/ www.sciencexpress.org / 9 August 2007/ Page 1 / 10.1126/science.1146324
Disturbance and deforestation have profound ecological
and socio-economic effects on tropical forests, but their
diffuse patterns are difficult to detect and quantify at
regional scales. We expanded the Carnegie forest damage
detection system to show that, between 1999 and 2005,
disturbance and deforestation rates throughout the
Peruvian Amazon averaged 632 km2 yr-1 and 645 km2 yr-1,
respectively. However, only 1-2% occurred within natural
protected areas, indigenous territories contained only
11% of the forest disturbances and 9% of the
deforestation, and recent forest concessions effectively
protected against clear-cutting. Although the region
shows recent increases in disturbance and deforestation
levels, and leakage into forests surrounding concession
areas, land-use policy and remoteness are serving to
protect the Peruvian Amazon.
Tropical forests play essential roles in ecological, climate and
biogeochemical processes, and in the lives of human
populations (1–4), but anthropogenic disturbances can disrupt
forest structure, function, and composition (5–7). Because of
its large, relatively contiguous area of primary rainforest, the
Peruvian Amazon has a major conservation value, and is
considered a priority in nearly all global biodiversity
inventories (8). Despite the internationally recognized
uniqueness and importance of Peruvian rainforest
ecosystems, the impacts of human activities throughout the
region remain poorly understood.
Increasing rates of large-scale forest damage in the
neighboring Brazilian Amazon have been linked to modern
road building and government policies supporting resource
extraction and settlement (9, 10). Peru’s 661,000 km2 of
Amazon tropical forest are also subject to elevated human
impacts that have not been well documented at the landscape
level: the paving of the Inter-Oceanic Highway and the
spreading road network throughout the Pucallpa region have
brought migrants mostly from the Peruvian Andes, along with
largely undocumented impacts on forest cover and structure.
However, in recent years the Peruvian government has also
established or extended large natural protected areas and
indigenous territories in the Peruvian Amazon, and forest
management legislation has placed 31% of its forests into
permanent resource production status (table S9), 104,970 km2
of which into long-term, timber-producing, commercial
concessions by 2005 (11). Small-scale studies have noted an
increase in forest damage within some protected areas, mostly
as a result of land conversion to agriculture and pasture near
human settlements and river valleys (12) associated with
proximity to roads, rural credit programs, and access to
markets (13), as well as inadequate land-use planning and
governance (14). However, a synoptic assessment of forest
disturbance and deforestation has not been derived for
Large-scale assessments of forest disturbance in the
Peruvian Amazon, typically diffuse and difficult to detect,
require complex detection algorithms for the analysis of high-
resolution satellite imagery (15, 16), but these methods are
just now proving critical for land management, conservation
analysis, and land-use policy assessments in tropical forest
regions (17). Here, we adapted a satellite-based forest
disturbance detection system, originally designed for
industrial-grade timber extraction monitoring in Brazil, to
Peru’s generally smaller-scale forest disturbance regimes. We
present an updated version of the Carnegie Landsat Analysis
System (CLAS©1.1), and apply it to a study area covering
79% of the Peruvian Amazon (18) from 1999 to 2005. The
core technology of the CLAS change detection algorithm (15,
19, 20) was improved with optimized, automated versions of
the atmospheric and haze correction and water/cloud masking
processes of the Monte Carlo Unmixing (AutoMCU)
approach (21). We also added an automated deforestation
Land-Use Allocation Protects the Peruvian Amazon
Paulo J. C. Oliveira,1 Gregory P. Asner,1* David E. Knapp,1 Angélica Almeyda,1,2 Ricardo Galván-Gildemeister,3 Sam Keene,4
Rebecca F. Raybin,1 and Richard C. Smith3
1Department of Global Ecology, Carnegie Institution of Washington, Stanford, CA, 94305 USA.
2Department of Anthropological Sciences, Stanford University, Stanford, CA, 94305 USA.
3Instituto del Bien Común, Av. Petit Thouars 4377, Miraflores, Lima 18 Perú.
4Department of Electrical and Computing Engineering, Boston University, 8 Saint Mary's Street, Boston, MA 02215 USA.
*To whom correspondence should be addressed. E-mail: firstname.lastname@example.org
detection component to provide an integrated analysis of both
diffuse forest disturbance and clear-cutting. We used 101
Landsat 5 TM and Landsat 7 ETM+ satellite images at a
spatial resolution of 30 x 30 m to derive annual incremental
damage maps for most of the human-impacted, timber-
producing regions—up to 24 images per year, with each non-
overlapping footprint covering 26,000 km2. The satellite
detection results were validated via a large field survey in the
Pachitea and Ucayali watershed regions, and regionally
evaluated against available land use, land cover and
We found that 632±230 km2 yr-1 and 645±325 km2 yr-1 of
Peruvian Amazon forests were subjected to new forest
disturbances and deforestation, respectively, between 1999
and 2005 (Table 1). Our forest disturbance values represent
previously unaccounted human impacts throughout the
region. The deforestation portion of our analysis is in
agreement with FAO deforestation estimates (table S1), but
are lower than those reported by the Peruvian government
Between 1999 and 2001, we found that 86% of all forest
damage was concentrated in only two regions. In this period,
the four satellite scenes covering the area around the Ucayali
logging center of Pucallpa, and along the road network that
emanates from it (Fig. 2), had the highest rates of forest
disturbance and deforestation, contributing 64% of the total
Peruvian Amazon damage. This was followed by the four
satellite images covering the corridor centered in the eastern
Madre de Dios capital city of Puerto Maldonado, which
extended along the Inter-Oceanic Highway, showing 23% of
the total damage (Fig. 1). We therefore concentrated on a five
satellite scene subset for more detailed analyses from 1999 to
2005. Within this subset, forest damage rates remained
relatively constant between 1999 and 2003, with average
forest disturbances and deforestation rates of 340±44 km2 yr-1
and 469±35 km2 yr-1, respectively (table S7). Total forest
damage rates then increased substantially between 2003 and
2005, particularly in the last year of our analysis, where
disturbance and deforestation rates of 995 km2 yr-1 and 1140
km2 yr-1 were 2.9 and 2.4 times higher than the average for
the initial four years, respectively (21). In particular, forest
disturbance greatly increased east of Pucallpa in 2004, and
west of the Iberia area of Madre de Dios in 2005, in regions
where forest concessions had recently been granted.
Forest disturbances and deforestation were detected in
other areas to the north near the Loreto capital of Iquitos,
where early indications of small-scale damage were seen in
1999, but these increased in intensity over the years of
analysis, spreading to nearby forest areas on both sides of the
Amazon River (Fig. 1). The northern Loreto forests close to
the Colombian border, which maintained relatively low
damage rates between 1999 and 2002, mostly in and around
/ www.sciencexpress.org / 9 August 2007/ Page 2 / 10.1126/science.1146324
native communities’ lands along rivers, showed only a slight
increase in forest disturbances by 2004-05. The remote Napo
moist forests of western Loreto showed very little damage
between 1999 and 2002, concentrated on river edges (17).
The concentration of forest damage along the Iquitos-to-
Nauta road is a clear indication that road access could be the
most important control over forest disturbance and
deforestation rates in the remote Peruvian Amazon, where
sheer distance and the intricate hydrologic network of the
Amazon and Marañón rivers likely prevent high damage
intensities, and where timber extraction may be limited by
current road access to markets (9).
Overall, only 2% of the forest disturbances and 1% the
deforestation detected in the entire study area occurred within
the boundaries of natural protected areas. Furthermore,
territories occupied by indigenous communities contained
11% and 9% of the total forest disturbance and deforestation,
respectively (Table 2). These results clearly show that these
two forms of land-use allocation can provide effective
protection against forest damage. However, a few exceptions
occurring on indigenous community lands in the Oxapampa
and Puerto Inca provinces, and, to a lesser extent, in the El
Sira natural protected area, appear to be related to proximity
to roads, indicating that the protection afforded by their legal
status may not be sufficient when the land is highly accessible
to markets (6). In fact, an estimated 75% of the total Peruvian
Amazon forest damage, including 66% of disturbances and
83% of deforestation, was detected within a 20 km distance
from the nearest roads (Figure 1). However, even within that
20-km buffer, forests within conservation units were more
than four times better protected against deforestation than
unprotected forests (21). Even after compensating for
differences in the geographic extent of each land-use type,
forest damage was about 18 and 10 times more likely in
undesignated and indigenous territories, respectively, than in
natural protected areas (21).
We also evaluated the impacts of recent timber harvest
legislation on rates of forest disturbance and deforestation,
before and after their enactment (11). Within all permanent
production forests allocated to long-term concessions
between 2002 and 2004, deforestation rates were up to two
orders of magnitude smaller than forest disturbance resulting
from the logging operations (table S5). However, outside the
concession areas granted in 2004 in the remote northern
Iquitos region, disturbance and deforestation rates increased
by 468% and 304%, respectively. This leakage effect was
also prevalent in the central Pucallpa logging region, where
deforestation and forest disturbances outside concessions rose
almost 400%, to a combined rate of 1086 km2 in 2005.
Furthermore, the Madre de Dios logging region observed an
increase within and outside concessions, but still at relatively
low rates. These results suggest that sanctioned forest
extraction activities may be an effective deterrent against
forest clear-cutting, but closer monitoring of neighboring
non-concession lands is critical to prevent leakage around
concession forests. A time-series analysis of our data shows
that the rate of clear-cutting previously disturbed forest was
1.8%, 7.2%, and 13.8% at one, three, and five years,
respectively, after the initial disturbance (table S10). These
relatively low values suggest that forest disturbances in the
Peruvian Amazon are not simply a precursor to deforestation.
Our field validation studies showed that the CLAS
methodology is precise and accurate in detecting forest
disturbance and deforestation in the Peruvian Amazon. Our
uncertainty was 10.5% for forest disturbances and 0.5% for
deforestation (table S6). Atmospheric correction, cloud cover,
and annualization errors in the satellite analyses were found
to be very low and had been proven nearly negligible
compared to manual audit uncertainty (15, 21).
The establishment of protected natural areas, the titling of
native territories, and the sanctioning of selective logging
activities, have combined with the Peruvian Amazon’s
traditional conservation allies—its remoteness and a complex
hydrological network—to ensure a moderate level of success
in the conservation of its forest ecosystems. Economic
development of the forest sector, which employed 279,000
people nationally in 2001 (24), is essential for the well-being
of human populations, but poorly monitored logging
concessions, along with the challenges of uncontrolled road
access, may hinder efforts to maintain ecological function and
diversity in Peruvian rainforests in the future. Deforestation
pressures, along with rising rates of forest disturbance, in
many tropical countries are often at odds with increasing
conservation efforts (25, 26). A balanced portfolio of forest
use and protection, along with substantive law enforcement,
could be used to sustain the services provided by tropical
forests to society while also protecting those forests.
Increased satellite monitoring of logging and other forest
disturbances will thus be essential to conservation,
management and resource policy development efforts in Peru
and other rainforest nations.
/ www.sciencexpress.org / 9 August 2007/ Page 3 / 10.1126/science.1146324
References and Notes
1. A. Gentry, Oikos 63, 19 (1992).
2. J. Shukla, C. Nobre, P. Sellers, Science 247, 1322 (1990).
3. J. Grace et al., Science 270, 778 (1995).
4. J. A. Foley, G. P. Asner, M. H. Costa, M. T. Coe, R.
DeFries, H. K. Gibbs, E. A. Howard, S. Olson, J. Patz, N.
Ramankutty, P. Snyder, Front. Ecol. Environ. 5, 25
5. A. R. Holdsworth, C. Uhl, Ecological Applications 7, 713
6. B. S. Soares-Filho, D. C. Nepstad, L. M. Curran, G. C.
Cerqueira, R. A. Garcia, C. A. Ramos, E. Voll, A.
McDonald, P. Lefebvre, P. Schlesinger, Nature (UK) 440,
7. D. Werth, R. Avissar, Journal of geophysical research 107,
8. T. M. Brooks et al., Science 313, 58 (2006).
9. J. Imbernon, Ambio 28, 509 (1999).
10. D. Nepstad et al., Science 295, 629 (2002).
11. PERÚ, Reglamento de la Ley Forestal y de Fauna
Silvestre, DS Nº 014.2001-AG, del 9 de abril, 2001.
12. INRENA/SZF/CDC-UNALM, Hacia un Sistema de
Monitoreo Ambiental Remoto Estandarizado para el
SINANPE. Piloto V: Parque Nacional Manu, Parque
Nacional Alto Purús, Reserva Comunal Purús y Santuario
Nacional Megantoni (2000-2005) (INRENA, Lima, Perú,
13. N. L. Alvarez, L. Naughton-Treves, Ambio 32, 269
14. S. Mäki, Environ. Conserv. 28, 199 (2001).
15. G. P. Asner et al., Science 310, 480 (2005).
16. G. P. Asner et al., PNAS 103, 12947 (2006).
17. L. M. Curran, S. N. Trigg, PNAS 103, 12663 (2006).
18. E. Dinerstein et al., A Conservation Assessment of the
Terrestrial Ecoregions of Latin America and the
Caribbean (The World Bank, Washington, D. C., 1995).
19. G. P. Asner, M. Keller, J. N. M. Silva, Glob. Change Biol.
10, 765 (2004).
20. G. P. Asner, D. E. Knapp, A. N. Cooper, M. M. C.
Bustamante, L. P. Olander, Earth Interactions 9, 1 (2005).
21. Materials and methods are available as supporting
material on Science online.
22. INRENA, Perú Forestal en Números 2002 (INRENA,
Lima, Perú, 2003).
23. CONAM/INRENA, National Environmental Council and
Program on National Capacity Strengthening for
Managing the Impact of Climate Change and Air
Contamination, National Institute of Natural Resources,
Office for Transectoral Environmental Management and
Natural Resource Evaluation and Information
("PROCLIM"), Mapa de Deforestación de la Amazonía
Peruana – 2000, Memoria Descriptiva, IM-03-02,
Volumen I, Texto (CONAM and INRENA, Lima, Peru,
24. FAO, Evaluación de los Recursos Forestales Mundiales
2005, Perú Informe Nacional 201 (FAO, Rome, 2006).
25. FAO, Global Forest Resources Assessment 2005: Main
Report, FAO Forestry Paper 147 (FAO, Rome, 2006).
26. D. Nepstad et al., Conservation Biology 20, 65 (2006).
27. We thank V. Barrena, E. Broadbent, C. Carlson, C. de la
Rosa, B. Haxo, N. Kroll and P. Summers for assistance
with various portions of this project. CLAS©1.1 was
developed by the Carnegie Institution of Washington.
Application of CLAS to the Peruvian Amazon was funded
/ www.sciencexpress.org / 9 August 2007/ Page 4 / 10.1126/science.1146324
by the John D. and Catherine T. MacArthur Foundation
and by the Gordon and Betty Moore Foundation.
Supporting Online Material
Materials and Methods
Tables S1 to S10
11 June 2007; accepted 20 July 2007
Published online 9 August 2007; 10.1126/science.1146324
Include this information when citing this paper.
Fig. 1. Cumulative spatial distribution of forest disturbance
(blue) and deforestation (red) in the Peruvian Amazon
between 1999 and 2005. Left: light gray areas show the
extent of native territories, and dark gray areas show natural
protected areas. Right: orange lines show road distribution,
magenta lines shows 20 km road buffers, and green areas
show the extent of forest concessions allocated by 2005;
Letters A and B denote the Pucallpa and Inter-Oceanic
Highway regions, respectively.
Fig. 2. Two high-resolution examples of forest disturbance
and deforestation detection results from CLAS1.1 overlaid on
satellite imagery, showing impacted forest: A. near Pucallpa
(left), where damage is more extensive in non-protected areas
accessible from roads or rivers; and B. near the remote area of
Iquitos (right) with small damage (see Figure S1 for location).
/ www.sciencexpress.org / 9 August 2007/ Page 5 / 10.1126/science.1146324
† Based on an assessment of the spatial distribution of forest damage in the first three
years, a subset of satellite path/rows was selected for analyses in the following three years,
as available: 5 path/rows in 2002-03, 3 in 2003-04, and 5 in 2004-05.
*The number of satellite path/rows analyzed in each year varied according to image
availability: 23 path/rows in 1999-2000, 23 in 2000-01, and 17 in 2001-02.
Table 1. Forest disturbance and deforestation area estimates for Peruvian
Amazon tropical forest based on CLAS methodology.
Damage rates (km2 yr-1)
* GIS spatial layers obtained from Peru 2000 Forest Map, INRENA.
† Spatial layers of Titled Indigenous Territories based on unpublished data collected and
prepared by the Instituto del Bien Común for an ongoing study, when territories of 80% of titled
indigenous groups had been mapped. Analysis also included Madre de Dios State Reserve
(Indigenous Peoples in Voluntary Isolation) spatial layer from CIF-INRENA, 2005.
Table 2. Percentage of detected forest disturbances and deforestation that
falls within the boundaries of Natural Protected Areas* and Indigenous
Territories† of the Peruvian Amazon.
Damage within Natural
Protected Areas (%)
Indigenous Territories (%)