Chapter 18
Detecting Intact Forests from Space: Hot Spots
of Loss, Deforestation and the UNFCCC
Fre
´de
´ric Achard, Hugh Eva, Danilo Mollicone, Peter Popatov, Hans-Jurgen
Stibig, Svetlana Turubanova, and Alexey Yaroshenko
18.1 Introduction
Changes in forest cover have become recognised as an important global environ-
mental issue. This chapter aims to synthesise what is known about areas and rates of
forest-cover change in the tropics and boreal Eurasia from the 1990s onwards,
based on data compiled from expert opinion and earth observation technology.
Since the early 1990s, changes in forest area can be measured with confidence from
space from the global to the regional scale (Mollicone et al. 2003).
Forest-cover change (including deforestation) at the regional scale is the process of
land-cover change that is most frequently measured. During the 1990s, rates of forest-
cover change were much higher in the tropics than in other parts of the world. In
particular, the Amazon basin and Southeast Asia contain a concentration of defores-
tation hotspots, and more regional remote sensing studies cover the tropics than boreal
zones. However, forest degradation in Eurasia, related mostly to unsustainable
logging activities or increases in fire frequency, has been growing in recent years.
In addition to reviewing the results from Earth observation studies, this chapter
presents a potential accounting mechanism in the context of the United Nations
Framework Convention on Climate Change (UNFCCC) question of reducing emis-
sions from deforestation in developing countries (UNFCCC 2006), which builds on
recent scientific achievements related to the estimation of tropical deforestation
rates from Earth observation technology.
18.2 Monitoring of Forest Areas from the Global
to the Regional Scale using Satellite Imagery
Combined with ground measurements, remote sensing plays a key role in determin-
ing the loss of forest cover. Technical capabilities have advanced since the early
1990s and operational forest monitoring systems at the national level are now a
C. Wirth et al. (eds.), Old‐Growth Forests, Ecological Studies 207, 411
DOI: 10.1007/978‐3‐540‐92706‐8 18, #Springer‐Verlag Berlin Heidelberg 2009
feasible goal for most developing countries (DeFries et al. 2006). Several appropri-
ate methods are now available to analyse satellite data to measure changes in forest
cover. These methods range from visual photo-interpretation to sophisticated digi-
tal analysis, and from wall-to-wall mapping to hot spot analysis and statistical
sampling. Clearings for large-scale mechanised agriculture are detectable with
medium resolution data (hundreds of metres spatial resolution), whereas small
agricultural or settlement clearings of 0.5 1 ha require higher resolution data
(tens of metres) to be detected accurately.
Analysis of remotely sensed satellite data is the only practical approach to
measure changes in forest area at the regional to global scale. High resolution
data, with almost complete global coverage, are available at low or no cost for the
1990s, early 2000s and around year 2005, in particular Landsat satellite data from
NASA (https://zulu.ssc.nasa.gov/mrsid), the USGS (http://edc.usgs.gov/products/
satellite/landsat ortho.html) or from the University of Maryland’s Global Land
Cover Facility (http://glcfapp.umiacs.umd.edu/). It has been demonstrated that
estimates of deforestation can be provided by using such data at the global or
continental level (Achard et al. 2002; FAO 2001), or at national level for very large
countries such as Brazil or India (INPE 2005; Forest Survey of India 2004).
Deforestation, defined as the conversion of forest land to non-forest land, is most
easily monitored. Estimating forest degradation resulting from practices such as
unsustainable timber production, harvesting of wood for fuel, and fires clearing the
edge of forest fragments is more technically challenging than measuring deforesta-
tion. Quantifying the accuracy of the result and ensuring that consistent methods are
applied at different time intervals is critical. Accuracies of 80 95% are achievable
for monitoring with high resolution imagery to discriminate between forest and
non-forest (DeFries et al. 2006). Accuracies can be assessed through in-situ obser-
vations or analysis of very high resolution aircraft or satellite data.
18.3 Information on Global Forest Extent
and Deforestation Rates
18.3.1 Distribution of Forest Areas at Global Scale
In the late 1990s, data from AVHRR (advanced very high resolution radiometer)
sensors at 1.1 km resolution on board the United States National Oceanic and Atmo-
spheric Administration’s polar orbiting meteorological satellites were used to produce
pan-tropical forest maps at around 1 km resolution (Fig. 18.1) with classification
techniques adapted to the ecological conditions of these areas, e.g. low seasonality
and nearly permanent cloud coverage (Achard et al. 2001). Recently, the
VEGETATION sensoron board SPOT-4 and SPOT-5 satellites, and theMODIS sensor
on board the Terra and Aqua satellites allowed for a spatial and thematic refinement of
the previous global maps. In the framework of the Global Land-Cover 2000 project
412 F. Achard et al.
(GLC-2000), teams of regional experts mapped each continent independently using
VEGETATION data for the year 2000 at 0.01geographic resolution, i.e. at around
1.1 km resolution at the equator (Bartalev et al. 2003; Eva et al. 2004; Latifovic
et al. 2004; Mayaux et al. 2004; Stibig et al. 2003, 2004). To complement mapping
data, a ‘‘vegetation continuous fields’’ algorithm has been developed using MODIS
data to map the global percent tree cover at 500 m resolution (Hansen et al. 2003).
To produce estimates of the global extent of tropical forests, different approaches
have been developed so far, based mainly on: (1) compilation of national inventories
or maps; (2) statistical sampling with high spatial resolution satellite images; or (3)
global coverage of forested areas by remote sensing data at medium to coarse
resolution.
Each method suffers from its own limitations as detailed in Mayaux et al. (2005),
and each assessment uses its own definition of ‘forest’, e.g. based on a different
cover threshold or with some specific land-use characterisation. Therefore, forest
area figures vary considerably among the assessments, as illustrated in Table 18.1.
18.3.2 Distribution of ‘Intact Forests’: from Boreal Eurasia
to the Global Scale
There are many definitions of forest degradation relating to canopy cover, ecologi-
cal function, carbon stocks, and other attributes of forests (Penman et al. 2003).
Degradation defined by changes in canopy cover is most readily observable
with remote sensing. The concept of ‘intact forest landscapes’ was first applied
by the Global Forest Watch network over Russia (Yaroshenko et al. 2001;
Aksenov et al. 2002). It was extrapolated across the world using a consistent
set of criteria and high-resolution satellite imagery from throughout the year
2000 (Greenpeace 2006). This new map of the world’s intact forests depicts the
remaining large forest areas where it can be assumed that human influence is
limited (Fig. 18.2).
Table 18.1 Tropical forest areas derived from the GLC 2000 map and from the FAO FRA
exercise. CS: Country Survey (compilation of national statistics), RSS: Remote Sensing Survey
GLC 2000
a
FAO FRA 2000
b
Humid
tropics
(10
6
ha)
Dry
tropics
(10
6
ha)
Flooded
forests
(10
6
ha)
CS RSS
Closed
forests
(10
6
ha)
Open
forests
(10
6
ha)
Forests (10
6
ha)
South America 630 147 25 858 69 780
Africa 233 415 13 353 289 518
Southeast Asia 231 145 13 416 58 272
Global 1,094 707 52 1,627 416 1,572
a
Global Land Cover 2000 project (Mayaux et al. 2005)
b
FAO Forest Resources Assessment 2000 (FAO 2001)
18 Detecting Intact Forests from Space 413
This forest distinction between ‘intact’ and ‘non-intact’ is based on experience
with satellite-based forest mapping and uses a ‘negative approach’; disturbance
such as the development of roads can be detected easily, whilst the absence of such
visual evidence of disturbance can be taken as evidence that what is left is ‘intact’
(Yaroshenko et al. 2001). Intact forest areas were originally defined for the boreal
ecosystems according to the following six criteria: situated within the forest zone;
larger than 50,000 ha, and with a smallest width of 10 km; containing a contiguous
mosaic of natural ecosystems; not fragmented by infrastructure; without signs of
significant human transformation; and excluding burnt lands and young tree sites
adjacent to infrastructure objects (with 1 km wide buffer zones). This definition has
been applied to all forest ecosystems of the world (Greenpeace 2006) but could be
easily adapted for other purposes (see Sect. 18.4). Disturbance is easier to identify
unequivocally from satellite imagery than the forest ecosystem characteristics that
would need to be determined if we followed the ‘positive approach’ i.e. identifying
intact forest and then determining that the rest is non-intact. Following the negative
approach, forest conversions between intact forests, non-intact forests and other
land uses can be measured easily worldwide through Earth observation satellite
imagery. In contrast, other definitions of forest status (e.g. pristine, virgin, primary/
secondary, etc.) are very difficult to quantify at large scale (Chap. 2 by Wirth et al.,
this volume).
18.3.3 Hot Spots of Forest Loss
For the humid tropics, areas of rapid deforestation were first identified through
expert knowledge (Achard et al. 1998). This information was used to sample areas
to be analysed with high resolution data (Achard et al. 2002). Experts with detailed
knowledge at the country or regional level ensured that areas of major change were
not overlooked. Databases such as transportation networks, population changes and
locations of government resettlement programmes can also be used to help identify
areas where the pressure to deforest is likely to be high and where a more detailed
analysis is required. Globally, the main forest conversion process in the humid
tropics is the transformation of closed, open or fragmented forests to agricultural
land. The major forest changes are largely confined to a number of ‘‘hot spot’’ areas
where forests are increasingly fragmented, heavily logged or burnt, and where rates
of change are alarmingly high. In Latin America, the transformation from forest to
agriculture by clear-cutting predominates. In addition, areas of mosaics or savannah
woodlands have been transformed for agriculture.
A more recent study based on this ‘hot spot’ assessment in the tropics identified
areas of recent and current rapid forest-cover change at a global level from expert
knowledge, and characterised the main drivers of these changes (Lepers et al.
2005). It concluded that, at the end of the 1990s, Asia had the greatest concentration
of areas of rapid land-cover changes, and that the Amazon basin remained a major
414 F. Achard et al.
hotspot of tropical deforestation. These results were supported by a national
Brazilian assessment through the PRODES monitoring system (INPE 2009),
which identifies ‘critical areas’ based on the previous year’s monitoring to prioritise
analyses for the following year.
More recently still, the broad geographic patterns of rapid forest-cover change
have been mapped for boreal Eurasia, with characterisation of their main causes
from expert opinion and remote sensing data (Achard et al. 2006). Around 40 mil-
lion ha of rapid change with clear-cutting activities and 70 million ha with increased
fire frequency were depicted. Rapid land-cover change is not randomly or uniform-
ly distributed but is clustered in some locations, e.g. high intensity logging
takes place mostly in the European part of Russia (e.g. in the Karelian Isthmus)
and along the southern border of the Taiga. Forest degradation in Siberia related
mostly to an increase in fire frequency and development of logging activities is
extending rapidly. Annual rates of forest-cover change in areas identified as ‘rapid
change areas’ may range from 0.26% year
–1
for diffuse logging activities to
around 0.65% year
–1
for areas affected by intense clear-cutting activities, up to
2.3% year
–1
for areas affected by fires or a combination of fire and logging (Achard
et al. 2006). While such an approach does not lead directly to quantitative
estimates of forest-cover changes, it highlights those areas where intensive moni-
toring would be required for an improved estimation of the changes at the
continental scale (Potapov et al. 2008).
18.3.4 Estimates of Forest Conversion Rates in the Tropics
During the 1990s, rates of forest-cover changes were much higher in the tropics
than in other parts of the world. To estimate deforestation over the whole tropical
belt, three main methods have been tested. (1) Gathering information through
reports, national statistics and independent expert opinions (FAO 2001). This
approach is limited by the heterogeneity of the applied methods and forest defini-
tions used. (2) Measuring change using fine resolution satellite imagery on a
sampling basis (FAO 2001; Achard et al. 2002). This approach exploits the fine
spatial resolution of satellite images but requires a well designed sampling strategy.
(3) Measuring change using coarse resolution satellite imagery (DeFries et al. 2002;
Hansen et al. 2005). This approach measures changes in ‘‘percent tree cover’’ but
must be carefully calibrated with local studies.
The TREES (tropical ecosystem environment observations by satellites) project
(Achard et al. 2002) estimated deforestation rates for four regions of the humid tropics:
(1) Pan Amazon and Central America, (2) Brazil Amazonia and Guyana, (3) Africa
and (4) Southeast Asia. The TREES forest definition corresponds closely with the
FAO definition of ‘closed broadleaved forest’ (FAO 2001). The resulting estimates
of global humid tropical forest area change for the period 1990 1997 showed a
marked reduction of closed forest cover: the annual deforested area for the humid
tropics is estimated at 5.8 1.4 million ha with a further 2.3 0.7 million ha of
18 Detecting Intact Forests from Space 415