Content uploaded by Ruth S. Defries
Author content
All content in this area was uploaded by Ruth S. Defries on Jul 30, 2014
Content may be subject to copyright.
Content uploaded by Ruth S. Defries
Author content
All content in this area was uploaded by Ruth S. Defries on Jul 02, 2014
Content may be subject to copyright.
LETTERS
PUBLISHED ONLINE: 7 FEBRUARY 2010 | DOI: 10.1038/NGEO756
Deforestation driven by urban population growth
and agricultural trade in the twenty-first century
Ruth S. DeFries
1
*
, Thomas Rudel
2
, Maria Uriarte
1
and Matthew Hansen
3
Reducing atmospheric carbon emissions from tropical defor-
estation is at present considered a cost-effective option for
mitigating climate change. However, the forces associated with
tropical forest loss are uncertain
1
. Here we use satellite-based
estimates of forest loss for 2000 to 2005 (ref. 2) to assess
economic, agricultural and demographic correlates across 41
countries in the humid tropics. Two methods of analysis—
linear regression and regression tree—show that forest loss
is positively correlated with urban population growth and
exports of agricultural products for this time period. Rural
population growth is not associated with forest loss, indicating
the importance of urban-based and international demands for
agricultural products as drivers of deforestation. The strong
trend in movement of people to cities in the tropics is, counter-
intuitively, likely to be associated with greater pressures for
clearing tropical forests. We therefore suggest that policies
to reduce deforestation among local, rural populations will
not address the main cause of deforestation in the future.
Rather, efforts need to focus on reducing deforestation for
industrial-scale, export-oriented agricultural production, con-
comitant with efforts to increase yields in non-forested lands
to satisfy demands for agricultural products.
Maintaining carbon stocks in tropical forests is widely recog-
nized as a relatively low-cost option for mitigating climate change
3,4
with ancillary benefits for biodiversity, regulating precipitation
and a host of other ecosystem services
5
. The United Nations
framework convention on climate change is considering whether
reduced emissions from avoided deforestation and degradation
(REDD) will be included as an allowable reduction strategy dur-
ing the second, post-2012 commitment period. Voluntary carbon
markets are also developing standards for incorporating REDD
activities
6
. These initiatives will be effective in reducing defor-
estation only if they address the factors that promote forest loss.
Yet demographic and economic factors associated with defor-
estation remain poorly understood, partly because these complex
factors vary throughout the tropics and over time
7
, and partly
because consistent and reliable data have not been available for
cross-country analyses
1
.
Major demographic and economic shifts are sweeping across
many places in the tropics (Fig. 1). Population growth rates are
slowing overall, but urban growth is vastly outpacing rural growth
8
.
In the next 20 years, 22 of the 41 countries included in this analysis
are projected to have fewer rural inhabitants than they do today,
whereas all of the countries have increasing urban populations. In
nine of the countries, urban populations are projected to more
than double in the next 20 years
9
. Many countries throughout the
tropics are also on a trajectory of increasing agricultural exports.
The question we address in this study is the likely impact of
1
Department of Ecology, Evolution, and Environmental Biology, Columbia University New York, New York 10027, USA,
2
Departments of Human Ecology
and Sociology, Rutgers University, Piscataway, New Jersey 08854, USA,
3
Geographic Information Science Center of Excellence, South Dakota State
University, Brookings, South Dakota 57007, USA. *e-mail: rd2402@columbia.edu.
these changing demographic and economic factors on pressures
to clear tropical forests.
Debates on whether these trends will reduce or increase
pressure to clear more forests are unresolved. Some argue that
depopulating rural landscapes will reduce pressures on forests
10
.
Others argue that reducing demand for agricultural lands in rural
areas coupled with increasing yields can result in land-sparing
11
.
Others implicate industrial-scale, mechanized agriculture as the
next wave of deforestation following the planned settlements and
small-scale farmers in previous decades
12
. Reliable data on changes
in forest area have not been available to quantify factors associated
with past changes in forest area to resolve these debates.
We use a newly available, spatially explicit analysis of forest
loss (not including regrowth) in the humid tropical forest biome
based on extensive samples of high-resolution satellite observations
(30 m) and regression estimators to extrapolate to lower resolution
(500 m) data
2
. We assess demographic and economic factors
associated with forest loss for 41 countries across the humid tropics
(see Supplementary Table S1 for criteria and list of countries). These
countries collectively cover 98% of all forest area in the humid
tropical forest biome
13
. For each data set we apply two methods,
multiple linear regression and regression trees
14
, to identify relevant
factors. The aim is to identify factors most strongly associated
with forest loss, rather than to predict forest loss per se. We
examined ten possible correlates: four demographic factors for
2000–2005 (urban, rural and total annual population growth and
per cent of population that is urban), four factors related to
agricultural production (net agricultural trade per capita, per cent
of agricultural production that is exported, agricultural exports per
unit production and per cent forest remaining in biome) and two
economic factors (gross domestic product per capita and annual
gross domestic product growth) (see Supplementary Table S2 for
sources and the Methods section for selection of variables). All pairs
of the variables used in the analysis have tolerance values greater
than 0.5 (Supplementary Table S3) and variables were tested for
multicollinearity (Supplementary Tables S4 and S5).
Both methods show a positive association of forest loss with
urban growth and agricultural exports. The most highly significant
factors associated with satellite-derived forest loss are urban
growth rate (p < 0.001) and net agricultural trade per capita
(p < 0.001) (Table 1 and Supplementary Table S2). Although these
associations do not prove causality, the positive correlations do
suggest that the traditional mode of clearing in frontier landscapes
for small-scale production to support subsistence needs or local
markets is no longer the dominant driver of deforestation in
many places
15–17
. Rather, our analysis indicates that higher rates of
forest loss for 2000–2005 are strongly associated with demands for
agricultural products in distant urban and international locations.
178 NATURE GEOSCIENCE | VOL 3 | MARCH 2010 | www.nature.com/naturegeoscience
© 2010 Macmillan Publishers Limited. All rights reserved.
NATURE GEOSCIENCE DOI: 10.1038/NGEO756
LETTERS
1980 1990 2000 2010
Value of agricultural exports
(US$ ) (1980 index = 1)
Africa
Asia
Latin America
Urban population (1980 index = 1)
Rural population (1980 index = 1)
0
1
2
3
4
0
2
4
6
8
1980 1990 2000 2010 2020
2030
0
2
4
6
8
1980 1990 2000
Year
Year
Year
2010 2020 2030
a
c
b
Figure 1 | Trends in agricultural exports, urban population and rural
population for the countries in this study. a–c, Trends in agricultural
exports (a), urban population (b) and rural population (c). The values are
indexed to 1980 for 41 countries included in this study aggregated by
continent. The per cent of the total population that is urban was 28, 21 and
64 in 1980 and is projected (medium variant) to be 59, 52 and 84 in 2030
in the African, Asian and Latin American countries included in the study
respectively. Demographic data are from ref. 9 and export data are from
ref. 30. See Supplementary Table S1 for list of countries.
Total population growth is mildly significant (p < 0.01), with a
negative coefficient refuting the claim that increasing population
causes deforestation. Rural population growth rates related to local
demand are not significantly correlated.
The regression tree provides a complementary, non-parametric
approach
14
and is used here to test the robustness of results from
the linear regression. The regression tree confirms the significance
of agricultural exports and urban growth as national-scale drivers of
forest loss in 2000–2005 (Fig. 2). The same variables significant in
the linear regression were selected by the regression tree. As with the
linear regression, the variables associated with high forest loss are
associated with demands for agriculture in urban and international
markets rather than rural population dynamics.
Net agricultural trade is the first split in the regression tree,
meaning that it is the most powerful discriminator between
countries with relatively high and low forest loss. The highest forest
Table 1 | Results of ordinary least-squares regression for
annual forest loss for 2000–2005 (ref. 2).
Variable Coefficient
Intercept 0.031 (0.003)*
Annual urban growth rate (2000–2005) 0.016 (0.004)*
Total annual growth rate (2000–2005) − 0.010 (0.004)
†
Net agricultural trade per capita (2003–2004) 0.008 (0.003)*
% of agricultural production exported 0.007 (0.004)
‡
R
2
0.52
Adjusted R
2
0.47*
Values are standardized coefficients with standard error in parentheses. Arcsine transformation
was carried out on the dependent variable. See Supplementary Table S2 for results with all
variables used in the analysis, and data sources, and see Supplementary Tables S3, S4 and S5
for multicollinearity diagnostics. n = 41.
* p < 0.001
†
p < 0.01
‡
p < 0.10
loss occurs in those countries with relatively high agricultural trade
and high urban growth (right-most node in the tree), although
these countries comprise only 2% of the remaining forest area
(see Supplementary Table S6). Almost 60% of the remaining forest
area occurs where net agricultural trade, per cent of agricultural
products exported and urban growth are all relatively low (the three
left-most nodes in the tree). As demands for agricultural products
grow in these countries, this large remaining forest area is likely to
experience increasing pressures.
Variables associated with forest loss vary within and between
regions
7
. Urban population growth, the most significant variable
in our analysis, is positively associated with forest loss in all three
regions of Africa, Asia and Latin America (Fig. 3) although to
varying degrees. In African countries in which forest loss is relatively
low, the association with urban growth is less pronounced (lower
slope) than in Latin American and Asian countries. Agricultural
trade is not significantly associated with forest loss for African
or Latin American countries, whereas it is significant (p < 0.02)
for Asian countries. These differences suggest varying pressures
operating unevenly across the regions.
The results build on previous cross-national studies by incorpo-
rating satellite-based data from the current decade. Previous studies
have been plagued by poor data quality
1,18
and have generally relied
on country-level data of net change in forest area
19
. This analysis
indicates the value of satellite-derived data on gross forest loss
(Supplementary Fig. S1). Country-level data of net change yielded
no significant correlations with the economic and demographic
variables used in this study.
Comparison of these results with previous studies indicates
a transition towards increasing influence of urbanization and
international trade as pressures on tropical forests. Most previous
studies do not consider distal demands for agricultural products as
drivers of deforestation. Those analyses that distinguish between
urban and rural population growth generally find a positive
association with rural growth rates and a negative association
with urban growth rates for the 1980s and 1990s (refs
20, 21),
suggesting that urban to rural migration reduced pressure on
forests in this time period. We conclude the opposite for 2000 to
2005. Collectively, these results indicate a shift from state-run road
building and colonization in the 1970s and 1980s to enterprise-
driven deforestation in later decades
22
.
There are two major implications of these results. First,
trajectories towards continuing urban growth and agricultural
exports in tropical countries (Fig. 1) are at present associated
with major pressures on forests. We cannot distinguish from this
NATURE GEOSCIENCE | VOL 3 | MARCH 2010 | www.nature.com/naturegeoscience 179
© 2010 Macmillan Publishers Limited. All rights reserved.
LETTERS
NATURE GEOSCIENCE DOI: 10.1038/NGEO756
N
Y
N Y
Y (0.24, n = 7)
Trade > 0.15?
Y (0.50, n = 4)
N
Y N
Y
0.01
(<0.01)
Export > 0.175?
Trade > 0.115?
Urban > 4.7?
Urban > 3.55?
N (0.11, n = 37)
N (0.11, n = 37)
(0.33, n = 5)
N (0.07, n = 28)
N (0.11, n = 10)
Y
0.26
n = 2
n = 18
n = 2 n = 2
0.46
n = 3
0.15
(0.03)
n = 2
0.74
(0.07)
n = 2
0.26
(0.03)
n = 2
0.04
0.34
(0.06)
Urban > 3.15?
Y
Total > 1.625?
Trade > ¬0.015?
(0.02)
(<0.01)
n = 8
(0.02)
(<0.01)
0.07
Figure 2 | Regression tree derived from 10 demographic, agricultural and
economic variables for countries in the study. Each split lists the mean
rate of annual forest loss for 2000–2005 and the number of countries. The
hexagons are terminal nodes with mean forest loss and deviance in
parentheses. The thickness of the lines represents the relative amount of
remaining humid tropical forest in each node (see Supplementary Table S6
for countries and forest area in each node). The residual mean deviance is
0.007. Y: yes, N: no. Trade: net agricultural trade per capita, export: per cent
of agricultural production exported, urban: urban population growth rate,
total: total population growth. See Supplementary Table S2 for derivation
of variables.
national-scale study whether urban growth and agricultural exports
lead directly to forest conversion, or indirectly through conversion
of non-forested lands that displaces other land uses into forested
areas. The pressures from these direct and indirect forces probably
vary among countries and require country-specific analysis to
develop effective approaches to reduce deforestation. Nevertheless,
in the absence of incentives to counter these pressures, future
pressures will increase on remaining forest areas where pressures
are low at present (Fig. 2).
Landscapes with stabilizing or depopulating populations are
likely to exacerbate rather than reduce pressure to clear forests,
primarily because urbanization raises consumption levels and
increases demand for agricultural products. Urban consumers
generally eat more processed foods and animal products than rural
consumers, thereby inducing commercial production of crops and
livestock
23,24
. This pattern contradicts the argument that pressures
on forests will decline as local populations urbanize
10
. Nearly all
population growth in the coming decades is projected to occur
in urban rather than rural areas
8
, which will place demands on
rural landscapes for commercial food production. Competing land
uses for other products such as biofuels
25,26
will exacerbate these
pressures on tropical forests.
Second, demands for agricultural production and pressures to
clear tropical forests are closely linked. Policies such as REDD aim
to reduce deforestation through incentives to maintain standing
forests. Such policies assume that the demand for agricultural
production will be fulfilled elsewhere. Approaches to meet the
dual goals of maintaining forest carbon and increasing agricultural
production are needed as countries face pressures to clear
more forests. Such approaches include incentives to maximize
agricultural production on already-cleared lands while minimizing
new forest clearing
27
.
Urban growth and agricultural exports are positively and
significantly correlated with forest loss, and tropical forests will
continue to face large pressures as urban-based and international
Forest loss (% yr
¬1
) 2000¬2005
¬0.2 ¬0.1
Forest loss (% yr
¬1
) 2000¬2005
Africa
Asia
Latin America
¬1 0 1 2 3 4 5 6 7
0
0.2
0.4
0.6
0.8
1.0
Annual urban population growth 2000¬2005
0
0.2
0.4
0.6
0.8
1.0
Net agricultural trade per capita (US$/person x 1000) 2003¬2004
0 0.1 0.2 0.3 0.4
a
b
Figure 3 | Per cent forest loss versus annual urban growth and
agricultural trade by region. a,b, Per cent forest loss versus annual urban
growth (a) and net agricultural trade per capita (b) for countries in the
three tropical regions. n = 12, 11 and 15 for African, Asian and Latin
American countries respectively. p = 0.006, 0.06 and 0.05 respectively for
annual urban growth rate. Only regression for Asian countries is significant
(p = 0.02) for agricultural trade.
demands for agricultural products continue to increase. These
patterns underscore the challenges that policy instruments such as
REDD face in reducing deforestation. Policies need to be flexible
and recognize that causes of deforestation vary through time and
space. As demands for agricultural products accelerate in the future,
policies designed to increase agricultural yields on already-cleared
lands will have to accompany REDD policies to reduce forest
clearing as a climate mitigation strategy.
Methods
We used previously published satellite-derived estimates of forest loss for
2000–2005 for the humid tropics aggregated to the country level
2
. The satellite data
are derived from Landsat and MODIS data using regression estimators to estimate
forest loss for the humid tropical biome per 18.5 by 18.5 km block. We calculated
the area of forest loss by aggregating the blocks to the national scale. The rate of loss
is calculated relative to the country’s area within the tropical humid forest biome as
determined from ref. 13 (see Supplementary Table S1).
We examined many possible correlates with forest loss at a national scale,
including demographic, economic and agricultural production statistics. After
eliminating variables that were highly correlated or poor data quality (see Supple-
mentary Table S2 for list of variables, sources and derivations), we formally tested
for collinearity using a number of regression diagnostics including variance inflation
factors (Supplementary Table S4), and condition indexes (ratios of eigenvalues) and
variance decomposition proportions of the design matrix (Supplementary Table S5;
ref. 28). If the largest condition index is large (that is, 30 or higher), then there may
be collinearity problems. All condition indices were less than five.
For each of the data sets, we carried out least-squares regression and derived
regression trees using the R software package
29
with routines lm() and tree(). These
approaches are complementary and are used together to lend robustness to the
analysis. Unlike linear regression, regression trees are non-parametric, nonlinear
and make few assumptions about the underlying relationship between the response
variable and predictor
14
. We used four as the threshold for the minimum node size,
as determined from the threshold that leads to the largest reduction in residual
deviance. For the linear regression, independent variables were standardized and
the dependent variable underwent arcsine transformation. We ran the linear
180 NATURE GEOSCIENCE | VOL 3 | MARCH 2010 | www.nature.com/naturegeoscience
© 2010 Macmillan Publishers Limited. All rights reserved.
NATURE GEOSCIENCE DOI: 10.1038/NGEO756
LETTERS
regression with and without weighting by forest area in 2000. The same variables
were significant in both (Supplementary Table S2), except in the linear regression
with weighting one variable (per cent of agricultural production exported) was not
significant. We report results without weighting.
The data sets used in this study have several limitations. The satellite-derived
data set identifies gross forest loss for the humid tropics. The extent of abandonment
and regrowth, which can partially offset emissions from deforestation, is
not captured and remains a major uncertainty. Another limitation of the
satellite-derived data set is that it provides estimates of forest loss rather than
deforestation, and may be capturing plantation harvest, reclearing of secondary
regrowth or forest loss from wildfires in some locations rather than clearing
of primary forest. Finally, we cannot infer causal relationships based on the
associations between forest loss and the independent variables. It cannot be
determined, for example, whether the association between urban population
growth and forest loss results from a ‘pull’ of farmers to cities with economic
opportunities (thereby accelerating consumer demands for agricultural products)
or a ‘push’ as higher agricultural prices attract large enterprises that drive small
farmers off the land. These results indicate general patterns, but analyses of specific
circumstances in individual countries are required to identify effective policies to
reduce forest loss in different circumstances.
Received 17 August 2009; accepted 23 December 2009;
published online 7 February 2010
References
1. Grainger, A. Difficulties in tracking the long-term global trend in tropical
forest area. Proc. Natl Acad. Sci. USA 105, 818–823 (2008).
2. Hansen, M. C. et al. Humid tropical forest clearing from 2000 to 2005
quantified using multi-temporal and multi-resolution remotely sensed data.
Proc. Natl Acad. Sci. USA 105, 9439–9444 (2008).
3. Canadell, J. G. & Raupach, M. R. Managing forests for climate change
mitigation. Science 320, 1456–1457 (2008).
4. Stern, N. The Economics of Climate Change: The Stern Review (Cambridge
Univ. Press, 2007).
5. Hassan, R., Scholes, R. & Ash, N. (eds) in Ecosystems and Human Well-Being:
Current Status and Trends Vol. 1 (Island Press, 2005).
6. Voluntary Carbon Standard, Voluntary Carbon Standard Association,
Voluntary Carbon Standard 2007.1. <http://www.v-c-s.org/docs/
Voluntary%20Carbon%20Standard%202007_1.pdf> (2008).
7. Geist, H. J. & Lambin, E. F. Proximate causes and underlying forces of tropical
deforestation. BioScience 52, 143–150 (2002).
8. Montgomery, M. R. The urban transformation of the developing world. Science
319, 761–764 (2008).
9. United Nations, World Urbanization Prospects: The 2005 Revision.
(Department of Social and Economic Affairs, Population Division, 2006).
10. Wright, S. J. & Muller-Landau, H. C. The future of tropical forest species.
Biotropica 38, 287–301 (2006).
11. Green, R., Cornell, S., Scharlemann, J. & Balmford, A. Farming and the fate of
wild nature. Science 307, 550–555 (2005).
12. Rudel, T. K., DeFries, R., Asner, G. P. & Laurance, W. Changing drivers
of deforestation and new opportunities for conservation. Conserv. Biol. 23,
13961405 (2009).
13. Olson, D. et al. Terrestrial ecoregions of the world: A new map of life on Earth.
BioScience 51, 933–938 (2001).
14. Breiman, L., Friedman, J. H., Olshend, R. A. & Stone, C. J. Classification and
Regression Trees (Wadsworth, 1984).
15. Chomitz, K. At Loggerheads?: Agricultural Expansion, Poverty Reduction, and
Environment in the Tropical Forests (World Bank, 2006).
16. Nepstad, D. C., Stickler, C. M. & Almeida, O. T. Globalization of the Amazon
soy and beef industries: Opportunities for conservation. Conserv. Biol. 20,
1595–1602 (2006).
17. Rudel, T. K. Tropical Forests: Regional Paths of Destruction and Regeneration in
the Late Twentieth Century (Columbia Univ. Press, 2005).
18. Angelsen, A. & Kaimowitz, D. Rethinking the causes of deforestation: Lessons
from economic models. World Bank Res. Observ. 14, 73–98 (1999).
19. Food and Agriculture Organisation Global Forest Assessment 2005 (FAO
Forestry Paper 147, 2006).
20. Ehrhardt-Martinez, K., Crenshaw, E. & Jenkins, J. C. Deforestation and the
environmental Kuznets curve: A cross-national investigation of intervening
mechanisms. Soc. Sci. Q. 83, 226–243 (2002).
21. Jorgenson, A. & Burns, T. Effects of rural and urban population dynamics and
national development on deforestation in less-developed countries, 1990–2000.
Soc. Inquiry 77, 460–482 (2007).
22. Rudel, T. K. Changing agents of deforestation: From state-initiated to
enterprise-driven processes, 1970–2000. Land Use Policy 24, 35–41 (2007).
23. Kennedy, G., Nantel, G. & Shetty, P. Globalization of Food Systems in Developing
Countries: Impacts on Food Security and Nutrition 1–26 (FAO Food and
Nutrition Paper 83, 2004).
24. Mendez, M. A. & Popkin, B. M. Globalization, urbanization and nutrition
change in the developing world. J. Agric. Dev. Econ. 1, 220–241 (2004).
25. Fargione, J., Jason, H., Tilman, D., Polasky, S. & Hawthorne, P. Land clearing
and the biofuel carbon debt. Science 319, 1235–1238 (2008).
26. Wise, M. et al. Implications of limiting CO
2
concentrations for land use and
energy. Science 324, 1183–1186 (2009).
27. Matson, P. A. & Vitousek, P. M. Agricultural intensification: Will land
spared from farming be land spared for nature? Conserv. Biol. 20,
709–710 (2006).
28. Belsley, D. A., Kuh, E. & Welsch, R. E. Regression Diagnostics: Identifying
Influential Data and Sources of Collinearity (John Wiley, 2004).
29. R Development Core Team, R: A Language and Environment for Statistical
Computing (R Foundation for Statistical Computing, 2008).
30. Food and Agriculture Organisation FAO Statistical Yearbook 2005–2006 (FAO
Statistics Division, 2007).
Author contributions
R.S.D. conducted data analysis and drafted the manuscript; T.R. participated in data
analysis and writing; M.U. advised on statistical analysis and participated in writing; M.H.
guided the use of the forest cover data and participated in writing.
Additional information
The authors declare no competing financial interests. Supplementary information
accompanies this paper on www.nature.com/naturegeoscience. Reprints and permissions
information is available online at http://npg.nature.com/reprintsandpermissions.
Correspondence and requests for materials should be addressed to R.S.D.
NATURE GEOSCIENCE | VOL 3 | MARCH 2010 | www.nature.com/naturegeoscience 181
© 2010 Macmillan Publishers Limited. All rights reserved.
