegal
832
2
192538
1723
20
1
806
23
3
0
0.4
1.5
14.3
0.0
9.5
23.8
23.8
Kazakhstan
828
239
2628744
13973
13954
17598
329
196
125
178
0.0
1.1
1.0
1.0
0.8
1.7
1.7
Bulgaria
779
678
68662
3910
6653
32070
18
20
71
670
0.7
1.8
1.9
2.1
1.8
3.7
3.4
Ireland
778
1238
60220
2358
3492
2899
26
38
122
592
1.1
8.6
11.2
20.4
19.4
30.5
29.0
Togo
768
24
48707
6809
1270
5
383
333
50
2
1.4
4.8
4.1
40.0
1.9
6.0
5.8
Swaziland
747
603
11310
3962
1386
597
48
84
204
412
4.3
11.8
31.1
69.0
30.4
61.5
41.9
Algeria
743
325
2294657
4212
3989
4967
67
136
211
329
0.0
5.1
6.0
6.6
3.6
9.7
9.5
Suriname
724
70
4791
420
567
138564
2
4
9
708
0.5
0.5
0.5
0.5
0.1
0.6
0.5
Guinea-Bissau
676
65
18081
11875
3106
35
164
315
186
11
2.0
3.4
6.3
31.4
2.1
8.3
7.9
Solomon Islands
630
203
528
122
441
26825
3
1
5
621
2.3
2.3
2.3
2.3
0.7
3.0
2.8
Belgium
601
373
21609
1139
2011
5756
7
11
59
524
2.0
6.7
7.5
9.1
4.8
12.3
11.9
El Salvador
567
86
9961
2231
3309
4710
25
56
206
280
2.8
5.3
6.1
5.9
1.1
7.1
7.1
Bangladesh
543
70
108675
5560
8449
6965
15
23
116
389
0.4
2.5
3.3
5.6
0.5
3.7
3.7
Denmark
533
322
35730
1281
2506
3105
8
13
68
444
1.3
7.6
9.1
14.3
5.7
14.9
14.5
Croatia
454
265
31487
2890
3396
18892
53
35
40
326
0.8
1.6
1.6
1.7
1.2
2.8
2.8
French Guiana
441
42
928
110
216
81317
1
1
4
435
0.5
0.5
0.5
0.5
0.1
0.6
0.6
Equatorial Guinea
439
56
199
251
1864
24513
2
12
76
349
1.6
1.6
1.6
1.4
0.2
1.8
1.7
Nepal
434
134
94352
9930
21761
21318
67
65
106
195
0.3
0.7
0.7
0.9
0.3
1.0
1.0
Jamaica
329
68
3185
479
753
6545
7
9
28
285
3.0
4.1
4.3
4.4
0.9
5.2
5.2
Morocco
315
196
405884
2870
2251
2110
58
65
85
107
0.1
3.6
4.4
5.1
4.5
8.9
8.8
Albania
311
74
21128
1719
1711
3642
20
25
55
212
1.1
4.1
5.0
5.8
1.4
6.4
6.3
Macedonia
296
104
16148
1234
1600
5359
11
11
31
244
1.2
3.5
4.0
4.6
1.5
5.4
5.0
Haiti
286
48
17810
2494
2238
4334
19
42
76
148
1.1
2.9
3.4
3.4
0.7
4.1
4.1
Serbia
267
356
49076
3338
5852
19132
5
4
13
245
0.3
0.9
1.0
1.3
1.4
2.5
2.4
Taiwan
267
61
12174
1158
2368
20098
6
7
27
227
0.7
1.1
1.1
1.1
0.3
1.4
1.4
Switzerland
227
104
24221
1255
3120
11363
3
3
19
201
0.6
1.4
1.5
1.8
0.7
2.2
2.2
Burundi
204
36
16600
6956
1189
222
43
86
65
11
0.8
1.9
5.4
5.0
2.6
7.9
7.7
Bosnia and Herzegovina
198
265
23549
2532
4225
20546
26
15
38
120
0.4
0.6
0.6
0.6
1.1
1.7
1.7
Fiji
194
119
2205
1834
1857
12339
1
2
17
174
1.1
1.2
1.3
1.4
0.8
2.2
2.1
East Timor
185
61
7265
1467
1871
4296
4
5
20
156
1.2
2.4
2.9
3.6
1.0
3.8
3.7
Rwanda
178
71
16805
4966
1171
889
22
65
63
27
0.7
2.2
4.4
3.0
3.4
7.8
7.4
Brunei
171
88
434
68
157
5066
2
1
7
161
3.0
3.2
3.2
3.2
1.7
4.9
4.3
Netherlands
166
71
28258
1503
1896
2866
6
8
32
121
0.5
2.6
3.2
4.2
1.5
4.7
4.7
Slovenia
162
35
6851
612
1304
11166
3
3
16
140
0.8
1.2
1.3
1.3
0.3
1.5
1.5
Trinidad and Tobago
154
16
1188
111
168
3664
2
2
9
140
3.0
3.8
3.9
3.8
0.4
4.3
4.2
Puerto Rico
141
64
3591
397
539
4381
10
7
19
105
1.6
2.5
2.5
2.4
1.3
3.8
3.8
Bhutan
129
22
13772
2471
7040
16652
9
13
35
73
0.3
0.5
0.5
0.4
0.1
0.5
0.5
Namibia
128
0
822966
122
5
1
104
21
3
0
0.0
18.8
50.0
0.0
0.0
50.0
50.0
New Caledonia
125
57
3883
4303
3261
7254
7
24
37
57
0.7
0.8
0.9
0.8
0.5
1.4
1.4
Gambia
111
0
10221
213
0
0
99
11
0
0
1.1
5.2
-
-
-
-
-
Tunisia
103
115
152233
568
712
1074
13
12
20
58
0.1
3.8
4.4
5.4
6.4
10.8
10.6
Pakistan
100
8
861788
4077
3320
3349
9
11
35
46
0.0
0.9
1.2
1.4
0.1
1.3
1.3
Bahamas
95
8
8628
571
575
1995
3
6
10
75
0.8
2.9
3.3
3.8
0.3
3.6
3.6
Syria
91
16
184360
319
366
455
10
10
17
54
0.0
7.1