Emma.

'55:};

 

fur;

 

 

 

vr: - ~ .
.46813

 

 

2563 LIBRARY
54025 7 577 Michigar State
University

 

 

 

This is to certify that the
thesis entitled

MULTI-TEMPORAL ASSESSMENT OF SELECTIVE
LOGGING USING REMOTELY SENSED DATA IN THE
BRAZILIAN AMAZON

presented by

Eraldo A T Matricardi

has been accepted towards fulfillment
of the requirements for the

Master of Arts degree in Geography

 

ISanéL

Major Professor’s Signature

U
Date

 

MSU is an Afirmativa Action/Equd Opportunity Institution

.ll.efi-o-O-OE_3-o-o-n-.-0—.—._.-.-c-

4 -n-m.-——-—-

PLACE IN RETURN Box to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

7110 3 1'4

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c#CIRCIDateDue.p65-p.15

 

 

MULTI-TEMPORAL ASSESSMENT OF SELECTIVE
LOGGING USING REMOTELY SENSED DATA
IN THE BRAZILIAN AMAZON

By

Eraldo A T Matricardi

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF ARTS
Department of Geography

2003

ABSTRACT
MULTI-TEMPORAL ASSESSMENT OF SELECTIVE LOGGING USING
REMOTELY SENSED DATA IN THE BRAZILIAN AMAZON
By
Eraldo A T Matricardi
Deforestation in tropical forests and its consequences on global climate change
have been monitored using remotely sensed data. Most of the currently available land
cover maps show only those sites where trees have been completely removed. Just a few
attempts to detect forest areas where trees were partially removed, as in the case of
selective logging, have been done. This study assesses forest impacts by selective logging
and extend current multi-annual Amazon-wide land use classification to include
selectively logged forests in addition to the more fiequently used thematic classes (forest,
deforestation, regrowth, cerrado, and water body). Remotely sensed data combined with
field studies, Geographic Information Systems (GIS), and related techniques were applied
in this research. This is the first multi-temporal assessment of selective logging in the
Brazilian Amazon in which 1 estimated that at least 5980, 10064, and 26085 10112 of
forest had been logged by 1992, 1996, and 1999, respectively. I also estimated that at
least 3689, 5107, and 11638 Km2 had been actively logged in 1992, 1996, and 1999,
respectively. Finally, I found that at around of 10% of former logging areas detected in
between 1992 and 1996 was revisited or logged again in 1999. I also observed that 13%
of logged forests in 1992 were deforested by 1996, an additional 15% by 1999, and 11%

of forests logged in 1996 were deforested by 1999.

DEDICATION

This thesis is dedicated to my wife (Cleusa), my daughters (Camila and Marcela),
and my parents (Mauro and Adelina) for their infinite love, and for their unconditional

and continual encouragement in completing my Masters degree.

Dedico esta tese a minha esposa (Cleusa), filhas (Camila e Marcela) e aos meus

pais (Mauro e Adelina) pelo infinito arnor e apoio irrestrito na realizacao deste curso dc

Mestrado.

iii

ACKNOWLEDMENTS

Firstly, I would like to thank my Committee Members, Dr. David Skole, Dr.
Jiaguo Qi, Dr. Mark Cochrane, and Dr. Marcos Pedlowski, for the support and guidance
during this manuscript preparation. I thank very much Dr. David Skole, my faculty
advisor, for this opportunity to be here being part of his working team, for his patience,
and direct support. I thank Dr. Marcos Pedlowski, for thrusting and supporting me since
the begging of this journey. I thank Dr. Mark C ochrane for his unconditional support,
critical commentary, and encouragement in accomplishing this research. I thank too Dr.
Jiaguo Qi for his support in this research, especially for teaching me digital image
processing and analysis, without it should not be possible. Special thanks are given to
Walter Chomentowski for his friendship and support during data processing and
fieldwork. Many thanks to Narumon Wiangwang (Nok) and Stephen Cameron for their
friendship and help during my master program. I would also recognize the help provided
by my brothers, Wagner Antonio Trondoli Matricardi and Mauro Lucio Trondoli
Matricardi, and friends Ana Lucia Sousa Lima, Wilson Scares Abdala, Vilmar Ferreira,
Antonio Carlos Nascimento, Herminio Fernandes da Silva Neto, Helio Gomcs de
Oliveira, Nicolau Priante Filho, Fernando Raiter, Jansen Luiz Trienweiler, and Daniel
Gomes de Oliveira, during the fieldwork in the State of Rondonia and State of Mato
Grosso. I thank Dr. Carlos Castro, Luiz Claudio F emandes, and Israel Xavier Batista for
providing data and comments on my research. Special thanks to my friend Flora
Cerqueira for her encouragement. I thank very much Deana Hanner and Diane Cox for

their friendship, patience, and support at Center for Global Change and Earth

iv

Observations. I would also like to recognize the help and support in my academic
pursuits by the Department of Geography, especially Dr. Randall Schaetzl, Dr. Richard
Groop, and Dr. Robert Walker. I am especially thankful for the expertise and contribution
from Dr. Julie A. Winkler, Dr. Antoinette Winklerprins, Dr. Catherine Yansa, and Dr.
Jiaguo Qi, during the coursework. Many thanks are given to Sharon Ruggles, Ellen
Schueller, Judy Hibbler, and Marilyn Bria for their academic and administrative support
at Department of Geography. Last but not least, many thanks to my colleagues, friends,
and classmates from the Center for Global Change and Earth Observations and
Department of Geography for helping me and sharing good moments during this time in

the USA.

TABLE OF CONTENTS

LIST OF FIGURES ........................................................................................................... ix
LIST OF TABLES .............................................................................................................. x
LIST OF APPENDICES .................................................................................................... xi
CHAPTER
1. Problem Statement .......................................................................................................... 1
1.1. Statement of Purpose ............................................................................................... 1
1.2. Background .............................................................................................................. 2
1.2.1. Selective Logging Concepts ............................................................................. 3
1.2.2. Impacts of Selective Logging ........................................................................... 5
1.2.2.1. Impacts on Carbon Cycle ........................................................................... 5
1.2.2.2. Forest Fire Susceptibility ........................................................................... 8
1.2.2.3. Biodiversity .............................................................................................. 10
1.2.2.4. Socioeconomic factors ............................................................................. 12
1.2.3. Synergism between selective logging and deforestation ................................ 14
1.2.4. Previous attempts to map selective logging .................................................... 16
1.3. Bibliography .......................................................................................................... 19
CHAPTER
2. Method for Detection and Validation ........................................................................... 24
2.1. Abstract .................................................................................................................. 24
2.2. Introduction ............................................................................................................ 25
2.3. Methodology .......................................................................................................... 27
2.3.1. Site description ................................................................................................ 27
2.3.2. Data Set ........................................................................................................... 27
2.3.2.1. Calibration and Geometric Rectification ................................................ 28
2.3.2.2. Pan-Sharpening Ikonos image ................................................................. 28
2.3.3. Detection of areas of selective logging ........................................................... 29
2.3.3.1. Digitization of areas of selective logging ................................................ 29
2.3.3.2. Automatic detection of selective logging ................................................ 32
2.3.4— Logging Area Estimates on Landsat imagery ................................................. 37
2.4. Field studies ........................................................................................................... 37
2.5. Results .................................................................................................................... 39
2.6. Analysis of accuracy assessment of the remote sensing techniques ...................... 40
2.7. Conclusion ............................................................................................................. 44
2.8. References .............................................................................................................. 46
CHAPTER
3. Assessment of selective logging detection for the Amazon State of Rondénia ........... 47
3.1. Abstract .................................................................................................................. 47
3.2. Introduction ............................................................................................................ 48
3.3. Methodology .......................................................................................................... 50

vi

3.3.1. Study site and data set ..................................................................................... 50
3.3.2. Field studies .................................................................................................... 50
3.3.3. Detection of selectively logged forest ............................................................. 52
3.4. Buffer radius assessment ........................................................................................ 53
3.5. Discussion and conclusion ..................................................................................... 57
3.6. References .............................................................................................................. 59
CHAPTER
4. Inter and Intra-annual Analysis ..................................................................................... 61
4.1 . Abstract .................................................................................................................. 61
4.2. Introduction ............................................................................................................ 62
4.3. Regional settings: Southern and Southwestern Brazilian Amazonia ..................... 63
4.3.1. Description of the study area in the State of Rondénia .................................. 63
4.3.2. Description of the study site in the State of Mato Grosso .............................. 64
4.4. Methods .................................................................................................................. 65
4.4.1. Satellite imagery and geometric rectification ................................................. 65
4.4.2. Detection of selectively logged forest ............................................................. 65
4.5. Field studies ........................................................................................................... 66
4.5.1. State of Mato Grosso ...................................................................................... 66
4.5.2. State of Rondénia ............................................................................................ 68
4.6. Results .................................................................................................................... 70
4.6.1. Case study: the State of Mato Grosso ............................................................. 70
4.6.1.1. Intra—annual analysis ................................................................................ 75
4.6.2. Case study in the State of Ronddnia ............................................................... 76
4.7. Discussion and conclusion ..................................................................................... 77
4.8. References .............................................................................................................. 79
CHAPTER
5. Amazon-Wide Selective Logging Detection and Measurement ................................... 82
5.1. Abstract .................................................................................................................. 82
5.2. Introduction ............................................................................................................ 83
5.3. Methodology .......................................................................................................... 85
5.3.1. Study Area and Data Set ................................................................................. 85
5.3.2- Calibration and Geometric Rectification ........................................................ 87
5.3.3. Logging Detection .......................................................................................... 87
5.3.3.1. Automatic Analysis .................................................................................. 88
5.3.3.2. Visual Interpretation ................................................................................ 90
5.4. Results .................................................................................................................... 92
5.5. Discussion and Conclusions .................................................................................. 95
5.5.1. Accuracy assessments ..................................................................................... 95
5.5.2. Multi-temporal assessment of selective logging Amazon-wide ..................... 98
5.5.3. Analysis and evaluation of results and Assumptions ...................................... 99
5.5.4. Final considerations ..................................................................................... 104
5.6. References ............................................................................................................ 106

vii

CHAPTER

6. Concluding Remarks ................................................................................................... 108
6.1.Seating the current study in the context of global change research ...................... 108
6.2. Research questions revisited ................................................................................ 110
6.3. Significance and implications of this study ......................................................... 112
6.4. Opportunities for further studies .......................................................................... 112
6.5. References ............................................................................................................ 1 13

viii

LIST OF FIGURES

Figure 2.1. Visible logged forest areas on Landsat ETM+ image, RGB 5/3/2, June 18,

2000 ................................................................................................................. 30
Figure 2.2. Visible logging on Ikonos image RGB 3/4/2, 2000. ...................................... 31
Figure 2.3. Methods used in the logging automatic detection ......................................... 33
Figure 2.4. Spatial profile of a textural image (Landsat ETM+ , band 5). ....................... 35
Figure 2.5. Textural image (band 5) and buffer radius of 180 meters .............................. 36
Figure 2.6. Selective logging detection using visual interpretation and automatic analysis
on Landsat ETM+ , testing different buffer radiuses. ..................................... 41
Figure 2.7. Selective logging detected on Ikonos and Landsat ETM+ imagery displayed
on Ikonos image RGB 3/4/2, April 30, 2000. ................................................. 43
Figure 2.8. Visual interpretation only, automatic analysis only (buffer radius of 180 m),
and overlap between logging areas detected by both techniques. .................. 44
Figure 3.1. Buffer radius around patios intersecting selectively logged forest ................ 54
Figure 3.2. Different buffer radiuses around storage patios (from the field studies) ........ 56
Figure 3.3. Different buffer radiuses around log landings detected using textural
algorithm, Landsat ETM+, band 5. ................................................................. 57
Figure 4.1. Land use and land cover change in the study site (path/ row 226/068) ......... 71
Figure 5.1. Images used in the logging Analysis .............................................................. 86
Figure 5.2. Log landings detected using textural algorithm, band 5, 226/068, 2002 ....... 88
Figure 5.3. Buffer radius of 180 meters around of log landings detected ........................ 89
Figure 5.4. Obvious logging on Landsat image, 226/068, RGB 5/3/2, 2001. .................. 91
Figure 5.5 - Subtle logging on Landsat image, 226/068, RGB 5/3/2, 2001. ................... 91

"Images in this Thesis are presented in color"

ix

LIST OF TABLES

Table 2.1. Land use in the study area by 2000 .................................................................. 39
Table 2.2. Logged forest detected using visual interpretation and buffer radius of 180
meters on Landsat ETM+ image .................................................................... 42
Table 4.1. General results for the path/row 226/068 ........................................................ 72
Table 4.2. Selective logging and deforestation increment for the path/row 226/068 ....... 73
Table 4.3. Deforestation and selective logging for the path/row 226/068 ........................ 74
Table 4.4. Selectively logged forest revisited (path/row 226/068) ................................... 75
Table 4.5. Intra-annual increment of selective logging (path/row 226/068) ................... 76
Table 4.6. Selectively logged for the path/row 232/067 ................................................... 77
Table 5.3. Selective logging by visual interpretation and automatic analysis .................. 92
Table 5.4. Logging area by visual interpretation and automatic analysis ......................... 93
Table 5.5. Selective-logging areas detected by State in the Brazilian Amazon ............... 94
Table 5.6. Basin-wide selective logging total detected and annual rates .................. 103

LIST OF APPENDICES

Appendix A.1. Timber production in Brazil’s Amazonian states - 1990 to 2001 .......... 116
Appendix 8.]. Study site location at Sinop municipality, State of Mato Grosso, Brazil.
................................................................................................................. 1 1 8
Appendix B.2. Satellite images used in the Mato Grosso case study. ............................ 120
Appendix 83. Logged forests detected using Landsat and Ikonos images and different
remote sensing techniques (automated and visual interpretation). ......... 122
Appendix C]. Study site location at Cujubim municipality in the State of Rondfinia,
Brazil. ...................................................................................................... 124
Appendix C.3. Textural algorithm image and buffer radius of 450 meters around log
landings captured using automated method. Path/row 232/066 ............. 128

Appendix C.4. Selective logging areas estimated using fieldwork measurements in the
study site at Manoa ranch, Cujubim municipality, State of Rondénia,
Brazil. ...................................................................................................... 130
Appendix C.5. Selective logging areas estimated using Textural algorithm and Landsat
ETM+, band 5. Study case in the State of Rondénia, Brazil. Path/row
232/066 acquired in 08/05/2002. ............................................................ 132
Appendix D.1. Landsat scenes used in this analysis (Path/row 232/062 and 226/068).. 134
Appendix D.2. Landsat (TM and ETM+) images (path/row 232/067) used for the inter-

annual analysis in the State of Rondc‘mia, Brazil .................................... 136
Appendix D.3. Landsat (TM and ETM+) images used for inter and intra-annual analysis

in the State of Mato Grosso, Brazil. ........................................................ 138
Appendix D.4. Study case results of deforestation and selective logging for 2001 in the

State of Mato Grosso. ............................................................................. 140
Appendix E. l. Landsat imagery used for the basin-wide logging analysis .................... 142
Appendix E.2. Basin-wide results of selective logging detection for 1992, 1996, and

1999 ......................................................................................................... 144

Appendix E.3. Basin-wide spatial distribution of selective logging detected in 1992. .. 146
Appendix E.4. Basin-wide spatial distribution of selective logging detected in 1996. .. 148
Appendix E.5. Basin-wide spatial distribution of selective logging detected in 1999. .. 150

xi

CHAPTERI

Problem Statement

In the past four decades tropical forests in the Brazilian Amazon have been
intensively deforested and selectively logged. Remote sensing techniques have been
successfully applied to monitor areas where forests were converted to agriculture,
pasture, or urbanization. Selectively logged forest areas have not been included in current
deforestation estimates in the Brazilian Amazon because they are not easily detectable
using unsupervised classification. Nevertheless, some attempts to quantify the area of
selective logging have been conducted in the Brazilian Amazon, resulting in indirect
estimation of areas where forests have been partially damaged by selective logging and
not completely removed as in the case of deforestation. Those estimates of selectively
logged forest in the Brazilian Amazon have raised questions: Is selective logging a
significant disturbance in terms of area increment of total forest cleared or damaged? Can
we efficiently measure it basin-wide as a form of degradation in addition to outright
deforestation? Is logging causing more deforestation than would otherwise occur? These

questions are addressed in this research.

1.1. Statement of Purpose

The Brazilian Amazon has been increasingly exposed to deforestation (Skole and
Tucker 1993, Houghton et a1. 2000, Achard et a1 2003, INPE 2003), forest fire
(Holdsworth and Uhl 1997, Uhl 1998, Cochrane 2003, Cochrane 2001, Cochrane 2002),
and to selective logging (Stone and Lefebvre 1998, Nepstad et a1. 1999, Souza and

Barreto 2000). Recent estimate is that up to 11.7% of the tropical forest of the Brazilian

Amazon have been deforested by 2001 (INPE, 2003). Forest fires annually burn
thousands of square kilometers of standing tropical moist forests in Amazonia, which
become more susceptible to recurrent and more severe fire (Cochrane and Schulze,
1998). Up to 413 million cubic meters of round wood were logged from 1990 to 2002 in
the Brazilian Amazon (IBGE, 2003).

The purpose of this study is to assess the extent of forest affected by selective
logging in the Brazilian Amazon through a multi-temporal analysis using remotely
sensed data, geographic information systems, remote sensing techniques, fieldwork
observations and measurements.

Forest areas impacted by selective logging are likely to be rapidly increasing,
making it a land use of growing concern in the Brazilian Amazon (Stone and Lefebvre
1998, Nepstad et a1. 1999, Alvarado and Sandberg 2001). In this study, 1 exam how
effectively selective logging can be detected and quantified both locally and basin-wide

using remote sensing techniques.

1.2. Background

The Brazilian Amazon region contains 40% of the world’s remaining tropical
rainforests and it assumes an important role in maintaining biodiversity, regional
hydrology and climate (F earnside 1999), and terrestrial carbon storage (F earnside 1999,
Houghton et a1. 2000).

Satellite-based remote sensing has been used for many years to monitor and
evaluate land cover change in the Amazon, concentrating on forest and deforested land

evaluation (Fearnside et a1. 1990, Skole and Tucker 1993). Data from these studies, such

as the rate of deforestation in Brazil’s Legal Amazon, are often used to estimate human
effects on the global carbon cycle (F earnside 1997, Houghton 1997). However, the
deforestation estimates for Brazilian Amazon have been detecting less than half of the
total area of damaged or impoverished forest each year. Most of those non-detected areas
are related to forest degradation by selective logging and fire (Nepstad et al. 1999).

The impacts caused by logging in tropical forests are considered significant in
terms of forest degradation and fire susceptibility (Stone and Lefebvre 1998, Nepstad et
al. 1999, Souza and Barreto 2000). Feamside (1997) estimated that in the Brazilian
Amazon impacts by selective logging in undisturbed forest create a net carbon flux equal
to approximately 4-7% of the annual carbon released from deforestation. Therefore, the
carbon flux estimates for the Amazon region may be low because of the increasing
logging areas that have not been accounted for (Nepstad et a1. 1999).

Selective logging concepts, its socio-economical implications, and consequences
on the carbon cycle, forest fragmentation, and biodiversity as well as previous work to

detect selective logging using remotely sensed data, are presented as following.

1.2.1. Selective Logging Concepts

The term selective logging is used to define timber extraction of a select group of
tree species, usually the more economically valuable trees (Verissimo et a1. 1995, Uhl et
a1. 1997). Unlikely deforestation, selective logging activities do not clear-cut and burn
forests, but harvests a portion of trees. Though less significant than clear-cutting,
selective logging can damage a large portion of forest and many others trees during
logging operations (Uhl et al. 1991, Verissimo et a1. 1992, Pinard and Putz 1996, Uhl et

a1. 1997, Huth and Ditzer 2001). Using traditional techniques, the harvesting of a single

tree can directly result in the death of thirteen other trees as result of the tree felling
direction and logging operations such as opening up of trails, roads, and log storage
patios (Verissimo et a1. 1992).

Recent reports suggest that selective logging activities have been occurring in
Brazil’s tropical forests for several years (Stone and Lefebvre 1998, Nepstad et a1 1999,
Alvarado and Sandberg 2001). Traditionally, these activities have been restricted to flood
plains (va’rzeas) due to the easy access through the rivers (fluvial transportation) in areas
annually inundated. However, the road network expansion during 19605 and 19708 in the
Brazilian Amazon may have permitted the expansion of selective logging to the
interfluvial (terrafirme) forest (Uhl and Vieira 1989, Uhl et al. 1997).

Based on these observations, selective logging can be characterized in terms of its
spatial distribution and intensity. Spatially, selective logging occurs in ‘varzea’ and ‘terra
firme’ forest in the Brazilian Amazon. Both ‘varzea’ and ‘terra firme’ selective logging
vary in terms of its intensity (low, moderate, and high), depending on the volume and
number of harvested trees (Uhl et a1. 1997). Low impact logging (highly selective
logging) removes as few as one or two high value tree species per hectare, as is the case
when harvesting for Virola (Virola surinamensis (Rol.) Warb ) and Mahogany
(Swietenia macrophylla, King) (Verissimo et a1. 1995, Uhl et a1. 1997). High impact
logging removes 100 or more species per hectare, which occurs highly managed logging
regions of the Amazon basin (Uhl et a1. 1997).

Uhl et a1. (1997) have also described a form of logging referred to as forest

‘mining’, a practice which impoverishes an intact forest over 30 years through a sequence

of progressively more intensive logging, leading to forest degradation or, in some cases,

outright deforestation.

1.2.2. Impacts of Selective Logging

Selective logging leaves behind a mixed landscape of intact forest, tree fall gaps,
road openings, log-loading patios, and damaged forest. It also results in large amount of
dead slash or dried biomass which enhances fuel load and forest fire susceptibility (Uhl
and Buschbacher 1985, Stone and Lefebvre 1998, Nepstad et al. 1999, Souza and Barreto
2000). Moreover, as many as 40% of the remaining standing trees are killed or severely
damaged during logging operations (Uhl et a1. 1991). These impacts can be devastating

even when a small volume of timber is harvested (Frumhoff, 1995).

1.2.2.1. Impacts on Carbon Cycle

The balance of net carbon flux is the balance between areas undergoing declining
C-stock and areas of increasing C-stock during vegetation growth (Palm et al. 1986,
Houghton 1996, Houghton 1998, Houghton et a1 2000). Rates of land use are used to
estimate the annual flux of carbon between terrestrial ecosystems and atmosphere.
Undisturbed areas are considered to have no net carbon flux (Palm et a1. 1986, Houghton
1998). Therefore, tropical deforestation as result from suite of social and economic
pressure on natural resources (Skole et a1. 1994), is a major component of the carbon
cycle (Palm et a1. 1986, Houghton 1996, Houghton 1998, Houghton eta12000).

Carbon is released to the atmosphere following deforestation and logging through

immediate oxidation of organic matter by burning and long-term oxidation through

decomposition (Houghton 1998, Houghton et a1. 2000). The highest carbon losses occur
when high biomass forest areas are converted to low biomass systems such as pasture or
agriculture (Palm et a1. 1986, Houghton 1996, Houghton 1998). If tree planting or
regeneration occurs in a cleared area, there is an initial loss of carbon followed by a
gradual accumulation of carbon (Palm et al. 2000).

However, uncertainties in estimates of forest biomass (Houghton et al., 2001) and
land use change (Achard et al. 2002) are responsible for current uncertainty in estimates
of carbon flux in Amazénia, the most important region of carbon emissions. Indeed,
Skole et a1. (1994) affirmed that estimates of deforestation in tropical forests is a research
challenge because of the geographic extent, spatial pattern, lack of accurate
measurements, and its causes. Kauffinan and Cummings (1995) also stated the lack of
information from quantitative sources for estimating biomass loss by combustion
processes.

Using rates of land use change for the period 1850 to 1990, Houghton (1998)
estimated that 108 x 1015 gC were released from forests to the atmosphere of which 67%
was from tropical forest clearing and 33% from Boreal and temperate forest clearing. The
estimate of annual net carbon flux in the 19808 was an average of approximately 2.0 x
1015 gC year", most of it from changes in land use in tropical regions. Recently, Achard et
a1. (2002) based on updated and more accurate remotely sensed information, presented a
new estimates for the total global carbon emission from land use changes for the 19908 of
about 0.8 to 2.4 x 10‘5gc yr".

DeFries et a1. (2002) based on remote-sensing data and a terrestrial carbon model,

estimated that net mean carbon fluxes for the 19808 and 19908 from tropical deforestation

and regrowth is 0.6 and 0.9 x 10'5gC yr", respectively. More specifically, Houghton et al
(2000) based on the annual rates of deforestation and spatial distribution of biomass,
deforestation and land cover changes, estimated that the Brazilian Amazon alone was a
source for about 0.2 x 10'5gC yr'I over the period 1989-1998.

In addition to outright deforestation, Amazonian forests are increasingly being
exposed to logging activities (Uhl et a1. 1997, Stone and Lefebvre 1998, Nepstad ct a1.
1999, Souza and Barreto 2000). These activities modify, at least temporally, the carbon
storage in the forest (Houghton, 1998) and its impacts, yet not included in the land use
classification, may add 4-7% to this carbon flux estimate. Forest fires may double the
carbon flux amounts in years following drought (Nepstad et a1. 1999).

The impacts of selective logging vary considerably with extraction intensity, but
in worst cases the impacts can be significant in area and amount of biomass removed.
Selectively logged forests accumulate carbon over time and recover to pre-harvest levels
of biomass if left undisturbed. However, many forests are revisited several times when
loggers return to harvest additional tree species as regional timber markets develop (Uhl
et a1. 1997, Verissimo et a1 1995). These forests become highly degraded and may have
40 — 50% of the canopy cover destroyed during these repeated logging operations (Uhl
and Vieira 1989, Verissimo et a1. 1992).

Forests heavily impacted by repeated selective logging become more susceptible
to fire (Holdsworth and Uhl 1997, Cochrane 1999) and, if they burn, will contribute more
emissions of carbon to the atmosphere than areas which area logged only once
(Houghton, 1997). This emission of carbon can vary fiom 7.5 to 7 x 10 MgC ha"l

depending on what previously happened in terms of land use and fire (Cochrane et al.,

1999). Combined, selective logging and fire degrade forest structure, creating different
land cover types characterized between natural forest and deforested areas (Gerwing,

2000)

1.2.2.2. Forest Fire Susceptibility

The spread of forest fires in closed tropical forest is a growing problem ( Uhl and
Buschbacher, 1985). The consequences of wildfire are devastating for tropical forests
because their thin bark makes them naturally low in resistance to fire (Uhl and Kauffamn,
1990). Both anthropogenic and natural forest disturbances can increase forest fire
susceptibility in the tropics by damaging canopies and, consequently, decreasing forest
moisture content (Cochrane, 2002).

In the tropical forest, fire possibility and intensity increase as the forest becomes
more fi'agmented through selective logging. The increase boundaries between forest and
deforested areas are more exposed to agricultural practices, which use fire for land
maintenance and weed control by clearing woody debris (Uhl and Buschbacher, 1985).
Indeed, agricultural plots in the Brazilian Amazon are the main ignition source for fires
(Uhl and Buschbacher 1985, Uhl and Kauffman 1990, Cochrane 2001).

Once burned for the first time, chances of recurring fire and intensity due to
increased fuel loads and drier microclimate in the interior of the forest (Cochrane and
Shulze, 1998). This fire sequence can rapidly degrade forest areas by destroying their
structure and composition, creating a new land cover type more similar to secondary

regrowth than to tropical forest (Gerwin, 2000).

Cochrane et al. (1999) conducted field studies and a multitemporal analysis of
remotely sensed imagery to understand forest fire dynamics in a case study at Tailandia
municipality in the State of Para. They established two sites of undisturbed forest as
control areas, and eight sites of fire disturbed forests for field observations and
measurements in 1996. They measured the rate of fire spread, fire recurrence, tree
mortality, biomass combustion levels, and fire characteristics (flame heights and depths).

Based on these studies Cochrane et a1. (1999) observed that the first fires within
undisturbed forest are low intensity fire, which moves slowly along the ground and burns
only dry leaf litter. This type of fire kills mostly trees less than thirty centimeters of dbh
(diameter of breast height). The second instance of fire is faster and more intense,
drastically increasing mortality of large trees. Only large, thicker barked tree may
survive. Tree mortality occurred up to two years after the second fire and possibly longer.

Uhl and Kauffman (1990) studied potential for sustained fire events within
different land cover types (undisturbed forest, selectively logged forest, secondary
regrowth, and pasture) at Vitora ranch in Paragominas municipally of the Amazon state
of Para In this research, fuel availability, microclimate, and rates of fuel moisture loss
were measured. The authors concluded that selectively logged forests are more
susceptible to fire because of the high fuel mass and special conditions for rapid fuel
drying created by logging activities. Specifically, the authors found that selective logging
significantly increased fuel mass (180 Mg/ha) compared with other land cover types.
They observed that harvesting 50 m3 ha'l of round wood will result in 150 m3 ha'1 of
woody debris or fuel for potential fires. They also observed that logged forest were

susceptible to fire after 5-6 rainless days during the dry season.

Holdsworth and Uhl (1997) conducted studies of fire impacts upon structure and
composition of selectively logged forests at two sites located in the municipality of
Paragominas in the Amazon state of Para. They observed that fire in logged forest causes
serious damages to forest structure and composition and contributes to additional forest
fiagmentation. Even with low-impact selective logging, which showed reduced
flammability compared with high-impact logging, there was a risk for fire because

adjacent areas of agricultural lands are a frequent source of fire ignition.

1.2.2.3. Biodiversity

Upper AmazonianI forests are considered to be some of the most biologically
diverse biomes in the world (Gentry, 1988). Although only a few valuable tree species
are harvested during selective logging, high intensity logging can have significant impact
to the entire forest ecosystem (Frumhoff, 1995) damaging nearby trees and soils (Uhl and
Vieira 1985, Johns et al 1996) and increasing the risk of local species extirpation (Martini
et al, 1994). Furthermore, hunters will have easier access to internal forests through the
road network created by logging operations, increasing species harvests by commercial
and subsistence hunting, which is usually aimed at primates and duikers2 (Frumhoff,
1995)

According to Martini et a1. (1994), selective logging affects both plant and animal

species, especially those species associated with the species of trees targeted for timber

 

' The term Upper Amazonian here was used by Gentry (1988) as the Peruvian Amazon
and part of the Amazon in the Venezuela-Brazil border.
2 Duikers are small antelopes comprising two genera (Cephalophus and Sylvicapra)

(Merriam-Webster Dictionary, 2003).

10

extraction (e. g. mammals and birds that eat fruits and disperse seeds from timber tree
species). The ecosystem changes caused by selective logging leads to animal
disequilibrium in terms of number of species and species diversity, where some species
tend to be more abundant than others, affecting the ecological dynamics within logged
forests and may also extend to undisturbed neighboring forests (Frumhoff, 1995). In
terms of plants, selective logging tends to eliminate the most valuable timber species and,
systematically, reduces individuals that display those traits (Uhl and Vieira, 1989).

Moreover, selective logging activities cause a loss of forest structural integrity
and cause further degradation, especially in open canopy forests, which have a presence
of vines (Uhl and Buschbacher, 1985). These effects of logging activities also provoke
habitat modification including site desiccation (Martini et al., 1994) and soil compaction
(Pinard and Putz 1996, Fredericksen and Pariona 2002). Such impacts by logging
directly modify the variability of the forest ecosystem ( Uhl and Vieira, 1989), destroy
habitats, and affect the abundance of amphibians and reptiles, which in turn may affect
seed dispersal and forest regeneration (Martini et al., 1994). The highest impact to the
natural habitat occurs in forests heavily exploited by logging that become depleted of
valuable tress and conversed to other types of land use (Pinard and Putz, 1996, Huth and
Ditzer, 2001).

Although selective logging in the Amazon floodplain ofien concentrates on few
tree species, it results in profound ecological changes in the forest environment (Macedo
and Anderson, 1993). They also observed a dramatic transformation of the under canOpy

vegetation occurred after logging, where a secondary vegetation predominantly formed

11

by vines and herbs took place. Additionally, they conclude that there is a trend to increase

logging intensity and, consequently, environmental damages in those areas.

1.2.2.4. Socioeconomic factors

The Brazilian economy is among the 10 largest economies in the world, with a
Gross Domestic Product (GDP) of US$ 596 billion in 2000. The timber sector only
contributes 10% of this or approximately 5.5 billion (ABIMCI, 2002). The timber
industries operating in the Brazilian Amazon contribute approximately 15% of the
regional GDP, employing 5% of the regional workforce (World Bank, 1999). It is
estimated that 61.1% of all productive native forests in Brazil are located in the Legal
Amazon, the main source of raw materials for the timber industries (ABIMCI, 2002).

Between 1992 and 1999 approximately 295 million cubic meters of round wood
was harvested in Brazil’s Amazonian states, an average of 36.88 million m3yr'l (IBGE,
2003). Based on these data provided by IBGE (2003), of this timber production, more
than 81.8% was harvested in the state of Para, 8.8% in the state of Mato Grosso, and
3.7% in the state of Rondonia, 2.2% in state of Maranhao, 1.6% in the state of Amazonas,
and 1.9% in the states of Acre, Amapa, Tocantins, and Roraima. In recent years, the state
of Amazonas has shown significant annual increase in harvested timber volume from
162, 622 to 792 thousand cubic meters of round wood in 1992, 1996, and 1999,
respectively, while all other Amazon states have shown overall reduction (appendix A.l).

In the state of Para only 676 timber companies processed approximately 11.3
million cubic meters of round wood in 1998, which corresponded to 4.25 million cubic

meters of processed wood and 2.8 million harvested trees. The timber industries in the

12

state of Para contributed up to US$ 1.0 billion to the regional GDP in that given year
(Verissimo et al. 2002). More specifically, Verissimo et a1. (1992) observed that 112
sawmills in the vicinity of the city of Paragominas, in Para, provided approximately 5700
jobs, including forest employees (e. g. forest timber extractors, truck drivers, machinery
operators, etc.) and industry employees. More than 50% of the urban population of
Paragominas was directly dependent on the timber industry for its income. Indirectly, the
timber industry generates significant tax revenues, which can be used for the benefit of
the local and regional population. Such work opportunities and economic benefits
attracted many migrants from other neighboring states to come and live in a new frontier.

Pinedo-Vasquez et a1. (2001) conducted a study case in a floodplain forest in the
State of Amazonas, after a period of logging “boom”3 or abundance of most valuable
trees. In this case, the local timber industry appeared to have collapsed after the high
value timber species were exhausted, which led industries and individuals to migrate to
other regions searching, respectively, for raw material and job opportunities.

Verissimo et a1. (2002) studied 24 timber centers located at Paragominas and
Novo Progresso municipalities, in the Amazon State of Para, involving 676 timber
companies by 1998. In that given year, 11.3 millions cubic meters of round wood (around
2.8 million tress) were processed by those timber companies. The authors observed that

timber oriented industrial areas are economically collapsing in Paragominas, an old

 

3 Logging boom is defined as the beginning of the logging activities in a new frontier
region. In this ‘boom’ phase many sawmills are installed nearby abundant sources
(natural forests) of raw material for the timber industries. It is also defined as predatory

and unsustainable forest exploitation activities (Schneider et a1. 2000).

13

frontier, and re-emerging in Novo Progresso, a new frontier surrounded by undisturbed
forest. In those regions where logging is collapsing, forest cover, industry employments,
and government taxes are also abruptly decreasing. Finally, they concluded that in
pursuing the same forest exploitation model, the municipality of Novo Progresso will

follow the same economic growth pattern (boom-collapse) rather than a sustainable one.

1.2.3. Synergism between selective logging and deforestation

Brazil has the world’s highest absolute rate of deforestation (Skole and Tucker
1993, Skole et a1. 1994, Houghton et a1. 2000, Laurance et al. 2001a). Multi-annual
estimates of deforestation in the Brazilian Amazon, provided by the Brazilian Space
Agency - INPE (2003), show a total of 152,200 kmz, 377,500 sz, and 551,782 km2
deforested, respectively, by 1978, 1988, and 1998. These estimates represent an annual
increment of 19,979 km2 or 362% in 20 years. The most recent measurements based on
remotely sensed data show a total of 587,727 sz deforested by 2000, equivalent to
11.75% of total Brazilian Amazon (INPE, 2003).

Studies in the Brazilian Amazon suggest that deforestation is spatially
concentrated in some regions. Indeed, most of deforestation growth occurs around the
major roads (Alves et a1. 1999, Laurance et al. 2001a, b, c) and regions of higher human
population density (Laurance et al. 2001b). Other factors such as rainfall and unpaved
roads seem to have lesser influence on deforestation growth (Laurance et al. 2001b).

In addition to outright deforestation, uncontrolled forest exploitation by loggers
may also catalyze deforestation by opening roads into unoccupied government lands and

protected areas that are subsequently colonized by ranchers and farmers, which will result

14

in increasing deforestation and release of carbon to the atmosphere (Verissimo et al.,
1995).

Uhl and Vieira (1989) studied the indirect socio-economical impacts of selective
logging in a case study from the Paragominas region of the State of Para. They also noted
that the access roads remaining afier selective logging operations cease, lead to the
occupation of the area by landless peasants, resulting in increased deforestation and
threats to indigenous communities.

Uhl et a1. (1991) conducted a case study of selective logging at Tailfindia
municipality in the Amazon State of Para. They observed that although selective logging
directly provoked mild damages in the forest, it contributed to deforestation by improving
the local economy during the initial colonization in this region that otherwise would not
successfully occur.

More specifically, Uhl and Buschbacher (1985) studied the synergism between
selective logging and deforestation for cattle ranch in a case study at Paragominas
municipality in the State of Para. The authors observed that pasture development was
economically subsidized by selective logging through the timber product
commercialization and the required road network for timber transportation. Furthermore,
fire used by farmers to manage crops and pastures often then spread to adjacent forests.
As mentioned earlier, compared to non-logged forest, fire spreads more readily in
exploited forests, causing extensive damages (Uhl and Buschbacher, 1985).

However, Stone and Lefebvre (1998) observed that 5 years after logging the
majority of logged forest was remaining in a case study at Paragominas municipality in

the State of Para. In this case, deforestation does not appear to be following selective

15

logging activities, even though former logging areas have been more fi'equently re-visited

by loggers.

1.2.4. Previous attempts to map selective logging

Remote sensing studies of selective logging have centered on the use of Landsat
TM imagery. Stone and Lefebvre (1998) quantified forests affected by selective logging
in a multitemporal case study area in Eastern Para, Brazil, using Landsat images from
1986, 1988, 1991, and 1995. Visual interpretation of satellite imagery was used to study
how fast selective logging was occurring in areas west and northeast of the city of
Paragominas. Timber transportation roads and patios were digitized on a computer screen
with polygons registered to image coordinates and areas identified as having been
affected by logging. Their results showed that most logged forests became visually
indistinguishable fiom surrounding forests within 3 years. Stone and Lefebvre (1998)
applied texture algorithm and NDVI (Normalized Difference Vegetation Index) using
Landsat TM imagery to investigate whether forest canopy texture and greenness in
logged forests were significantly different fi'om that of undisturbed forest. They
concluded that texture and NDVI images were not helpful in defining selectively logged
forest.

Souza and Barreto (2000) conducted a multi-temporal analysis of 82.5 hectares of
selectively logged forest in the State of Paul, Brazil. They used Landsat TM scenes,
bands 1-5 and 7, from June 1984, July 1991, and July 1996. They applied a linear
mixture model, based on spectrally pure pixels, to estimate the soil, vegetation, and shade

fractions within each TM pixel. Subsequently, they used a morphologically constrained

l6

classification of the soil fraction image to identify logging patios where cut logs are
temporarily stored. Based on fieldwork, Souza and Barreto (2000) determined that a
buffer zone of 180 meters around the detected logging landings (patios and roads)
provided an excellent estimation of the actual forest area affected by selective logging.
However, Souza and Barreto (2000) cautioned that, due to the rapid vegetative regrowth,
their technique detected only 60% of field-verified logging patios that were one year old
and none that were three years old.

J aneczek (1999) estimated the forest area affected by selective logging throughout
the Brazilian Amazonia in 1992, using Landsat TM imagery and a GIS. In this study,
J aneczek (1999) used an automated texture analysis of Landsat TM band 5 to detect
logging landings (patios and roads). Similar to Souza and Barreto (2000), Janeczek
(1999) applied buffer zones of 180 meters around logging landings to estimate areas
affected by selective logging. In addition, J aneczek mapped selective logging activities
(visible canopy disturbance on Landsat images) through visual interpretation, and
circumscribed them digitally on a computer screen. In the analysis, 84% of the logging
landings were detected automatically when compared with the visual identification on the
Landsat images. Areas that were missed using automatic analysis were added by visual
interpretation. A total of 5,406 km2 of forest was estimated to have been selectively
logged over 1-2 years.

Nepstad et a1. (1999) interviewed 1393 sawmills located in 75 timber polos in the
Brazilian Amazon in order to estimate areas affected by selective logging basin-wide.
They used sawmill records of the amount of harvested round wood for 1996 and 1997

and the round wood harvest intensity, expressed in cubic meters of timber per hectare of

17

forest. Based on these information, they estimated that 10 to 15 thousand sz of
undisturbed forests were annually harvested in 1996 and 1997. In this research, Nepstad
et al. (1999) concluded that, besides deforestation, selective logging in the Brazilian
Amazon has been contributing to increase CO2 in the atmosphere.

Asner et al. (2002) conducted a field study of forest canopy damages at Cauaxi
ranch in the Paragominas municipality of the Amazon State of Para, Brazil. The field
studies encompassed 50 hectares of non-logged forest, 200 hectares of high intensity
logged forests, and 200 hectares of low intensity selectively logged forests. They
surveyed and mapped those areas before and after selective logging, where trees and log
landings (roads, patios, skid and trails) were identified and located. They also measured
canopy gap fraction to report leaf area index (LAI). Additionally, they performed visual
analyzes using different bands and texture analysis on Landsat images from different
years and compared with field study results.

Asner et al. (2002) noted that only the highest canOpy damage type of selective
logging damages can be easily detected using Landsat multi-spectral images. In that case
study, Landsat reflectance or texture analysis could detect and estimate forest damage
when the forest gap fraction is less than 50%. They concluded that although basic
Landsat reflectance data and texture analysis could not quantify logging intensity in that
study site, these data and technique are useful for broad scale logging detection, as is
needed for environmental monitoring in the Brazilian Amazon. The author suggested
efforts are needed to develop alternative approaches using high spatial resolution and

hyper-spectral imagery in order to quantify selective logging intensity.

18

1.3. Bibliography

ABIMCI — Brazilian Association for Mechanically Processed Timber. 2002.
Mechanically Processed Timber — Sectorial study 2001. Available fiom
www.synteko.com.br.

 

ACHARD, F., EVA, H.D., STIBIG, H., MAYAUX, P., GALLEGO, J., RICHARDS, T.,
and MALINGREAU, J. 2002. Determination of Deforestation Rates of the World’s
Humid Tropical Forests. Science. 297, 999-1002.

ALVARADO, E. and SANDBERG, D V. 2001. Logging in Tropical Forest: Literature
review on Ecological Impacts. University of Washington. USDA Forest Service.
Internet site www.fs. fed.us/pnw/fera.

 

ALVES, D. S ., PEREIRA, J. , S OUSA, C. SOARES, J., and YAMAGUCHI, F. 1999.
Characterizing landscape changes in central Rondénia using Landsat TM imagery.
International Journal of Remote Sensing. 20, 2877-2882.

ASNER, G. P., KELLER, M., PEREIRA, R. and ZWEEDE, J.C. 2002. Remote sensing
of selective logging in Amazonia - Assessing limitations based on detailed field
observations, Landsat ETM+, and textural analysis. Remote Sensing of
Environment. 80, 483-496.

COCHRANE, M. A. and SCHULZE, M. 1998. Forest Fire in the Brazilian Amazon.
Conservation Biology. 12, 948-950.

COCHRANE, M. A., ALENCAR, A., SCHULZE, M.D. SOUSA jr, C.M., NEPSTAD,
D.C. LEFEBVRE, P., and DAVIDSON, E. 1999. Positive feedbacks in the fire
dynamic of closed canopy tropical forests. Science. 284, 1832-1835.

COCHRANE, M.A. 2001. Synergistic Interactions between Habitat F ragrnentation and
Fire in Evergreen Tropical Forests. Conservation Biology. 15, 1515-1521.

COCHRANE, M.A. 2002. Spreading like Wildfire. Tropical Forest Fires in Latin
America and the Caribbean: Prevention, Assessment and Early Warning (United
Nations Environment Programme (UNEP), Mexico City. Mexico. 96p.

COCHRANE, M.A. 2003. Fire science for rainforests. Nature. 421, 913-919.
DEFRIES, R.S., HOUGHTON, R.A., HANSEN, M.C., FIELD, C.B., SKOLE, D. L., and
TOWNSHEND, JOHN. 2002. Carbon emissions fi'om tropical deforestation and

regrowth based on satellite observations for the 19808 and 19908. PNAS. 99, 14256—
14261.

19

FEARNSIDE, P.M., TARDIN, A. T., and FILHO, L.G.M. 1990. Deforestation in
Brazilian Amazonia. PR/SCT, Instituto de Pesquisas Espaciais, Instituto Nacional de
Pesquisas da Amazonia.

FEARNSIDE, RM. 1997. Greenhouse gases from deforestation in Brazilian Amazonia:
net committed emissions. Climate Change. 35, 321-360.

FEARNSIDE, P. M. 1999. Green Gases from Deforestation in Brazilian Amazénia: Net
Committed Emissions. Climate Change 35, 321-360.

F REDERICKSEN, T. S. and PARIONA, W. 2002. Effect of Skidder disturbance on
commercial tree regeneration in logging gaps in a Bolivian tropical forest. Forest
Ecology and Management. 171, 223-230.

FRUMHOFF, RC. 1995. Conserving Wildlife in Tropical Forests Managed for Timber to
provide a more viable complement to protected areas. BioScience. 45, 456-464.

GENTRY, A. H. 1988. Tree Species Richness of Upper Amazonian Forests. Ecology. 85,
156-159.

GERWING, J. J. 2000. Degradation of forests through logging and fire in the eastern
Brazilian Amazon. Forest Ecology and Management. 157, 131-141.

HOLDSWORTH, A. R and Uhl, C. 1997. Fire in Amazonian Selectively Logged Rain
Forest and The Potential for Fire Reduction. Ecological Aplications. 7, 713-725.

HOUGHTON, RA. 1996. Terrestrial sources and sinks of Carbon inferred fiom
Terrestrial data. Tellus. 48B, 420-432

HOUGHTON, R. A. 1997. Terrestrial Carbon Storage: global lessons for Amazonian
Research. Ciencia e Cultura. 49, 58-72.

HOUGHTON, RA. 1998. The annual net flux of carbon to the atmosphere from changes
in land use 1850 — 1990. Tellus. SIB, 298-313.

HOUGHTON, R.A., SKOLE, D. L., NOBRE, C. A., HACKLER, J. L., LAWRENCE,
K.T., CHOMENTOWSKI, W.H. 2000. Annual fluxes of carbon from deforestation
and regrowth in the Brazilian Amazon. Nature 403, 301-304.

HOUGHTON, R.A., LAWRENCE, K.T., HACKLER, J .L., and BROWN, SANDRA.
2001. The spatial distribution of forest biomass in the Brazilian Amazon: a
comparison of estimates. Global Change Biology. 7, 731-747.

HUTH, A. and DITZER, T. 2001. Long-term impacts of logging in a tropical rain forest —
a simulation study. Forest Ecology and management. 142, 33-51.

20

IBGE — Instituto Brasileiro de Geografia e Estatistica [Internet]. 2003 - Sistema IBGE de
Recuperacao Automatica. Extracao Vegetal. Rio de J aneiro, RJ. Available from
http://www.ibge.gov.br/

INPE- BRAZILIAN SPACE AGENCY. 2003. Monitoring of the Brazilian Amazon by
Satellite. 85o Jose dos Campos, SP, Brazil. Available fi'om http://www.inpe.br/

J ANECZEK, DJ. 1999. Detection and Measurement of Amazon Tropical Forest Logging
using Remote Sensing Data. A Master’s Thesis presented to Michigan State
University, Department of Geography.

JOHNS, J.S., BARRETO, P., UHL, C. 1996. Logging damage during planned and
unplanned logging operations in the eastern Amazon. Forest Ecology and
Management. 89, 59-77.

KAUFFMAN, LB. and CUMMINGS, BL. 1995. Fire in the Brazilian Amazon: 1.
biomass, nutrient pools, and losses in slashed primary forests. Oecologia. 104, 397-
408.

LAURANCE, W.F., ALBERNAZ, A.K.M., and COSTA, C. 20013. Is deforestation
accelerating in the Brazilian Amazon? Environmental Conservation. 28, 305-311.

LAURANCE. W.F., ALBERNAZ, A.L.M., SCHROTH, G. FEARNSIDE, P.M.,
BERGEN, S., VENTICINQUE, E.M., COSTA, C. 2001b. Predictors of deforestation
in the Brazilian Amazon. Journal of Biogeography. 29, 737-748.

LAURANCE, W.F., COCHRANE, M., BERGEN, S., FEARNSIDE, P.M.,
DELAMONICA, P., BARBER, C., D’ANGELO, S., and FERNANDES, T. 2001c.
The Future of the Brazilian Amazon. Science. 291. 437-439.

MACEDO, D. S. and ANDERSON, A. B. 1993. Early Ecological Changes Associated
with Logging in an Amazon Floodplain. Biotropica. 25, 151-163.

MARTINI, A. M. Z., ROSA, N. A., UHL, C. 1994. An attempt to predict which
Amazonian tree species may be threatened by logging activities. Environmental
Conservation. 21, 152-161.

MERRIAN-WEBSTER DICTIONARY. 2003. Springfield, MA, USA. Available from
http://www.m-w.com/cgi-birfldictionary.

 

NEPSTAD D.C., VERISSIMO A., ALENCAR A., NOBRE C., LIMA E., LEFEBVRE
P., SCHLESINGER P., POTTER C., MOUTINHO P., MENDOZA E., COCHRANE
M.A., BROOKS V. 1999. Large Scale impoverishment of Amazonian forests by
logging and fire. Nature. 398, 505-508.

21

NEPSTAD, D., McGRATH, D., ALENCAR, A., BARROS, A.C., CARVALHO, G.,
SANTILLI, M., DIAZ, M.C.V. 2002. Frontier Governance in Amazonia Science’s
Compass. 295, 629-631.

PALM, C.A., HOUGHTON, R.A., MELILLO, J .M., and SKOLE, BL. 1986.
Atmospheric Carbon Dioxide from Deforestation in Southeast Asia. Biotropica. 18,
177-188.

PALM, C.A., WOOMER, P.L. ALEGRE, J. AREVALO, L., CASTILLA, C.,
CORDEIRO, D., FEIGL, B., HAIRIAH, K., KOTTO-SAME, J ., MENDEZ, A.
MOUKAM, A., MURDIYARSO, D., NJOMGANG, R., PARTON, W., RBISE, A.,
RODRIGUES, V., SITOMPUL, S., and NOORDWUK, M. 2000. Climate Chnage
Working Group Final Report, Phase 11: Carbon Sequestration and Trace Gas
Emissions in Slash-and-Burn and Alternative Land Uses in the Humid Tropics. ASB-
ICRAF and Tropical Soil Biology and Fertility (TSBF) Programme. Available from
www.asb.cgiar.org.

PINARD, M. A. and PUTZ, PE. 1996. Retaining Forest Biomass by Reducing Logging
Damage. Biotropica. 28. 278-295.

PlNEDO-VASQUEZ, M., ZARIN, D.J., COFFEY, K., PADOCH, C., and RABELO, F.
2001. Post-Boom logging in Amazonia. Human Ecology. 29, 219-239.

SCHNEIDER, R. ARIMA, E., VERISSIMO, A., BARRETO, P., SOUSA jr, C. 2000.
Amazdnia Sustentavel: lirnitantes e oportunidades para o desenvolvimento rural,
Banco Mundial/IMAZON. Brasilia, Brazil. 58p.

SOUZA, C. Jr. and BARRETO A., 2000. An alternative approach for detecting and
monitoring selectively logged forests in the Amazon. International Journal of
Remote Sensing, 21. 173-179.

STONE, T. A. and LEFEBVRE, P. 1998. Using multi-temporal satellite data to evaluate
selective logging in Para, Brazil. International Journal of Remote Sensing, 19.
25 17-2526.

SKOLE, D. and TUCKER, C. 1993.Tropical deforestation and habitat fragmentation in
the Amazon: satellite data from 1978 to 1988. Science. 260. 1905-1910.

SKOLE, D.L., CHOMENTOWSKI, W., SALAS, W. NOBRE, AD. 1994. Physical and
human dimensions of deforestation in Amazonia. BIOSCIENCE. 44, 314-325.

UHL, C., VERISSIMO, A., MATTOS, M.M., BRANDINO, Z., VIEIRA, CG. 1991.

Social, economic, and ecological consequences of selective logging in an Amazon
fiontier: the case of Tailandia. Forest Ecology and Management. 46, 243-273.

22

UHL, C., BARRETO, P., VERISSIMO, A., VIDAL, E., AMARAL, P., BARROS, A.C.,
SOUZA, C. JOHNS, J., GERWING, J. 1997. Natural Resources Management in the
Brazilian Amazon: An integrated research approach. BioScience. 47, 160-168.

UHL, C. and KAUFFMAN, J .B. 1990. Deforestation, Fire Susceptibility, and Potential
Tree Responses to Fire in the Eastern Amazon. Ecology. 71, 437-449.

UHL, C. and VIEIRA, I.C.G. 1989. Ecological impacts of selective logging in the
Brazilian Amazon: A case study from the Paragominas region of the state of Para.
Biotropica. 21, 98-106.

UHL, C. and BUSCHBACHER, R. 1985. A Disturbing Synergism Between Cattle Ranch
Burning Practices and Selective Tree Harvesting in the Eastern Amazon. Biotropica.
17, 265-268.

UHL, C. 1998. Perspective on Wildfires in the Humid Tropics. Conservation Biology.
12, 942-943.

VERISSIMO, A., P. BARRETO, M. MATTOS, R. TARIFA, and c. UHL. 1992.
Logging impacts and prospects for sustainable forest management in an old
Amazonian frontier: the case of Paragominas. Forest Ecology, and Management.
55, 169-199.

VERISSIMO, A., BARRETO, P., TARIFA, R., and UHL, C. 1995. Extraction of a high
value resource fiom Amazdnia: The case of mahogany. Forest Ecology and
Management. 72, 39-60.

VERISSIMO, A., LIMA, E., and LENTINI, M. 2002. Polos Madeireiros do Estado do
Para. IMAZON. Belem do Para, Brazil. 72p.

WORLD BANK. 1999. I Workshop of Sustainable Production of timber in the Brazilian
Amazon. Business opportunities. Manaus, AM, Brazil.

23

CHAPTER II

Method for Detection and Validation

2.1. Abstract

Selective logging is an important issue in terms of forest degradation and
fragmentation. However, few attempts have been made to detect and estimate the area of
forest where trees were partially removed, as in the case of selective logging. Most
attempts to detect selectively logged forest in the Brazilian Amazon are based on remote
sensing techniques combined with fieldwork information. Although these studies have
important findings in terms of the specific sites, the majority of them are located in the
Amazon State of Para.

This research tested previously developed remote sensing techniques to detect and
estimate forest impacted by selective logging in a case study in the Amazon State of
Mato Grosso. Field studies, automatic analysis, and visual interpretation of Landsat
ETM+ and Ikonos images were used. The results suggest that (1) visual interpretation of
Ikonos imagery can detect most of the forest impacted by selective logging (2) by
combining automatic analysis (textural algorithm plus a buffer of 180 meters) and visual
interpretation using Landsat ETM+ image it was possible to detect around 91.3% of the
selectively logged forest detected using Ikonos imagery (3) the total amount of logged
forest estimated using Landsat image omitted 8.7% and created 19.3% of commission
areas when compared with results obtained using Ikonos imagery, (4) the total logged
forest estimated using Landsat ETM+ is approximately 11% overestimated in comparison
with Ikonos results, (5) the total forest area impacted by selective logging in the study

area is almost twice the amount of actual deforestation.

24

2.2. Introduction

As presented in the previous chapter, the process of logging can result in a heavily
degraded forest environment. In spite of its damages and consequences, few attempts to
detect and quantify areas impacted by selective logging using remote sensing have been
made. Furthermore, most of these attempts were conducted in study cases in the Amazon
State of Para, Brazil, such as the studies by Watrin and Rocha (1992), Stone and Lefebvre
(1998), Sousa and Barreto (2000), and Asner et al. (2002).

Stone and Lefebvre (1998) applied supervised and unsupervised classification and
visual interpretation techniques of Landsat imagery to detect selective logging in a case
study in the State of Para. In this study, the authors could quantify areas impacted by
selective logging by performing visual interpretation of Landsat images. However, the
authors affirmed that images have to be acquired within a short period following logging
and eventual extrapolation to larger areas has to be cautioned.

Sousa and Barreto (2000) developed an alternative approach to estimate selective
logging areas in a case study in the State of Para. Based on field studies, they applied a
buffer radius of 180 meters around log landings detected on satellite images by using a
linear mixture model to estimate forest areas previously logged, but not detectable on
satellite imagery. The authors concluded that further research is necessary to test the
applicability of this methodology for other regions.

In a broader study, J aneczek (1999) applied visual interpretation and automatic
analysis of Landsat imagery in order to estimate selective logging. She quantified the
amount of selective logging in the Legal Amazon Basin for 1992. The author concluded

that these techniques could be improved by integrating with high spatial resolution

25

satellite imagery, such as that provided by Ikonos, which would increase accuracy of
estimating selectively logged areas.

In a more recent study, Asner et al. (2002) compared fieldwork information with
Landsat textural analysis to assess selective logging in a case study also in the State of
Para. The authors conclude that even though textural analysis is useful for broad
delineation of logged forest, this technique cannot assess the intensity of canopy damage.
They suggested new approaches using high spatial resolution and hyperspectral satellite
imagery in order to improve assessment of selective logging impacts.

In this research, I tested remote sensing and Geographic Information System
techniques to detect and measure forests impacted by selective logging in a study case in
the Amazon State of Mato Grosso, Brazil. Logged forests were mapped using satellite
imagery fi'om different sensors (Landsat ETM+ and Ikonos). Automatic detection
deve10ped by Janeczek (1999) plus a 180 meter buffer radius suggested by Sousa and
Barreto (2000) as well as visual interpretation using Landsat image were applied.

Applying visual interpretation to Ikonos imagery and complementary field
observations, the total area of forest impacted by selective logging was digitized. I
mapped logged forests into different GIS layers. Finally, I compared and analyzed the
results of different remote sensing techniques in order to assess their efficiency and

accuracy.

26

2.3. Methodology

2.3.1. Site description

The study area is located in the Brazilian Amazon State of Mato Grosso, about 55
kilometers North of the city of Sinop at 11° 34’ S latitude and 54° 39’ W longitude
(center point). This area encompasses 4,893.00 hectares (approximately 7 Km x 7 Km),
completely covered by an Ikonos scene and by part of a Landsat scene. In this analysis,
the Ikonos scene defines a portion of the Landsat scene and the study area. Most of this
study site is part of the Continental ranch in Sinop municipality of Mato Grosso State
(appendix B. 1).

The climate at Continental ranch is humid tropical with annual precipitation
averaging 2000 mm. The mean annual temperature is 24° C. The predominant natural
vegetation is semi-deciduous forest with emergent canopy, on dystrophic red-yellow
Latossols (RADAMBRASIL, 1980). The land use has been partially changed from

undisturbed forest to pasture, reforestation, and selectively logged forest.

2.3.2. Data Set

The data used in this study are drawn from the collection of Amazonian imagery
at the Center for Global Change and Earth Observations (CGCEO), Michigan State
University. This included deforestation layers and digital Ikonos and Landsat ETM+
images. The Lansat and Ikonos images were acquired on April 30 and June 18, 2000,
respectively These dates roughly correspondto the end of the wet season and to the
beginning of the dry season in this part of the Brazilian Amazon. Additional Landsat

images from 1992 to 1999 were also examined to search for selective logging in the

27

previous years. The satellite images (appendix 32) covering the study site, evidencing
selective logging, free of clouds and shadows, were used in this analysis.

The deforestation GIS layer was generated using unsupervised classification of
Landsat imagery. This layer was reviewed and updated using the high spatial resolution
Ikonos pan-sharpened image. Subsequently, the deforestation layer was used to mask
non-forest and forest, whether selectively logged or not, on both Ikonos and Landsat

images.

2.3.2.1. Calibration and Geometric Rectification

Geometric rectification was done on the Ikonos scene using Bicubic resampling
method, supplied with the Ikonos images at the time of ordering. Additionally, the Ikonos
image rectification was tested using collected GPS ground points in July 2002, from
several locations in the study area. A portion of the Landsat scene covering whole study
area also was rectified using control points derived from Ikonos image scene and nearest-

neighbor resampling method.

2.3.2.2. Pan-Sharpening Ikonos image

Pan-sharpening is a remote sensing technique that improves the spatial resolution
of a multi-spectral image by using a higher spatial resolution band. This spatial resolution
improvement can be done by using principal component merge and nearest neighbor
resampling techniques, which improves the spatial resolution and maintains the spectral

information (Erdas, 1997).

28

In this specific case, the resolution merge technique combined one panchromatic
high spatial resolution (1 m) Ikonos image with another multi-spectral lower spatial
resolution (4 m) Ikonos image, which generated a new multi-spectral image with 1 m

spatial resolution.

2.3.3. Detection of areas of selective logging
To detect and map selectively logged forest, visual interpretation was used on
both Ikonos and Landsat image. Additionally, automatic analysis (texture algorithm +

buffer zone) was applied to the Landsat image, band 5.

2.3.3.1. Digitization of areas of selective logging

Visual interpretation to detect selectively logged forests was done using RBG
3/4/2 and 5/3/2 color composites of pan-sharpened Ikonos and multi-spectral Landsat
image, respectively, displayed at full resolution on a computer screen. The logged forests
were digitized manually on each image at varied scales from 1:1,500 to 1:100,000 using
Arc/Info. The minimum scale for digitizing manually on Landsat image was 1:25,000.

The logged forests were identified by obvious canopy degradation, since logging
activities leave log landings, and tree-fall gaps along with obvious canopy disturbance.
According to Janeczek (1999), this land use creates a characteristic pattern of white
points on the Landsat images, which are patios or roads, embedded in the red hues of the

forest canopy (figure 2.1).

29

 

Undisturbed forest

Previously logged .A
forest (1999).. 5 ’/ '

Ongoing logging a 7,

Secondary Previously logged

“my“ forest (1998)</’

'\
:1

 

Scale

1 0 l 2 Kilometers

 

 

 

Figure 2.1. Visible logged forest areas on Landsat ETM+ image, RGB 5/3/2,

June 18, 2000.

According to Janeczek (1999), areas of obvious logging have well defined

logging roads and patios and extensive canopy degradation. Areas of subtle logging, have

30

lesser degrees of canopy disruption or visible infrastructure, either due to being early in
the logging process or because of substantial forest regrth in previously logged forests.
In this research, visible logged forest includes spectrally bright patios, roads, and
obvious canopy disturbance as well logged areas that exhibit faded log landings. The
areas around the log landings together with areas of obvious canopy degradation were
digitized as polygons into vector GIS layer, classified as logged forest. The digitizing was

done on the very edge of canopy disturbance (figure 2.2).

 

Deforestation

Undisturbed forest
Secondary regrowth "

i
Selectively logged forest/

Deforestation

 

Scale

1 0 1 Kilometer

 

 

 

 

Figure 2.2. Visible logging on Ikonos image RGB 3/4/2, 2000.

Logged forests that did not have visible canopy disturbance on satellite images

were not digitized. In this analysis, a single individual made all visual interpretations.

31

2.3.3.2. Automatic detection of selective logging

As previously discussed, log landings are small clearings cut into the forest for the
purpose of temporary storage and staging areas (patios) for timber transportation to the
sawmills by loggers (Stone and Lefebvre, 1998, Sousa and Barreto, 2000).

J aneczek (1999) tested a textural algorithm on Landsat bands 3, 4, and 5
individually (red, near infrared, and middle infrared, respectively) to detect forests
directly affected by selective logging activities or log landings. Log landings were
identified most effectively with band 5, because dry bare soil reflects more incoming
radiation at that wavelength than vegetation, resulting in a good contrast on the images.

According to Pratt (1991) cited in ERDAS (1997), the textural algorithm can be
used to segment an image and classify its segments, giving the image sharper edges. It
generally indicates the spatial variation in neighboring pixel values. The addition of
texture, to an image, adds structural information that assists in the detection of cryptic
deforestation.

In order to automatically detect patios and logging roads in the areas affected by
selective logging, a textural algorithm was applied to Landsat imagery of the study area.
The texture analysis algorithm used in this study produced a variance from texture
analysis on band 5. The texture algorithm used was the 5 x 5 moving window operator
described above. All details and steps applied in the automatic analysis to detect logged
forest, including texture analysis, noise reduction, mask, buffer zones, and coverage

overlaps, are shown in figure 2.3.

32

 

 

 

  
    

Deforestation
GIS layers

Landsat imagery

 

 

 

Forest and non-forest

Landsat 333d 5 ‘ R5911” 1.. (regular and expanded

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(deforestation mask) mask 3 pixels grid)
Texture Analysis _~ Logging GIS layers
(band 5) (edition)
Noise reduction Building buffer zone
(texture image) around log landings
Texture image ‘ Expanded H Deforestation and

(deforestation mask) logging overlap

 

 

 

3 pixels mask

 

 

 

I I

 

 

 

 

 
 
    

 

 

 

 

 

Final texture image f , ,
(reclassified) Selective Logging
(final layers)
Image to Polygon I
(conversion)

 

 

 

 

Figure 2.3. Methods used in the logging automatic detection

33

 

As part of the deforestation analysis, the image was classified by the Center for
Global Change and Earth Observations (CGCEO) using an unsupervised image
classification model of bands 2, 3, 4, and 5, into seven thematic classes: forest,
deforestation, regeneration, cerrado, clouds, shadows, and water. Non-forest areas
(deforestation, cloud, shadow, cerrado, water body, and regenerating vegetation) were
subsequently masked out of band 5 of the Landsat image. The texture analysis was then
run on the masked image (band 5) using a variance algorithm with a 5x5 window. A
median filter was applied to the resulting texture analysis to reduce ‘noise’. Thus, the
automatic analysis (textural algorithm) to detect selective logging using Landsat and
Ikonos images was applied solely on areas classified as forest, either selectively logged or
undisturbed.

The images were masked a second time using an expanded non-forest mask.
Specifically, the mask was expanded by 3 pixels so as to remove artifacts in the texture
analysis caused by the 5x5 variance window near forest/mask edges. This greatly reduced
the editing time necessary.

The texture layer was then reclassified using ERDAS, selecting and capturing the
pixels values (mostly from 6 to 11), correlated to areas directly affected by selective
logging, such as patios and roads (figure 2.4). Janeczek (1999) tested bands 3, 4, and 5
using 3x3 and 7x7 windows in the texture analysis algorithm, but the 5x5 window

provided the best results.

34

 

 

 

 

 

 

 

 

 

 

g 20 LL“ 1;. .23... . ,.. 1:73“ _ A . , I. -7 L 1A... $30-$2- “ '7 .3." "
to . .m: - .32 -.. . ‘ 1‘” '~ - '
> F . i’ - a ‘- C: q
E 10 A " I : T F.
X . I
E - 1 . ' ~ . EF
0 H.431;
0 2000 4000 6000 8000

Distance (meters)

 

 

 

Figure 2.4. Spatial profile of a textural image (Landsat ETM+ , band 5).

The classified texture images were converted to vector format and edited to
remove any remaining extraneous features that were not associated with logging. A
buffer radius of 180 meters, as suggested by Souza and Barreto (2000), was applied to the

detected log landings (patios and roads) in order to estimate the amount of forest actually

35

affected by logging (figure 2.5). Janeczek (1999) described those areas as “cryptic

logging”.

 

 

1 0 1 2 Kilometers

 

 

 

Figure 2.5. Textural image (band 5) and buffer radius of 180 meters

around the log landings (patios and roads).

Subsequently, the non-forest mask was applied again to remove any regions
where the suspected-logging areas overlapped with non-forested land cover classes.
Different buffer radiuses of 150, 210, 240, and 270 meters were also tested to assess the

efficiency and accuracy of automatically detecting selectively logged forest.

36

2.3.4- Logging Area Estimates on Landsat imagery

The final estimate of logged forest areas was made using the union of the logged
forest polygons detected by the automated analysis and visual interpretation. The overlap
between areas of logged forest and non-forest (deforestation) was removed to avoid

double counting.

2.4. Field studies

This field study was conducted during July 2002, the dry season in the Brazilian
Amazon. These studies consisted basically of empirical observations of the previously
logged forests identified on the satellite images. Measurements and personal
communication with the landowner and technical staff fi‘om the Continental ranch
responsible for enforcing the logging operation plans and activities were also conducted.
Areas of logged and non-logged forests were identified by empirical observations of
selective logging evidence such as former patios and roads and damaged forests (tree-fall
gaps, stumps, wood debris, etc.). Patio and road dimensions were measured in different
sites several years after logging. Forest regeneration within previously logged forest was
also examined.

By examining Landsat imagery from 1992 to 2000, no evidence of selective
logging activities prior to 1997 were found. From 1998 to 2000, selective logging became
a predominant land use type in the area. In personal communication with the Mr. Annor
Zanchette (landowner of the Continental ranch), it was confirmed that in the study area

selective logging had begun by 1998.

37

Undisturbed forests were also observed in 2002 during the field studies, though
they are located mostly in wetlands and areas protected for scientific purposes.

The harvested volume in the study site was around of 40 m3/ha, involving roughly
50 tree species (personal communication with Mr. Armor Zanchette, 2002). The average
diameter of the 10 patios in logged forest prior to 2001, was 61.7 meters. The average
width of former access roads in the same logged forests was 10.7 meters.

Tree fall gaps, stumps, logging access roads, Skidder trails, and patios were quite
evident a year after selective logging activities, though forest regeneration or secondary
regrowth was already taking place. Two years after logging, the patios and roads were
still visible, but showing a stronger regeneration and most of the tree fall gaps and
Skidder trails were covered by secondary regrowth (vines, fast-growth tree species, etc).
Three years after selective logging, most of the forest damage caused by logging
activities were covered by strong secondary regrowth.

However, some areas severely impacted by heavy trucks and other machinery,
such as main access roads and patios, lacked regeneration. These areas were evident
because of the compacted soil and soil exposure left behind by logging activities. Four
years after logging, forest regeneration was very strong and only a few areas were
showing partial soil exposure, though it was quite clear that those areas were previously
logged because of the damaged forest canopy and secondary regrowth. Wild fire does not

seem to have affected the study area so far.

38

2.5. Results

General results of this research showed that 14% of the study area had been
deforested by 2000 and 60.1% had been preserved as undisturbed forest (table 2.1).
Additionally, a total of 1,266.5 hectares (25.9%) of selectively logged forests were
detected by using visual interpretation of an Ikonos image. This area was adopted as
ground truth, hence it represented the total of selectively logged forest, used in this

analysis to compare with results obtained using Landsat imagery.

Table 2.1. Land use in the study area by 2000

 

 

 

Land Use Area (ha) %
Logged forest 1266.50 25.9
Undisturbed forest 2939.80 60.1
Deforestation 686.68 14.0
Total 4,894.81 100.0

 

Source: Visual interpretation on Ikonos image

The deforestation layer was used to mask out non-forest on satellite imagery,
hence the logging detection methodologies were applied only for forest areas, either
selectively logged or not.

Small features lefi behind in ongoing selective logging areas such as tree-fall
gaps, and Skidder trails were detected by visual interpretation on pan-sharpened Ikonos
image, hence the border between logged and undisturbed forest was well defined. Logged

forests prior to 1999 were also detected by visual interpretation on the Ikonos image, but

39

the definition of the border between logged and undisturbed forest was made more
difficult by forest regeneration. Logged forests in 1998 showed a few spots with soil
exposure. The former log landings (patios and access roads) were substituted by fast-
growing secondary vegetation. Thus, besides log landings, the additional criteria to
differentiate logged an undisturbed forest was the presence of secondary regrowth in

between the tree-canopies.

2.6. Analysis of accuracy assessment of the remote sensing techniques

In this research visual interpretation and automatic analysis of Landsat imagery
techniques, developed by J aneczek (1999), were applied in order to detect selective
logging. For the automatic analysis method, a different buffer radius, varying from 150
meters to 270 meters, in multiples of 30, according to the Landsat spatial resolution were
tested. The overlap, omission, and commission areas mapped using Landsat image were
than compared with the ground truth (Ikonos + field work) results (appendix B.2).

Visual interpretation only detected around 76.4% of the total logged forest
(1,266.5 hectares) estimated using Ikonos and complementary field studies. By increasing
the buffer radius the omission area decreased and the commission area increased, creating
an overestimation of the total of logged forest. By increasing the buffer radius from 150
meters to 270 meters, detection of logged forest significantly increased from 74% to
92.5% of the total area, and the omission area decreased from 26.6% to 7.5%,
respectively. When combining visual interpretation and automatic analysis using different

buffer radiuses and Landsat image, the detection of selective logging areas increased

40

from 88.4% to 97.1% and the omission area decreased from 11.6% to 2.9%, for buffer

radiuses ranging from 150 to 270 meters in 30 meter intervals (figure 2.6).

 

 

     
  

 

   
     

  

 

160.0

140.0 AVG” ,

120.0

100.0 “- -
s- 80.0» H H l1 [I I1

60.0 ~
40.0 —
20.0 -

0.0 —

 

 

--
Buffer 150 m + visual ,- _l--
--
. --
--
Buffer 210 m + visual _.--
--

Buffer 240 m + visual _‘-
--

Buffer 270 m + Visual ——J-

    

            

Visual only
Buffer150m L A, 9;
Buffer 180 m c
Buffer 210m , .

Buffer 180 m + visual

Technique

 

! Logging area detected I Commission areas

 

 

El Omission areas D Logging area estimated.

 

 

 

Note: Total logged forest (Ikonos): 1,266.5 hectares (100%)

Figure 2.6. Selective logging detection using visual interpretation and automatic analysis

on Landsat ETM+ , testing different buffer radiuses.

Specific results of testing visual interpretation, automatic analysis using a

buffer radius of 180 meters developed by Sousa and Barreto (2000), and both techniques

combined are presented as following (table 2.2).

41

Table 2.2. Logged forest detected using visual interpretation and buffer radius of 180

meters on Landsat ETM+ image

 

Technique

 

Logged forest Visual + Automatic

Visual Automatic
Area (ha) % Area (ha) % Area (ha) %

 

 

Detected (overlap between

 

Ikonos x Landsat) 930.3 73.5 1,025.0 80.9 1,156.7 91.3
Commission 19.1 1.5 238.7 18.9 244.8 19.3
Omission 336.1 26.5 241.5 19.1 109.8 8.7
Total 949.4 75.0 1,263.7 99.8 1,401.6 1 10.7

 

Note: Total logged forest (Ikonos): 1,266.5 hectares (100%)

 

Using automatic analysis (textural analysis plus a buffer radius of 180 meters) 1
could detect more selectively logged forest than visual interpretation, 80.9% and 73.5%,
respectively, resulting in less omission areas (underestimation) than visual interpretation,
19.1% and 26.5%, respectively. However, automatic analysis created more commission
areas (overestimation) than visual interpretation, respectively, 18.9% and 1.5%. When
compared the total area (detected plus commission areas) of logging estimated using
automatic analysis, visual interpretation and both techniques, automatic analysis showed
to be more accurate in estimating the total logging areas (99.8%) than visual
interpretation and both techniques (75% and 110.7%, respectively). In spite of
overestimating in 10.7% the total logging areas, when combining automatic analysis and
visual interpretation, it substantially increased precision by reducing omission areas from
19.1% and 26.5% (automatic analysis and visual interpretation only, respectively) to

8.7% (figure 2.7).

42

 

Undisturbed forest

Logging detection ,1" , I
by Landsat 1” '
\34’“ ‘V‘ .-

~

Logging detection [if A I ‘
by lkonosa ;. ' ii. a...“

 

i — —O 1 2 Kilometers

 

 

 

 

Figure 2.7. Selective logging detected on Ikonos and Landsat ETM+ imagery

displayed on Ikonos image RGB 3/4/2, April 30, 2000.

Automatic analysis only detected 99.8% of the total of logged forest while visual
interpretation underestimated by 25.0% and both techniques overestimated by 10.7%.
Visual interpretation only contributed about 11% in detecting selectively logged forest
while automatic analysis about 20%. Both techniques overlapped around 69% in areas of

selective logging (figure 2.8).

43

 

Visual
a interpretation
7 I only
( 1 1%)

Automatic
\7 analysis
only
(20%)

Overlap
between visual
interpretation J/
and automatic
analysis
(69%)

   

 

 

 

Figure 2.8. Visual interpretation only, automatic analysis only (buffer radius of 180 m),

and overlap between logging areas detected by both techniques.

2.7. Conclusion

Based on these research results and fieldwork observations, forest disturbances by
selective logging activities are visible and measurable using high spatial resolution (lm
pan-sharpened) Ikonos and Landsat imagery. Ongoing or recent selective logging damage
such as patios, roads, Skidder trails, tree-fall gaps, etc, are visible on Ikonos imagery
while on Landsat imagery only patios, roads, and the heaviest canopy damage are visible.
Hence, visual interpretation on Ikonos imagery and complementary field studies could
support a reliable estimate of selective logging for the study site. The Ikonos product
therefore is useful to test other sensors and satellites with coarse spatial resolution.

In spite of the spatial resolution limitation, the Landsat ETM+ image showed

reasonable efficiency and accuracy in estimating logged forests using different remote

44

sensing techniques. Although automatic analysis alone created more commission errors,
which varies according to the buffer radius size, it showed to be more efficient than
visual interpretation in estimating selectively logged forests.

Although automatic analysis using only a buffer radius of 180 meters showed to
be more accurate in estimating logging areas, the combination of visual and automatic
analysis using the given buffer radius, showed to be more precise by significantly
reducing the omission areas from 19.1% to 8.7%, even though it overestimated in 10.7%

the total logging areas. Indeed, combining visual interpretation and automatic analysis

 

(buffer radius of 180 meters) 1 could detect 91.3% of the total logged forests detected
using the Ikonos image and fieldwork observations in this study site at Sinop
municipality in the State of Mato Grosso.

Visual interpretation alone showed to be more interpreter dependent and results
are expected to vary between individuals. Additionally, neither visual interpretation nor
automatic analysis on Landsat images could detect small forest disturbances, detectable
using Ikonos imagery.

Finally, both automated detection and visual interpretation combined form an
efficient methodological approach for estimating areas affected by selective logging in
similar sites. However, the amount of selective logging detected, commission, and
omission areas are variable according to the buffer radius. It is important therefore to
dedicate more effort to testing different buffer radiuses for other sites or regions,
considering that the Brazilian Amazon is a vast area with different ecosystems, land use,

and logging patterns and intensities.

45

2.8. References

ASNER, G. P., KELLER, M., PEREIRA, R. and ZWEEDE, J .C. 2002. Remote sensing
of selective logging in Amazonia - Assessing limitations based on detailed field
observations, Landsat ETM+, and textural analysis. Remote Sensing of
Environment. 80, 483-496.

ERDAS. 1997. Erdas Field Guide. 4‘h edition. Atlanta, Georgia.

JANECZEK, DJ. 1999. Detection and Measurement of Amazon Tropical Forest
Logging using Remote Sensing Data. A Master’s Thesis presented to Michigan
State University, Department of Geography.

RADAMBRASIL, 1980. Mining and Energy Ministry. Department of Mineral
Production. Levantamento dos Recursos Naturais. Folha SC.21. Juruena. Volume
20.456p.

SOUZA, C. Jr. and BARRETO., 2000. An alternative approach for detecting and
monitoring selectively logged forests in the Amazon. International Journal of
Remote Sensing, 21, 173-179.

STONE, T. A. and LEFEBVRE, P. 1998. Using multi-temporal satellite data to evaluate
selective logging in Para, Brazil. International Journal of Remote Sensing, 19,
2517-2526.

WATRIN, 0.8. and ROCHA, A.M.A. 1992. Levantamento da vegetacao natural e do uso

da terra no municipio de Paragominas, Para, utilizando images Landsat (TM).
Boletim de Pesquisa, 124, EMBRAPA/ CPATU, Belém, PA, 40p.

46

CHAPTER III

Assessment of selective logging detection for the Amazon State of Rond6nia

3.1. Abstract

This Chapter assesses the methods for estimating forest areas affected by selective
logging in the Amazon State of Rondénia. I combined a detailed field study with remote
sensing techniques to estimate selectively logged forest areas. The study site
encompassed 2,460 hectares of logged forest, where 37,966 commercial trees were
inventoried prior to harvesting. Post-harvesting inventory reported that 4,138 commercial
trees were logged. The trees, access roads, and log storage patios were properly located
and plotted. Different buffer radiuses were tested around patios located during the
fieldwork in order to quantify the spatial overlap between the buffer zones and selectively
logged forest areas. An automated method (textural algorithm) on bands, Landsat ETM+,
path 232 row 066, was applied to detect forest impacted by selective logging. Different
buffer radiuses around log landings (patios and roads) detected by the automated method
on Landsat imagery were compared with field information. I found that a 540 meter
buffer radius around the patios located on the ground could estimate about 90% of the
forest affected by logging, creating no commission areas. By applying a 450 meter buffer
radius around log landings captured on satellite imagery, it overlapped about 89% of the
forest affected by logging, creating no commission areas. Visual interpretation was also
tested but it was unsuccessful because selective logging in this study site was not
intensive and therefore canopy degradation was not enough to be visible on Landsat

imagery.

47

 

3.2. Introduction

Selective logging is a process of removing a limited number of trees that leaves
behind a mixture of undisturbed forest with canopy damaged forest and soil exposure
areas such as access roads, storage patios, tree fall gaps, and Skidder trails (Uhl and
Buschbacher 1985, Stone and Lefebvre 1998, Nepstad et al. 1999, Souza and Barreto
2000).

Various alternative approaches based on remotely sensed imagery have been
developed in order to detect and estimate areas of selective logging in the Brazilian
Amazon. Most of these attempts have been conducted in the State of Para, for example
the case studies by Watrin and Rocha (1992), Stone and Lefebvre (1998), Sousa and
Barreto (2000), and Asner et al. (2002).

Sousa and Barreto (2000) applied linear texture model to detect log landing on
Lansat images. In complementary field studies, they defined a 180 meters buffer radius
around areas of evident soil exposure, which estimated accurately areas affected by
selective logging in that study site. However, the authors affirmed that further work is
necessary to test the applicability of the methodology for other sites.

Janeczek (1999) developed the first approach for detecting selective logging
Amazon-wide, using remotely sensed data. In this research, J aneczek (1999) improved a
procedure to locate and estimate the extent of cryptic logging and degradation in the
Brazilian Amazon. The author applied techniques from previous studies (visual
interpretation) and developed an automated method to detect logged forests. The
automated method was based on textural analysis of the Landsat, band 5, and

incorporated the 180 meters buffer radius studied by Sousa and Barreto (2000) for the

48

case study in the State of Para. Although J aneczek’s results were successful in that
research, she concluded that “it would be beneficial to firrther test the accuracy of the 180
meters, such as measuring the distance between patios to determine the spatial
variability”.

Specifically in the State of Ronddnia, Janeczek (1999) classified this State as
“low logging region”, because she could not detect a significant amount of logged forest
using her procedures and methods. In spite of this result, the State has been

systematically logged and deforested since the 1970’s. The total deforested area increased I

 

from 420 Km2 in 1978 to more than 5000 Kin2 in 1999 (INPE, 2003). More than 14.9
million m3 of round wood was exploited from natural forest in the period of 1990-2002,
an average of 1.2 m3 yr" (IBGE, 2003).

The applicability of these methods and techniques for other regions therefore
requires additional examination and validation. The purpose of this study was to test the
methods and techniques for estimating areas affected by selective logging activities in the
Amazon State of Rondénia. I conducted a detailed field study in 2,460 hectares located at
Manoa ranch in the Cujubim municipality. I also tested visual interpretation and the
automated method developed by Janeczek (1999) on Landsat ETM+ band 5. Finally, the
different buffer radiuses were compared with fieldwork and remote sensing results for

validation.

49

3.3. Methodology

3.3.]. Study site and data set

This study was conducted at the Manoa ranch, located in Cuj ubim municipality of
the Amazon State of Rondénia, Brazil, approximately 70 kilometers North of the city of
Cujubim at (center point) 8° 59’ S latitude and 62° 20’ W longitude (appendix CI). The
study area encompasses 2,460.86 hectares of selectively logged forest.

The climate at Manoa ranch is humid trOpical. The mean annual precipitation is
2,400 mm. A dry season extends from June through September. The mean annual
temperature is 26° C. The predominant soil types are red-yellow and yellow latossols
covering with open canopy tropical forest (RONDONIA, 2002). The predominant land
uses are undisturbed forest and a small amount of logged forest.

Digital Landsat ETM+ image (path 232, row 066) fi’om the Center for Global
Change and Earth Observation at Michigan State University, acquired in August 5, 2002
was used. This Landsat scene covering the entire study area was rectified using ERDAS
nearest-neighbor resampling method and 15 ground control points. The control points

were acquired using GPS Trimble Pro-XR a Trimble GeoExplorer 11.

3.3.2. Field studies

This study area was previously forest inventoried in 1999. A forest engineer,
Vilmar Ferreira, and a forest technician, Henninio Femandes da Silva Neto, conducted
this inventory. This forest area was selectively harvested during the dry season (May to

October) of 2001. Eraldo Matricardi, Walter Chomentowski, Stephen Cameron, and

50

Heminio F. da Silva Neto conducted complementary post-logging field verification in
June 2002.

During the last field verification in June 2002, almost a year after logging, the
forest regeneration was very strong. The patios and roads were still visible and most of
the tree fall gaps and skidder trails were covered by secondary regrowth such as vines
and fast-growth tree-species. The complete soil exposure was observed only on the main
and secondary roads and partial soil exposure on storage patios. There was no evidence
of wild fire in this study area.

All patios and access roads were located using GPS. All trees with Diameter at
Breast High (DBH) of more than 45 centimeters were identified, measured, and located
within the forest. Transects were plotted within the forest area heading North direction,
100 meters distant from each other. During the previous forest inventory, trees were
located within the ranges of 50 meters in the left and right side of each transect. After
logging, the location of logged trees was once again examined. The total of 4,138 trees
were harvested from a total of 37,966 commercial trees inventoried. The total volume
harvested was 27,847 m3 of round wood, which correspond to an average of 11.3 m3 ha",
considered a low intensity selective logging. Subsequently, the log landings and tree
locations were plotted on a map (appendix C.2).

A total of 65 patios were built in the study area. The patio diameters vary from 25
to 35 meters taking into account the edge-affected forest. The width of main access road
varies from 10 to 12 meters wide, while a secondary access road ranges from 6 to 8
meters wide. The skidder trails are 2 to 3 meters wide and the forest canopy covered most

of them by June 2002. According to Vilmar Ferreira (in personal communication during

51

 

the fieldwork) the tree-fall gaps vary in dimensions depending on the size of the
harvested trees, though an average of 300 m2 of undisturbed forest are damaged by
harvesting a single tree.

Based on a possible relationship between the distance of the harvested trees and
the log storage patios, different buffer radiuses varying from 150 to 720 meters were

tested. The buffer zones were then used to estimate the amount of logged forest.

3.3.3. Detection of selectively logged forest

 

As presented in the Chapter 2, the technique developed by J aneczek (1999) was
applied to this case study. A textural algorithm was performed on band 5 of Landsat
ETM+. The textural algorithm can be used to segment an image and classify its
segments, giving the image sharper edges. It generally indicates the spatial variation in
neighboring pixel values. The addition of texture, to an image, adds structural
information that assists in the detection of cryptic deforestation (ERDAS, 1997).

The textural image creates a good contrast between dry bare soil and forest
canopy. This image was re-classified to separate log landings from the undisturbed forest
canopy and, subsequently, was converted to Arc/Info coverage in order to remove any
remaining extraneous features that were not associated with logging.

Different buffer radiuses were tested. I used a multiple of the Landsat ETM+
pixel size (30 m) for the buffer routines, varying from 150 to 720 meters. These buffer
radiuses were applied around log landings that were detected using automatic analysis

(textural algorithm) in order to estimate the amount of forest areas actually affected by

52

logging but not necessarily visible on satellite imagery. Janeczek (1999) described those
areas as “cryptic logging”.

Although logging roads and patios were easily observed by visual interpretation,
this technique was unsuccessfirl in this study area because of the low intensity logging
characteristics, which do not create significant amount of soil exposure or canopy

damage to be detectable on Landsat imagery.

3.4. Buffer radius assessment

According to the field study results 2,460.86 hectares were logged until June
2002. Approximately 27 hectares was classified as log landings (patios and timber
access roads) showing evident soil exposure. The storage patios only occupied 7 hectares.
The total of 65 patios were located and measured in the study area, showing an average
of 32.5 meters of diameter. The total of 20 hectares was occupied by timber access roads,
showing the total length of 9.3 kilometers and the width of 11 meters for the main access
roads and the length of 13.6 kilometers and the width of 7 meters for the secondary
access roads.

Although Stone and Lefebvre (1998) and J aneczek (1999) could detect selective
logging using visual interpretation in the Amazon state of Para and basin-wide,
respectively, in this study this technique did not produce satisfactory results because of
the low intensity selective logging observed at the study site.

Sousa and Barreto (2000) tested different buffer radiuses around log landings in
order to estimate selectively logged forest for a case study in the Amazon State of Para.

They concluded that a 180 meter radius was the optimal distance for the buffer. In the

53

 

research presented here, I tested different buffer radiuses, varying from 150 meters to
720 meters, around storage patios located during the fieldwork. The buffer zones were
spatially overlapped with the logged forest area to determine the best buffer radius for
estimating selective logging. The overlap between areas of logged forest and within the
buffer zone was identified as logged forest detected. Additionally, I measured the
omission and commission areas of selective logging, created by the buffer zones (figure

3.1).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

i , ~.
5: .. 1:3" . 1:) ' '.. Q
OVd’lap \
N LEGEND .. 1 ' " ~.
. I .’ ‘ ‘ . O .
A N Studyarea(border) ‘ -, _‘ ’fi ' ~ "
I: 540metersbufi‘erradius 'j L _‘
E21 Storagepafios Ornrssrorr,‘ '. ~VL'.’
Logged trees ' Commission
0.3 0 0.3 0.6K!!!
======

 

 

 

 

 

 

 

Figure 3.1. Buffer radius around patios intersecting selectively logged forest

54

The buffer radius of 150 meters overlapped only 27.7% of the total logged trees
(stumps) and 22.7% of the total logged forest area from the ground truth data. This buffer
radius omitted 77.3% of the total logged forest and 72.3% of the total stumps (logged
trees). The buffer radiuses from 180 to 540 meters significantly increased detection of
selectively logged forest from 29.3% to 89.4%, respectively. Additionally, these buffer
radiuses reduced the omission areas from 70.7% to 10.6%, respectively, and showed no
commission errors (appendix C.4).

Buffer radiuses from 570 to 720 meters showed an increase in the detection of

 

selectively logged forest from 91.6% to 97.8%, respectively, and a slight decrease in the
omission area, from 8.4% to 2.2%, respectively. Practically 100% of the logged forest
and stumps were detected using 720 meter buffer radius. However, this buffer distance
created 2.2% and 21.9% of the omission and commission errors, respectively, which
represent a reduction of only 0.9% of the omission errors, and an increment of 18% of the
commission areas when compared with the 570 meters buffer radius (figure 3.2).

By using remotely sensed data and a textural algorithm, a total area of 243.26
hectares of log landings (patios and roads) were detected. The initial buffer radius of 150
meters around those detected log landings intersected with 45.8% of the total logged trees
(stumps) and 43.3% of the total logged forest, omitting 56.7% of the total logged forest

(appendix C5).

55

 

100.00

A as oo
.0 9 .°
0 o o
o o o

% Logged forest detected
N
9
o
o

 

.°
0
o

CO
m”
——

O O O O O O O O O O O O O
ass§asgswssra§es§s
Buffer radius (m)

! Detected logged forest I Omission E1 Commissiorfl

 

 

 

 

 

 

Figure 3.2. Different buffer radiuses around storage patios (from the field studies)

By increasing the buffer radius from 180 to 450 meters (appendix C.3), the
detection of selectively logged forest significantly increased from 49.3% to 88.1%,
respectively, while the omission errors decreased from 50.7% to 11.9% of the total
logged area. Commission errors were not present in these cases. Increasing buffer
radiuses fiom 480 to 720 meters, also increased the detection of selective logging from
90.1% to 98.3% of the total logged forest, respectively, and slightly decreased the
omission area from 9.8% to 1.7%, respectively. However, it created more than 29% of

commission (overestimation) errors, increasing from 1.1% to 30.7% (figure 3.3).

56

 

80.00
60.00 ”"-

40.00

% Logged forest detected

20.00

 

0.00

 

COCO
NWOO
mvvv

Buffer radius (m)
I Detected bgged fbrest I Om'ssion E1 Commission I

OOOOOOgO
“VFOMS N
WWWWVO \ON

360
9

OCOOOSO
WOO—1V5 M
v-‘v-‘NNNMM

 

 

 

 

 

Figure 3.3. Different buffer radiuses around log landings detected using textural

algorithm, Landsat ETM+, band 5.

3.5. Discussion and conclusion

This research developed an alternative approach to estimate selectively logged
forest using remote sensing techniques, based on the field study conducted in the Amazon
State of Rondonia.

There is a growing demand for research to better understand selective logging in
the Brazilian Amazon due to its potential effects on the environment and its interaction
with the process of land use and land cover change (Stone and Lefebre 1998, Nepstad et

al. 1999, Sousa and Barreto 2000, Alvarado and Sandberg 2001).

57

Based on the ground truthing data, I observed that the increasing size of the buffer
around storage patios would increase the detection of logged forests and decrease the
missing areas. Therefore, at certain size of buffer radius, a process of overestimating
would begin. Thus, the optimum buffer size for estimating selectively logged forest
depends on the desirable accuracy of the estimation. The buffer radius of 540 meters
showed a good estimation (almost 90%) of selective logging areas, small amount of
omission errors (less than 11%), and no commission errors.

Based on remote sensing techniques, the buffer radius of 480 meters showed to be
more precise for logging detection, overlapping more than 90% of the logged forest,
creating less than 10% of omission errors, and approximately 1% of commission errors. I
observed an exaggeration of the log landings detected using the automated method
because of effect of the textural algorithm, moving window 5 x 5 pixels, applied on
Landsat ETM+ band 5. Hence, the smaller buffer radius produced a similar buffer zone
compared to larger buffer radiuses based on the ground truthing data.

It is noted that the detection and estimation of low intensity selective logging
using Landsat imagery depends on the contrast between forest and soil exposure,
observed mostly in patios and road areas. In this study area, soil reflectance in tree-fall
gaps and skidder trails were not noticeable enough to be visually or automatically
detected on Landsat ETM+ imagery because of the low intensity (11.3 m3 ha") of
logging. Consequently, the total area impacted by selective logging had to be estimated
by using automatic analysis only (texture algorithm plus buffer radius around log

landings).

58

Finally, I observed that loggers often use the road network built to serve the
colonization settlement in the Cujubim municipality as forest access road. In this case, no
infrastructure is installed prior to logging operations. Logs are dragged hundreds of
meters from the forest interior to the roads using small trucks, winch tractors, or skidders.
I also observed that the margins of the roads are often used as temporary log storage area,
serving as an intermediate stage between forest and the sawmills, thus limiting the
number of patios. In this case, evidence of selective logging and forest degradation are

not easily detectable using satellite imagery because of lacking of soil exposure by l

 

logging features on the ground. Therefore, other experimental efforts to develop

methodological approaches to estimate these types of selective logging are needed.

3.6. References

ALVARADO, E. and SANDBERG, D V. 2001. Logging in Tropical Forest: Literature
review on Ecological Impacts. University of Washington. USDA Forest Service.
Internet site www.fs.fed.us/pnw/fera.

 

ASNER, G. P., KELLER, M., PEREIRA, R. and ZWEEDE, J .C. 2002. Remote sensing
of selective logging in Amazonia - Assessing limitations based on detailed field
observations, Landsat ETM+, and textural analysis. Remote Sensing of
Environment. 80, 483-496.

ERDAS. 1997. Erdas Field Guide. 4th edition. Atlanta, Georgia.

IBGE — Instituto Brasileiro de Geografia e Estatistica [Internet]. 2003 - Sistema IBGE de
Recuperacao Automatica. Extracao Vegetal. Rio de Janeiro, RJ. Available from
http://www.ibge.gov.br/

INPE- BRAZILIAN SPACE AGENCY. 2003. Monitoring of the Brazilian Amazon by
Satellite. S50 Jose dos Campos, SP, Brazil. Available from http://www.inpe.br/

JANECZEK, DJ. 1999. Detection and Measurement of Amazon Tropical Forest Logging

using Remote Sensing Data. A Master’s Thesis presented to Michigan State
University, Department of Geography.

59

NEPSTAD D.C., VERISSIMO A., ALENCAR A., NOBRE C., LIMA E., LEFEBVRE
P., SCHLESINGER P., POTTER C., MOUTINHO P., MENDOZA E., COCHRANE
M.A., BROOKS V. 1999. Large Scale impoverishment of Amazonian forests by
logging and fire. Nature. 398, 505-508.

RONDONIA, State Government. State Environmental Agency. 2002. Atlas
Geoambiental de Rondénia. Porto Velho, RO, Brazil. 140p.

SOUZA, C. Jr. and BARRETO A., 2000. An alternative approach for detecting and
monitoring selectively logged forests in the Amazon. International Journal of
Remote Sensing, 21. 173-179.

STONE, T. A. and LEFEBVRE, P. 1998. Using multi-temporal satellite data to evaluate
selective logging in Para, Brazil. International Journal of Remote Sensing, 19.
2517-2526.

 

UHL, C. and BUSCHBACHER, R. 1985. A Disturbing Synergism Between Cattle Ranch
Burning Practices and Selective Tree Harvesting in the Eastern Amazon. Biotropica.
17, 265-268.

WATRIN, 0.8. and ROCHA, A.M.A. 1992. Levantamento da vegetacao natural e do uso

da terra no municipio de Paragominas, Para, utilizando images Landsat (TM).
Boletim de Pesquisa, 124, EMBRAPA/ CPATU, Belem, PA, 40p.

60

CHAPTER IV
Inter and Intro-annual Analysis
4.1. Abstract
This chapter addresses the annual rates and the fates of forests impacted by
selective logging through inter and intra-annual analyzes of remotely sensed data and
field observations conducted in two study sites in the Brazilian Amazon. The study areas
are covered by two Landsat scenes, path 232 row 067 and path 226 row 068, located in

the Amazon States of Ronddnia and Mato Grosso, respectively. Visual interpretation and L.

 

an automated method (texture analysis) were applied. The fieldwork examined the
consistency of the remote sensing techniques by visiting areas identified as logged forests
on Landsat imagery. Based on this research results, I found that selective logging
increased 1,028 sz per year while deforestation increased 371 sz per year during
1993 to 2002 in the State of Mato Grosso study site. However, compared with the total
area detected, the increment of new selective logging areas in the Mato Grosso study site
decreased 22% from 1992 to 1999, respectively. More than 18% of the total logged forest
detected from 1992 to 2001 was deforested by 2002 in that study site. Deforestation
immediately following logging was observed in 2.9% of the total logging increment from
May 2001 to April 2002, which represents 13.22% of the deforestation increment in this
given period. In the Rondonia study site, new areas of selective logging increased an
average of 91 .42% per year compared with the total area of logging detected from 1989
to 2001, which corresponds to an increase of 40 sz per year of new selective logging
areas. From the total of 479 sz of selectively logged forests detected on satellite

imagery in the given period, more than 15% was deforested by 2001.

61

4.2. Introduction

As presented in the previous Chapters, although selective logging only harvests a
portion of trees, it can damage a large portion of forest and many others trees during
logging operations (Pinard and Putz 1996, Nepstad et al. 1999, Huth and Ditzer 2001).

The damages caused by selective logging vary according to the extraction
intensity and and affect the ability of the forest to recover if it is abandoned after the
harvest. If logged forests are revisited several times to harvest additional tree species (Uhl
et al. 1997, Verissimo et a1 1995), these forests become highly degraded and may have 40
— 50% of the canopy cover destroyed during the logging operations (Uhl and Vieira 1989,
Verissimo et a1. 1992).

Stone and Lefebvre (1998) observed different intensities of selective logging in
the case study in the State of Para, Brazil. The study showed that the intensity of logging
directly affected the extent of forest damages. The impacts of heavy machinery were
observed within the forest even several years after the harvest. The authors affirmed that
although selective logging impacts were evident in the field observations, it was very
difficult to distinguish logged and undisturbed forest four years following harvest using
satellite imagery.

Sousa and Barreto (2000) reported that log landings (patios and access roads)
from 1987 could not be detected using remote sensing techniques in 1991. Even areas of
logging in 1993 could not be detected in 1996. The authors stated that forest regeneration
covered the areas of bare soils, which became difficult to detect using remotely sensed
data. Hence, they suggested that a multi-temporal analysis of selective logging must use

images acquired no more than two years apart.

62

 

In this study, multi-annual and intra-annual analysis were performed using
remotely sensing data, acquired annually from 1989 to 2002, and complementarily field
verification, in two case studies located in the States of Mato Grosso and Rondénia.
These analyzes are intended to assess annual rates of selective logging increment and the

fate of logged forests such as subsequent logging or deforestation following harvest.

4.3. Regional settings: Southern and Southwestern Brazilian Amazonia

This study was conducted at the Sinop municipality in the state of Mato Grosso, l

 

Southern Brazilian Amazon, and at the Ariquemes municipality in the state of Rondénia,
Southwestern Brazilian Amazon. Two Landsat scenes defined the subsequent areas of
study used in this analysis path/row 226/068 for the state of Mato Grosso and path/row
232/067 for the state of Rondénia. Each scene encompasses approximately 187 x 186 km

(appendix D.1).

4.3.1. Description of the study area in the State of Rond6nia

The climate at the study area in the state of Rondénia is humid tropical with
annual precipitation of 2400 mm. A dry season extends from June through September.
The mean annual temperature is 26° C. The predominant vegetation type is open canopy
tropical forest, though savanna and dense forest occur sparsely (RADAMBRASIL,
1978). Changes in land use reflecting an increase in selective logging and pasture have
affected the previous closed canopy forest cover.

According to Pedlowski et a1. (1997), the state of Rondonia was part of the new
frontier occupation program created and enforced by the Brazilian federal government.

This program supported a road network improvement and construction in order to

63

promote the occupation of a new frontier with landless people from other Brazilian
regions and, at the same time, to promote economic growth (Dale et al. 1993, Pedlowski
et al. 1997).

As a consequence of this occupation process, the State of Rondonia has
experienced increasingly rapid landscape change. The area of deforestation increased
from 4.2 thousand sz to 58.1 thousand sz from 1978 to 2000, an increase of
approximately 1383%, which corresponds to 9.9% of the total Brazilian Amazdnia
deforestation (INPE, 2003). Additionally, more than 15 million in3 of round wood had

been exploited in the State of Rondonia from 1990 to 2001 (IBGE, 2003).

4.3.2. Description of the study site in the State of Mato Grosso

The climate in the study area (path/row 226/068) in the state of Mato Grosso is
humid tropical with annual precipitation of 2000 mm. The mean annual temperature is
26° C. A dry season extends from June through September. This area was originally
completely covered by semi-deciduous forest with emergent canopy. The predominant
soil type is dystrophic red-yellow Latossols (RADAMBRASIL, 1978).

In the state of Mato Grosso land use has been significantly changed from natural
vegetation to pasture and agricultural lands. Deforestation increased from 20 sz in 19??
to 143.9 thousand sz by 2000, an increase of 719%, corresponding to 24.49% of the
total Brazilian Amazonia deforestation (INPE, 2003). The round wood production from
1990 to 2001 reached more than 36 million In:3 exploited from natural forest (IBGE,

2003)

64

_— ‘3‘ FWLICE‘

4.4. Methods

4.4.1. Satellite imagery and geometric rectification

Digital Landsat ETM+ images for the state of Mato Grosso (path 226 row 069)
and for the state of Rondénia (path 232 row 067) from the Center for Global Change and
Earth Observation (CGCEO) at Michigan State University were used. These images were
acquired annually from 1989 to 2001 (appendix D.2) and from 1992 to 2002 (appendix
D.3) for the States of Rondc‘lnia and Mato Grosso, respectively. Three additional images
(path 226 row 068) were acquired in 2001 for an intra-annual analysis. The images fi'om
different years were rectified using ERDAS nearest-neighbor resampling method and
ground control points acquired using a GPS Trimble Pro-XR and Geo-Explorer 11 within

each Landsat scene.

4.4.2. Detection of selectively logged forest

Automatic analysis and visual interpretation techniques for selective logging
detection were applied. Inter-annual analysis of selective logging was performed on the
Landsat images acquired annually. Additional intra-annual analysis of selective logging
in the state of the Mato Grosso was performed using different acquisition dates within
2001.

The automated method, discussed in the chapter 2 and 3, created textural images
from Landsat TM and ETM+, band 5. Sequentially, these images were re-classified to
separate log landings from the undisturbed forest canopy and converted to Arc/Info
coverage in order to remove any remaining extraneous features that were not associated

with logging. A buffer radius of 180 meters around log landings was applied basin-wide,

65

except for the State of Rondénia and Acre (as discussed in the chapter 3), where a
regional buffer radius of 480 meters was applied. These buffer radiuses were applied in
order to estimate the amount of forest areas actually affected by logging and not
necessarily visible on satellite imagery. J aneczek (1999) described those areas as “cryptic
logging”.

A8 presented in the Chapter 2, visual interpretation to detect selectively logged
forests was done using RBG 5/3/2 color composite of the Landsat images, displayed at

full resolution on a computer screen. In this research, the areas of visible logging were ;

 

described as obvious logging. Logged forests were manually digitized on each image at
varied scales from 1:100,000 using Arc/Info. The minimum scale for manually digitizing
on Landsat imagery was 1:25,000. In this research, visible logged forest includes
spectrally bright patios, roads, and obvious canopy disturbance as well as logged areas
that exhibit faded log landings. The areas around the log landings together with areas of
obvious canopy degradation were digitized as polygons into vector GIS layer. They were
then classified as logged forest. The digitizing was done on the very edge of canopy
disturbance. Logged forests that did not have visible canopy disturbance on satellite

images were not digitized.

4.5. Field studies
4.5.]. State of Mato Grosso

The field study was conducted during July 2002 in the at Sinop municipality. A
Landsat scene (path 226 row 068) encompassed the whole study area. In this part of the

research, two senior students fi'om Forest Engineering (Fernando Raiter and Jansen Luiz

66

Trienweiler), Federal University of Mato Grosso, and two researchers from the Michigan
State University (Walter Chomentowski and Stephen Cameron) contributed with this
field verification. We visited abandoned and current selective logging sites. We visited
and verified areas of selective logging previously detected on Landsat imagery, focusing
on the image acquired in 2001 (the year prior our visit).

Selective logging activities in the State of Mato Grosso were shown to be
intensive and mostly located on ‘latifundios’ (extensive areas of private land). Therefore,
unequivocally we observed evidence of logging such as forest canopy damages, tree fall
gaps, and logging infrastructure such as access roads and patios.

As already presented in the Chapter 2, selective logging damage such as tree fall
gaps, stumps, logging access roads, skidder trails, and patios are quite evident in the field
two years after selective logging activities. These forest disturbances were evident in the
field from three to five years after logging and most of the forest damages, caused by
logging activities were covered by a strong secondary regrowth. In some cases of severe
impacts by logging activities, evidence of soil exposures (roads and patios) were still
observed even after 5 years after logging.

According to Annor Zanchette (the owner 'of Continental ranch), personal
communication 7/2002), the regional average of logging harvest is around of 25 m3 of
round wood per hectare. This harvesting intensity is variable depending on the market
demand for timber species and forest characteristics in terms of volume availability.

Wild fire does seem to affect logged forest in this study area. Indeed, Cochrane
(2001) affirmed that selective logging increases forest fragmentation and, consequently,

forest fire susceptibility. Logged forest and deforested areas converted to pasture or

67

agriculture are sharing the same space in that region. According to Uhl and Buschbacher
(1985), Uhl and Kauffinan (1990), and Cochrane (2001) agricultural plots in the Brazilian
Amazon are the main widespread ignition source of fire. Some of the areas visited
showed intensive forest degradation. Evidence of fire was observed in all of them, which
can be classified as a new land use type. Those areas form a blend of logged forest and
secondary regrowth. When using unsupervised classification of landsat images, most of
these areas were classified as secondary regrth or deforestation. Gerwing (2000),
observed that a combination of selective logging and fire degrades forest structure and
creates different land cover types characterized between natural forest and deforested

areas.

4.5.2. State of Rondi‘rnia

The field study was conducted during June 2002 at Ariquemes, Campo Novo,
Buritis, and Monte Negro municipalities in the State of Rondenia. The study area was all
included in a Landsat ETM+ scene (path 232 row 067). Two local forest engineers
(Wilson Soares Abdala and Vilmar Ferreira) contributed with this field study.

We visited selective logging areas previously detected on satellite imagery.
Twenty sites out of one hundred and six digitized polygons on Landsat imagery acquired
in 2001, were visited. We observed that all of them were logged in the previous year, still
showing clear evidences of selective logging such as visible canopy disturbances,
temporary logging roads, and patios. Although the patios and roads were still visible in
the field, we observed a strong regeneration and most of the tree fall gaps and skidder

trails were already covered by secondary regrowth (vines, fast-growth tree-species, etc).

68

In general, we observed that loggers distribute the patios and roads randomly
within the forest rather than following previous planning. The logging roads seemed to be
plotted according to the source of the most market valuable trees (raw material for timber
industries), therefore showing irregularly curves. Nearby logging areas at the margins of
the road network are often used as temporary log storage area.

According to the forest engineers Abdala and F erreira (2003), an average of
around 15 in3 per hectare had been harvested in the visited areas. Additionally, they
demonstrated that selective logging infrastructures (patios and roads) are very expensive.
Loggers prefer to drag logs farther away (as far as a thousand meters) rather than building
more access roads and patios. In this case, rather than trucks, loggers are likely to use
more skidders or tractors for the inside forest logging operations, therefore, there are less
visible features on satellite imagery such as access roads and patios.

Although the purpose of this research was the ground verification of selective
logging sites detected on satellite imagery, many other sites showed evidence of selective
logging , at certain stages of recovery. Those sites are located at the Burareiro and
Marechal settlement projects.

In these settlement projects, I observed that most of the remaining forests are part
of legal reserves required by the national forestry code (federal law) for individual farms.
Although deforestation is not allowed in those legal reserves, logging is permitted under
Special requests to the local Federal Environmental Agency (Instituto Brasileiro do Meio
Arnbiente e Recursos Naturals Renovaveis — IBAMA). Consequently, the logging harvest
cycle depends on the economic or timber demand of the landowner. F urthermorc,

selective logging requires low investment in infrastructure in those areas since a road

69

network is already present to serve each individual farm. We observed that the road
network built to serve the settlers for agro-products transportation and access was often
used as a log storage and loading areas at its margins.

We also observed a low to very low intensity of logging in these settlement
projects, varying from 3 to 10 m3 ha". Logs are dragged hundreds meters from the
interior forest to the roads using winch tractors, small trucks, or skidders.

In spite of the low intensity of logging, from the ground, logged forests were

easily identified because of the visible forest damages (logging trails, tree-fall gaps, and

 

stumps) caused during logging activities. However, we observed that forests disturbed by i
logging were rapidly recovering from the direct logging impact. Those areas showed very
strong vine and fast-growth species regeneration. These might be the reasons why

detection of selective logging using Landsat imagery is very difficult in that study area.

4.6. Results

4.6.1. Case study: the State of Mato Grosso

The results of this research show that deforestation area derived from the Landsat
scene, path /row 226/068, increased from 12% to 24% of the total study area between
1992 and 2002. Cumulative selective logging areas in the study site increased from 4%
to 36% of the total study site area by 1992 and 2002, respectively. By 2002, only 40% of

the study area remained as undisturbed forest (figure 4.1).

70

 

 

I Logged forest
I Deforestation
1:] Forest

%

 

 

 

 

1992 2002

Year

 

 

 

Figure 4.1. Land use and land cover change in the study site (path/ row 226/068)
An average of 371 .4 sz yr" were deforested from 1992 to 2002. The total areas

of selective logging detectable from satellite imagery increased from 1.2 thousand sz in

1992 to 4.4 thousand sz in 2002, accounting for an increase of 366% (table 4.1).

71

Table 4.1. General results for the path/row 226/O68

 

 

 

 

Land Use (sz)
Year Deforestation
Increase
Forest Deforestation Logged Forest Water Body (sz)
1992 24,527.1 3,403.6 1,166.0 36.1 -
1993 23,3355 3,900.3 1,860.9 36.1 496.7
1994 22,961.0 4,246.4 1,889.6 35.9 346.0
1995 22,9560 4,663.3 1,477.2 36.4 416.9
1996 22,4575 5,036.8 1,602.1 36.5 373.5
1997 21 ,584. 1 5,420.8 2,091.4 36.5 384.0
1998 20,3702 5,733.4 2,992.8 36.3 312.6
1999 18,6462 6,221.4 4,228.8 36.4 488.0
2000 18,425.? 6,357.7 4,313.0 36.4 136.3
2001 17,914.4 6,844.0 4,338.0 36.4 486.3
2002 17,581.7 7,117.5 4,397.2 36.4 273.5
371.4

Annual rate of deforestation (sz)
Note: A mask of 972 Km2 of the overall clouds, shadows, and smoke was applied.

 

The increase of new areas of selective logging observed from 1993 to 2001 was
about 1,080 Km2 yr". A small reduction was verified during the 1994 through 1996
period, an average of 730 sz yr". A significant increase was verified in 1999, an
average of 1,800 sz yr". From 2001 to April 2002, an increase of approximately 600
sz was observed, causing a partial increment for that given year. Although overall areas
of selective logging have been increasing from 1993 to 2002, the relative annual increase
(total increase compared with total logging detected) has been decreasing from 62.99%,

49.44%, to 27.22% in 1993, 1997, and 2001, respectively (table 4.2).

72

Table 4.2. Selective logging and deforestation increase for the path/row 226/068

 

 

 

 

 

Logging Deforestation
Year
Area (sz) Area (sz)
Total Increment % Total Increment %

1992 1,165.98 - - 3,403.65 - -
1993 1,860.92 1,172.13 62.99 3,900.32 496.67 12.73
1994 1,889.56 719.99 38.10 4,246.36 346.04 8.15
1995 1,477.16 676.18 45.78 4,663.27 416.91 8.94
1996 1,602.11 802.44 50.09 5,036.81 373.54 7.42
1997 2,091.44 1,033.94 49.44 5,420.82 384.01 7.08
1998 2,992.83 1,113.02 37.19 5,733.44 312.62 5.45
1999 4,228.75 1,838.86 43.48 6,221.43 487.99 7.84
2000 4,313.00 1,169.78 27.12 6,357.70 136.27 2.14
2001 4,337.99 1,180.95 27.22 6,843.99 486.29 7.1 1
2002 4,397.24 575.38 13.09 7,117.50 273.51 3.84

 

Deforestation of logged forests is more likely to occur in the oldest logging areas.
For example, deforestation of previously logged forests by 2002 decreased from 23%,
14%, 6%, to 0.8% for areas detected as logging in 1992, 1995, 1998 and 2001,

respectively (table 4.3).

73

Table 4.3. Deforestation and selective logging for the path/row 226/068

 

Logging areas deforested

 

 

Year of logging Total logged forest by 2002
Area (sz) Area (sz) %
1992 1165.98 265.30 22.75
1993 1860.92 338.60 18.20
1994 1889.56 307.97 16.30
1995 1477.16 206.76 14.00
1996 1602.11 158.06 9.87
1997 2091.44 138.89 6.64
1998 2992.83 185.59 6.20
1999 4228.75 145.60 3.44
2000 4313.00 122.74 2.85
2001 4337.99 36.15 0.83

 

Remote sensing analysis showed that selective logging could be detected on
satellite imagery for years. However, according to the field observations logging
generally stays active at a specific place no more than a year. Based on that, I estimated
that, fiom 1993 to 2000, an average of 7% of the active or ongoing logging was, in fact, a
‘revisit’ of previously logged forest. The areas of revisit logging had increased from 32,

319, to 455 sz in 1994, 1998, and 2002, respectively (table 4.4).

74

Table 4.4. Selectively logged forest revisited (path/row 226/068)

 

 

 

Total logging Net logging increment Revisited logging

Year Area (sz) Area (sz) Area (sz) (%)

1992 1,165.98 -
1993 1,860.93 1,172.13 - -
1994 1,889.56 719.99 31.99 1.69
1995 1,477.16 676.18 51.43 3.48
1996 1,602.11 802.44 101.55 6.34
1997 2,091.44 1,033.94 148.91 7.12
1998 2,992.84 1,1 13.02 319.28 10.67
1999 4,228.75 1,838.86 431.70 10.21
2000 4,313.00 1,169.78 491.61 11.40
2001 4,337.99 1,180.95 394.53 9.09
2002 4,397.24 575.38 455.26 10.35

 

Although revisited logging areas are relatively small compared with the total
selective logging, they account more than 23% of the total increase of new logging areas

from 1992 to 2001.

4.6.1.1. Intra-annual analysis

The intra-annual analysis was conducted using four Landsat images (path/row
226/068), three acquired in May, August, October 2001, and one acquired in April 2002
(appendix D3). The average of the total logging area detected was 4,134 sz for the
four scenes. The increase of new logging areas was 1,248.84 sz from May 2001 to

April 2002. The total deforestation estimated for the given period was 273.51 sz. Part

75

of this total deforestation (36.15 sz) was from land previously affected by selective
logging (appendix D.4). Those areas were deforested immediately after logging and
represent 2.9% of the total net logging increase detected on satellite imagery in that

period of time (table 4.5).

Table 4.5. Intra-annual increment of selective logging (path/row 226/068)

 

 

 

 

Logging net
Period of time increment

Area (sz)
May 2001 to August 2001 483.42
August 2001 to October 2001 273.53
October 2001 to April 2002 491.45
Total net logging increment (2001 to 2002) 1,248.4
Total of logged forest deforested by 2002 36.15

 

Although deforestation immediately following selective logging is only 2.9% of
the logging increment, it represents 13.22% of the deforestation increment in the period

of 2001 to 2002.

4.6.2. Case study in the State of Rondenia

The results of this research show that it is not possible to detect selectively logged
forest from Landsat imagery in this study area as much as in the case study in the State of
Mato Grosso.

Based on the time-series imagery 1989 to 2001 of the path/row 232/067, the total

selective logging detected averaged only 43.4 sz yr". Approximately 8.3% of detected

76

logging on satellite imagery was derived from the previous year logging. An additional
0.3% was detected from two years previous logging. Consequently, most of logging areas
detected using Landsat imagery (91.4%) were considered new logging areas averaging

39.9 sz yr" of net increase (table 4.6).

Table 4.6. Selectively logged for the path/row 232/067

 

Selective logging

 

 

 

 

Year Total Net increment
Area (sz) Area (sz) %

1989 40.29

1990 33.84 22.08 65.24
1991 24.48 23.26 95.02
1992 10.32 10.32 100.00
1993 66.38 63.78 96.08
1994 15.48 9.28 59.96
1995 18.47 14.33 77.58
1996 28.53 27.69 97.06
1997 49.95 42.92 85.92
1998 31.67 29.52 93.21
1999 124.27 121.41 97.70
2000 37.67 35.70 94.77
2001 82.93 78.74 94.95

Total 523.98 479.02 91.42

 

4.7. Discussion and conclusion
According to the results from the case study in the State of Mato Grosso, the total

area of selective logging significantly increased from 1992 to 2002. The increase of new

77

areas of selective logging has been slightly changed the last ten years. The average
increase of new areas of selective logging was 39.15% of the total detected from 1992 to
2001. Nevertheless, the relative increase of new areas compared with the total logging
area detected using Landsat imagery has decreased from 60.2% in 1992 to 47.52% in
1996, and to 37.97% in 1999, represented by the following linear equation with R2=

0.5757.

Y = -3.1794x + 6393.6

where:

Y= % of increment of selective logging areas compared with total detected or
mapped on Landsat imagery

x = year of logging detection

Based on these research results, deforestation is not significantly following
selective logging activities in the Mato Grosso case study, even though the total of
around 1.9 thousand Krn2 or 18.11% of logged forest were deforested from 1992 to 2001.
Moreover, loggers are revisiting former logging areas more often and, consequently,
increasing forest degradation in the State of Mato Grosso study case. Hence, based on the
studies by Uhl and Kauffrnan (1990), Holdsworth and Uhl (1997), and Cochrane et al.
(1999), it is more likely that forest fire will be a significant risk for selectively logged
forest in that region.

Results from the intra-annual analysis in the Mato Grosso study area show that

deforestation immediately following logging activities may be reduced by 2.9% of the

78

total logging areas detected on satellite imagery. These areas are classified as
deforestation instead.

In the State of Rondénia study area, the total of 479 sz of logged forest was
detected from 1990 to 2001. The average increase of new selective logging areas
compared with the total logging detected in this study site was 91.42% or 40 sz yr" in
the given period. The same increment showed to be slightly increasing fi'om 82.5%,
87.87%, to 91.94% in 1992, 1996, and 1999, respectively, when applying the following

linear equation with R2: 0.132.

Y = 1.3565x - 2619.7

where:

Y= % of increment of selective logging areas compared with total detected or
mapped on Landsat imagery

x= year of logging detection

The total of 72.6 sz (15.2%) of logged forest were converted to agricultural
land use by 2001. Although the field verification targeted logged areas previously
detected on the satellite imagery, I observed several logged forest areas that were not

detected on satellite imagery.

4.8. References

ABDALA, W. S. and FERREIRA, V. F. 2003. Personal communication during the
fieldwork, in June 2002. Ariquemes, State of Rondenia, Brazil.

79

COCHRANE, M.A. 2001. Synergistic Interactions between Habitat Fragmentation and
Fire in Evergreen Tropical Forests. Conservation Biology. 15, 1515-1521.

DALE, V.H., ONEILL, R.V., PEDLOWSKI, M. 1993. Causes and effects of land-use
change in central Rondénia, Brazil. Photogrammetric Engineering & Remote
Sensing, 59, 997-1005.

GERWING, J. J. 2000. Degradation of forests through logging and fire in the eastern
Brazilian Amazon. Forest Ecology and Management. 157, 131-141.

HUTH, A. and DITZER, T. 2001. Long-terrn impacts of logging in a tropical rain forest —
a simulation study. Forest Ecology and management. 142, 33-51.

IBGE - Instituto Brasileiro de Geografia e Estatistica [Internet]. 2003 - Sistema IBGE de
Recuperacao Automatica. Extracao Vegetal. Rio de Janeiro, RJ. Available from
http://www.ibge.gov.br/

INPE. BRAZILIAN SPACE AGENCY. 2003 — Monitoring of the Brazilian Amazonian
by Satellite. 2000-2001. Available on Internet site www.inpe.br

JANECZEK, DJ. 1999. Detection and Measurement of Amazon Tropical Forest
Logging using Remote Sensing Data. A Master’s Thesis presented to Michigan
State University, Department of Geography.

PINARD, M. A. and PUTZ, F .E. 1996. Retaining Forest Biomass by Reducing Logging
Damage. Biotropica. 28. 278-295.

NEPSTAD D.C., VERISSIMO A., ALENCAR A., NOBRE C., LIMA E., LEFEBVRE
P., SCHLESINGER P., POTTER C., MOUTINHO P., MENDOZA E.,
COCHRANE M.A., BROOKS V. 1999. Large Scale impoverishment of
Amazonian forests by logging and fire. Nature. 398. 505-508.

PEDLOWSKI, M.A., DALE, V.H., MATRICARDI, E.A.T., SILVA FILHO, ER. 1997.
Patterns and impacts of deforestation in Rondénia, Brazil. Landscape and Urban
Planning. 38, 149-157.

RADAM BRASIL. 1978. Projeto Rac_l_ambra_sii Programa de Integragi’to Nacional, 1978.
Rio de Janeiro, Ministerio das Minas e Energia, Departamento Nacional de
Produccao Mineral.

SOUZA, C. Jr. and BARRETO., 2000. An alternative approach for detecting and
monitoring selectively logged forests in the Amazon. International Journal of
Remote Sensing, 21, 173-179.

80

STONE, T. A. and LEFEBVRE, P. 1998. Using multi-temporal satellite data to evaluate
selective logging in Para, Brazil. International Journal of Remote Sensing, 19,
2517-2526.

UHL, C. and BUSCHBACHER, R. 1985. A Disturbing Synergism Between Cattle Ranch
Burning Practices and Selective Tree Harvesting in the Eastern Amazon. Biotropica.
17, 265-268.

UHL, C. and VIEIRA, I.C.G. 1989. Ecological impacts of selective logging in the
Brazilian Amazon: A case study from the Paragominas region of the state of Para.
Biotropica. 21, 98-106.

UHL, C. and KAUFFMAN, J .B. 1990. Deforestation, Fire Susceptibility, and Potential
Tree Responses to Fire in the Eastern Amazon. Ecology. 71, 437-449.

UHL, C., BARRETO, P., VERISSIMO, A., VIDAL, E., AMARAL, P., BARROS, A.C.,
SOUZA, C. JOHNS, J ., GERWING, J. 1997. Natural Resources Management in the
Brazilian Amazon: An integrated research approach. BioScience. 47, 160-168.

VERISSIMO, A., P. BARRETO, M. MATTOS, R. TARIFA, and C. UHL. 1992.
Logging impacts and prospects for sustainable forest management in an old
Amazonian frontier: the case of Paragominas. Forest Ecology, and Management.
55, 169-199.

VERISSIMO, A., BARRETO, P., TARIFA, R., and UHL, C. 1995. Extraction of a high

value resource from Amazé‘lnia: The case of mahogany. Forest Ecology and
Management. 72, 39-60.

81

CHAPTER V

Amazon-Wide Selective Logging Detection and Measurement

5.1. Abstract

Amazonian deforestation is a significant perturbation of the global terrestrial
carbon cycle. In the Amazon, forest clearing is quite apparent using Landsat satellite
imagery. Most available land cover data maps show only sites where trees have been
completely removed, not partially removed or degraded as in the case of selective
logging. Large-scale selective logging is a relatively new activity in the Amazon, the full
consequences of which have yet to be evaluated. Forest degradation caused by logging
and accidental fire is visible in Landsat ETM+ images, however, rapid regrowth (within
one to five years) can quickly obscure these areas. Forest impoverishment due to logging
in Amazonia has been estimated to create a net carbon flux equal to approximately 4-7 %
of the annual carbon release from deforestation. Therefore, current carbon flux estimates
for the region may be low because the growing effects of logging have not been
accounted for. In this research, visual interpretation and textural analysis remote sensing
techniques were applied to identify and map selective logging in tropical “terra-firrne”
(high land) forests together with the correlated multi-annual measurement results for
1992, 1996, and 1999, for the Brazilian Amazon. Logged areas detected using this
methodology indicate that selective logging is increasing rapidly in both intensity
(regional) and area (basin-wide). Specifically, by 1992, at least 5,980 sz of forest had
been logged. During the 1992-1996 and 1996-1999 intervals, the area impacted expanded
by an additional 10,064 sz, and 26,085 sz, respectively. Based on a multi-annual

analysis, it was estimated that part of this total logged forest, at least 3,689 sz had

82

being actively logged by 1992, an additional 5,107 sz, and 11,638 sz, had been
logged by 1996 and 1999. It was also estimated that at least 10% of logging areas
detected in 1999 were previously logged. This methodology is considered conservative
since it may miss very-low intensity logging and logging in riparian forests. The data
presented here represent the first multi-temporal assessment of the amount and locations

of selective logging in the Brazilian Amazon.

5.2. Introduction

Satellite-based remote sensing has been used for many years to monitor and
evaluate land cover change in the Amazon, concentrating on forest and deforested land
evaluation (F earnside et a1. 1990, Skole and Tucker 1993). Data from these studies, such
as the rate of deforestation in Brazil’s Legal Amazon, are often used to estimate human
effects on the global carbon cycle (F earnside 1997, Houghton 1997). Nevertheless,
according to Nepstad et al. (1999), “present estimates of annual deforestation for
Brazilian Amazonian capture less than half of the forest area that is impoverished each
year, and even less during years of severe drought.”

In addition to outright deforestation, Amazonian forests are increasingly being
exposed to logging activities. Logging in these forests is often selective in that only the
more economically valuable trees are removed. This type of logging can be based on as
few a one or two high value species, as is the case for mahogany (Verissimo et al. 1995),
or it can encompass 100 or more species in more developed logging regions (Uhl et a1.
1997). Using traditional felling techniques, the felling of a single tree can directly result
in the death of 6 other nearby trees (Verissimo et al. 1992). Furthermore, the sum total of

damages from logging operations, with the creation of logging roads and skidder trails,

83

often lead to as many as 40% of the remaining trees being killed or severely damaged
(Uhl et al. 1991). The impacts of selective logging vary with extraction intensity, but can
be substantial. Selectively logged forests would be expected to accumulate carbon over
time and recover to pro-harvest levels of biomass if left undisturbed. However, many
forests are revisited several times when loggers return to harvest additional tree species as
regional timber markets develop (Uhl et al. 1997; Verissimo et al. 1998). These forests
become very degraded and may have 40 - 50% of the canopy cover destroyed during
these logging operations (Uhl and Vieira 1989, Verissimo et a1. 1992). The effects of
selective logging include increased fire susceptibility (Holdsworth and Uhl, 1997),
damage to nearby trees and soils (Johns et a1 1996), increased risk of local species
extirpation (Martini et al 1994), and emissions of carbon (Houghton 1996). Furthermore,
uncontrolled exploration by loggers catalyzes deforestation by opening roads into
unoccupied government lands and protected areas that are subsequently colonized by
ranchers and farmers (Verissimo et a1 1995).

Selective logging is becoming an increasingly important activity in Amazonian
forests. Using field surveys of 1393 wood mills in the Brazilian Amazon, Nepstad et al.
(1999) estimated that 10,000 to 15,000 kmzyr" of forest are severely damaged by
logging. Moreover, IBGE (2003) estimates that, between 1990 and 1999, more than 413
million cubic meters of round wood was extracted from the North region of Brazil, an
average ofalmost 41.3 million m3 yr".

Selective logging has been occurring in Brazil’s tropical forests for several years,
but, although the damages are readily visible in some Landsat TM images, the activity

has not been detected or quantified by most Landsat TM classification techniques. For

84

example, previous estimates of deforestation for Brazil’s Legal Amazon, reported by
Michigan State University - MSU and the Brazilian Space Agency - INPE, did not
include most selectively logged areas (J aneczek, 1999).

Visual interpretation and a textural algorithm were applied to identify and map
selective logging areas throughout the Brazilian Amazon. These data extend current
Amazon-wide image classification techniques to include selectively logged forests in
addition to the current seven thematic classes used in basin-wide land-cover assessments:

forest, deforestation, regrowth, cerrado, cloud, cloud shadow, and water.

5.3. Methodology
5.3.1. Study Area and Data Set

The study area was Brazil’s Legal Amazon. The Legal Amazon is considered an
administrative area within the country of Brazil. It is 5x106 km2 and includes all of the
states Acre, Amapa, Amazonas, Para, Rondenia, Roraima, plus parts of Mato Grosso,
Maranhao, and Tocantins. This analysis was based on the more than 600 Landsat images,
covering the Legal Amazon, that are part of the ongoing deforestation monitoring at the
Center for Global Change and Earth Observations (CGCEO) at Michigan State
University. Signs of logging were searched for throughout the basin. Evidence of
selective logging was found in 30 Landsat TM images acquired in 1992 and 1996, and 38
Landsat ETM+ images in 1999 (figure 5.1). These images defined the subsequent areas

of study used in this analysis.

85

 

 

serene, mafia 2: a eon: swap: .2 2:5

 

 

an van .

8.6 3.53:3. 8:03 wan—Ea..—

c.1393 .
casing .

 

 

 

53% 2:. - mm?
203 am.— 9420

L

«he 3...:

 

 

 

 

    

:38! was a... 8.23 .35 I
83.5, .33 when has

 

42m

 

 

86

For dates prior to 1999, Landsat 5 TM imagery was used. For 1999, Landsat 7
ETM+ imagery was used for the analyses. Although I classify results as being from three
different years (1992, 1996, and 1999) for measurement and analysis purposes, some of
the imagery was acquired in other years depending on its availability. A firll listing of the
individual scenes used is provided in the appendix E1

The deforestation dataset produced by the CGGEO for the whole Brazilian
Amazon was also used. In the deforestation analysis, the imagery was classified, using an
unsupervised image classification model of bands 2, 3, 4, and 5, into seven thematic
classes: forest, deforestation, regeneration, cerrado, clouds, shadows, and water. This

dataset is available in Arc/Info and Grid format for 1992, 1996, and 1999.

5.3.2- Calibration and Geometric Rectification

Geometric rectification was done using nearest-neighbor resampling with the four
points derived from ephemeris data, supplied with the images at the time of ordering.
Additionally, the rectified images were validated and tested at BSRSI using collected

GPS ground points from several locations in the Amazon.

5.3.3. Logging Detection
As presented and discussed in Chapters 2, 3, and 4, visual interpretation and
automatic analysis (a textural algorithm) were applied to detect and quantify selective

logging Amazon-wide.

87

5.3.3.1. Automatic Analysis

The automatic analysis consisted in applying a textural algorithm (variance), 5 x 5
moving window operator, on band 5 of Landsat TM and ETM+. The resultant texture
images were used to detect log landings, which are small clearings cut into the forest for
the purpose of acting as temporary access and transportation, storage, and standing areas
by loggers (figure 5.2). Those images were re-classified using Erdas modeling tools,
selecting and capturing the pixels values (mostly fi'om 6 to 11) correlated to areas directly

affected by selective logging, such as patios and roads.

 

 

 

2 0 4 8 Kilometers

 

 

 

Figure 5.2. Log landings detected using textural algorithm, band 5, 226/068, 2002.

88

The classified texture images were converted to vector format and edited to
removing any remaining extraneous features that were not associated with logging. A
buffer radius of 180 meters, as suggested by Souza and Barreto (2000), was applied
around the detected log landings (patios) in order to estimate the amount of forest
actually affected by logging (Figure 5.3). A buffer radius of 450 meters, as presented in
the Chapter 3, was applied for the State of Rondénia and Acre, because of their specific
characteristics in terms of selective logging. Those areas of selective logging detected

using a buffer radius here are described as Cryptic Logging.

 

 

 

2 0 4 8 Kilometers

 

 

 

Figure 5.3. Buffer radius of 180 meters around of log landings detected using

textural algorithm on band 5, path/row 226/068, 2002.

89

Non-forest areas (deforestation, cloud, shadow, cerrado, bodies of water, and
regenerating vegetation) were subsequently masked out of band 5 of the Landsat images.

Subsequently, the non-forest mask was applied again to remove any regions

where the suspected-logging areas overlapped with non-forested land cover classes.

5.3.3.2. Visual Interpretation

Visual interpretation of logged forests was done using RBG 5/3/2 color composite
images, displayed at full resolution. The logged forests were digitized manually on each
image at varied scales from 1:30,000 to 1:100,000, in the three different study periods
(1992, 1996, and 1999).

The logged forests were identified by obvious canopy degradation, since logging
activities leave log landings, and tree-fall gaps along with obvious canopy disturbance.
This land use creates a characteristic pattern of white points on the Landsat images,
which are patios or roads, embedded in the red hues of the forest canopy. The areas
around the logging landings together with areas of obvious canopy degradation were
digitized as polygons.

After areas affected by logging were identified, they were digitized into vector
GIS layers. The digitizing was done on the very edge of canopy disturbance. The logging
areas were classified into two separate vector coverages; “obvious” and “subtle” logging.
“Obvious” logging includes spectrally bright patios, roads, and obvious canopy

disturbance (figure 5.4).

90

 

 

Scale

1 0 l 2 3 Kilometers

 

 

 

 

Figure 5.4. Obvious logging on Landsat image, 226/068, RGB 5/3/2, 2001.

Subtle logging refers to logged areas that exhibit visible canopy disturbance and

faded log landings, or no log landings (figure 5.5).

 

 

Scale

55::
l 0 l Kilometer

 

 

 

Figure 5.5 - Subtle logging on Landsat image, 226/068, RGB 5/3/2, 2001.

Logged forests that did not have visible canopy disturbance on satellite images

were not digitized.

5.4. Results

In 1992, the basin-wide estimate of logged forests was 5,980 kmz. The detected
logging areas were 10,064 km2 in 1996 and 26,085 km2 in 1999 (appendix E.2 for details
by Landsat path/row). These areas were not detected by the deforestation classification,
done by the Center for Global Change and Earth Observations, and represent an
increment of degraded forest areas over those estimates.

In the period of analysis, the results show that visual interpretation detected an
average of 83.4% (66.2% classified as “obvious logging” and 17.2% as “subtle logging”),

while texture analysis (“cryptic logging”) detected an average of 63.7% of the total

logged forests (table 5.3).

Table 5.3. Selective logging by visual interpretation and automatic analysis

 

 

 

 

Visual interpretation Automatic analysis
Year Obvrous Subtle Cryptic Total 0,)
Area (Km’) % Area (km’) % Area (Km’) %
1992 3,946.41 66.00 1,269.36 21.23 3,163.03 52.90 5,979.80
1996 6,553.37 65.12 1,359.23 13.51 7,451.55 74.04 10,064.10

1999 17,600.08 67.47 4,428.84 16.98 16,767.06 64.28 26,085.40
* The overlap between obvious-subtle and cryptic logging areas was not double counted.

92

Areas of selective logging detected in common between visual interpretation and
automatic analysis were 36%, 53%, and 49%, respectively, in 1992, 1996, and 1999.
Visual only contributed in detecting an additional of 47%, 26%, and 36%, respectively in

1992, 1996, and 1999 (table 5.4).

Table 5.4. Logging area by visual interpretation and automatic analysis

 

Year Visual only Automatic only Overlap (V/A) Total
Km: % Krn2 % Kin2 % (sz)

 

 

1992(') 2,817.17 47.11 1,009.01 16.87 2,153.63 36.02 5,979.80
1996(2) 2,615.61 25.99 2,155.10 21.41 5,293.39 52.60 10,064.10

1999(2) 9,323.76 35.74 4,060.90 15.57 12,700.74 48.69 26,085.40
‘ done by Janeczek (1999) and revised by Matricardi (2003); 2 done by Matricardi (2003)

A state by state analysis of the results show that the States of Mato Grosso and
Para contain the vast majority of the Legal Amazon’s detectable selective logging
(90.63% in 1992, 91.89% in 1996, and 89.92% in 1999). Mato Grosso and Para also have
the highest increment of selective logging areas in 1992, 1996, and 1999 (table 5.5).

Using this methodology, no logging was detected by in the states of Amazonas,
Amapa, and Roraima. Additionally, only a small amount of logged forests was detected
in states of Acre and Rondenia, even by applying a bigger regional buffer radius for these

states presented in the Chapter 3.

93

Table 5.5. Selective-logging areas detected by State in the Brazilian Amazon

 

 

 

 

State 1992 1996 1999

Area (Km2) % Area (Km2) % Area (Km2) %
Acre 7.84 0.13 29.52 0.29 23.94 0.09
Arnapa - - - - - -
Amazonas - - - - - -
Maranhao 506.12 8.46 571.87 5.68 2,276.91 8.73
Mato Grosso 3,013.58 50.40 4,196.50 41.70 12,662.47 48.54
Para 2,405.45 40.23 5,051.16 50.19 10,793.87 41.38
Rondenia 41.81 0.70 203.05 2.02 315.20 1.21
Roraima - - - - - -
Tocantins 5.00 0.08 12.00 0.12 13.00 0.05
Total 5,979.80 100.00 10,064.10 100.00 26,085.40 100.00

 

Moreover, multi-annual comparisons of detected logging in 1992-1996, 1992-

1999, and 1996-1999 indicate areas in common of 267, 253, and 4,062 km2 respectively

among the given periods of time. These results also show that repetitive (revisited)

logging in 1992, 1996, and 1999 is around of 148 kmz.

Multi-annual comparisons between logging and deforestation data indicate that

around 796 km2 (13%) of logged forests in 1992 were deforested by 1996 and an

additional 896 km2 (15%) by 1999. Around 1,074 km2 (11%) of the logged forests in

1996 were deforested by 1999. Around of 342, 587, 2,706 sz of overlap between

deforestation and visual interpretation of selective logging, respectively by 1992, 1996,

94

and 1999 was observed. Those areas were detected as logged forest and had been
classified as deforestation, mostly as secondary regrowth, by using unsupervised
classification in the deforestation analysis, because of their heavy canopy degradation. In
this logging analysis, after fieldwork observation, those areas were re-classified as
selective logging.

Approximately 123, 204, and 820 km2 of overlap between logged and deforested
areas, respectively, in 1992, 1996, and 1999. These overlap areas were due to the buffer
zones that expanded beyond the forest areas into deforested areas. These areas of overlap

were not included in the total area of logging.

5.5. Discussion and Conclusions
5.5.1. Accuracy assessments

Ecosystem damages from selective logging and subsequent impacts following the
operation such as desiccation, fire, and access-based colonization are severe (Verissimo
et al. 1992, Nepstad et al. 1999), causing canopy disturbance and soil exposure (Souza
and Barreto, 2000). These changes are detectable using remote sensing and allow for
estimation of the area of selectively logged forests in the Brazilian Amazon. Although
these techniques are not applicable to all forests (e. g. varzea forests), they are applicable
in areas of “terra firrne” (dry land), which make up 80% of all forests and account for the
vast majority of logging in the Amazon.

By using a “textural algorithm”, the buffer radius developed by Souza and Barreto
(2000), and a buffer radius developed in this research for States of Rondénia and Acre

(Chapter 3), automatic analysis incorporated roughly 64% of the total (automated plus

95

visual) estimated logging areas in 1992, 1996, and 1999. Visual interpretation
incorporated approximately 83% of the total in the given years. These areas were
classified as either ‘obvious’ or ‘subtle’ depending on their characteristics. Specifically,
areas of obvious logging had well defined logging roads and patios and extensive canopy
degradation. Areas of subtle logging had lesser degrees of canopy disruption or visible
infrastructure, either due to being early in the logging process or because of substantial
forest regrowth in previously logged forests. The division of the total detected logged
forests by class, obvious (visual), subtle (visual), and cryptic (automatic) is presented in

figure 5.6.

 

20,000.00

1 8,000.00

1 6,000.00
14,000.00

N 12,000.00
1 0,000.00
8,000.00
6,000.00
4,000.00
2,000.00

)

Area(Km

 

l 992 l 996 1999

Year

 

 

 

EJObvious ISubtle ICryptic

 

 

 

 

Figure 5.6. Logging areas classified according to the used detection methodology -

Cryptic (texture analysis) and Obvious and Subtle (visual interpretation)

96

Visual interpretation only provided roughly 47%, 26%, and 36% of the detected
logged forests, in 1992, 1996, and 1999 respectively, while texture analysis only added
17%, 21%, and 16% in the given years. The total overlap of detected logging using the

two techniques was 36%, 53% and 49% for the study years (figure 5.7).

 

1 00%

80%

60%

40%

20%

 

0%

l 992 1 996 l 999

 

E1 Visual only Autormtic only I Overlap (visual + automatic)

 

 

 

 

Figure 5.7. Visual interpretation, automatic analysis, and overlap areas.

Although visual interpretation was more successful in detecting logged forests
than automated detection, it was also more tirne-consuming. Furthermore, visual
interpretation is interpreter dependent and results are expected to vary among individuals.
In this analysis, all visual interpretations were made by a single individual, Eraldo

Matricardi.

97

Automated texture analysis provides a standardized technique for detecting
selectively logged forests that are relatively rapid and interpreter independent. Although
the technique provided less unique area (17-21%) of selective logging than the visual
techniques (26-47%), the areas detected may be important since they are often small
regions of newer logging that may expand in future years. It is important to point out that
before applying automatic texture analysis, Landsat images, band 5, were masked for
deforestation. Hence, some areas visually observed on satellite images were not analyzed
by automatic analysis, partially contributing with this difference in detecting selective
logging between the two techniques. In areas of overlap the automated procedure may be
more accurate as well since it is better able to deal with complex spatial mosaics of
logging with complex edges and unlogged islands than the visual methods.

Automated texture analysis and visual interpretation combine to form an efficient
methodological approach for estimating the area affected by selective logging in the
Amazon region. Moreover, the overlap areas that are common between the techniques are
useful in providing confidence in the estimates since they were detected twice by using

different methodologies.

5.5.2. Multi—temporal assessment of selective logging Amazon-wide

The final estimate of logged forest area was made using the union of the logged
forest polygons detected by the automated analysis (“cryptic” logging) and visual
interpretation (“obvious” and “subtle” logging). It also removed duplicated polygons in
the overlap areas among paths and rows of Landsat images in order to avoid double

counting.

98

Based on that, it was observed that the mapped logging areas are concentrated in
Para and Mato Grosso (90%), with 75% of all logging concentrated in the vicinity of two
major logging centers, Paragominas, State of Para and Sinop, State of Mato Grosso (see
appendix E3, E4, and E5). The balance of logged forests was found in the states of
Rondénia, Tocantins, Maranhao, and Acre. No logging was detected in the states of
Amazonas, Amapa, and Roraima.

Biophysical and socioeconomic factors, such as forest density, logging intensity,
clear cutting, and differing logging practices could be some of the main reasons for not
detecting some logged forest by using the present methodologies. Thus, the present
methodology is considered conservative, since, in different circumstances, logging areas
cannot be detected, especially in wetlands (varzeas) and in areas where logging was
followed immediately by deforestation.

It was observed that 13% of logged forests in 1992 were deforested by 1996, plus
15% by 1999, that is, 28% of logged forests in 1992 were deforested by 1999.

Additionally, 11% of forests logged in 1996 were deforested by 1999.

5.5.3. Analysis and evaluation of results and Assumptions

The process of logging can result in a heavily degraded forest environment.
Selective logging activities leave a mixture of intact forest, treefall gaps, roads, log-
loading patios, and damaged trees (Stone and Lefebvre 1998, Nepstad et al. 1999, Souza
and Barreto 2000). In spite of the damages and consequences of selective logging, few
attempts have been made to estimate the area of impact for the entire basin. Depending

on the methods used, the estimates have varied from between 10,000-15,000 km2 '1,

99

using survey data (Nepstad et al. 1999) to as little as 1,561 kmzyr", using visually
interpreted Landsat prints (Krug 2000). The resultant polemic over the importance and
scale of selective logging in the Amazon has recently been discussed by Alvarado and
Sandberg (2001).

The controversy about selective logging in the Brazilian Amazon is directly
related to the lack of accurate and comprehensive data on the spatial and temporal
distribution of this land use.. In this research, mapped distributions of logging activity in
Amazonian forests are presented for 1992, 1996 and 1999.

Due to the rapid disappearance of detectable selective logging fiom imagery,
annual rates of logging are hard to determine. Without annual imagery, it is necessary to
estimate logging rates based on specific knowledge of how the damage from logging
activity and subsequent ecosystem recovery interact across the landscape. Previous
research has shown that some logging activity can be detected in imagery more than a
year after the timber extraction but that none of the logged forests detected should be
more than 2-3 years old (Stone and Lefebvre 1998; Souza and Barreto 2000). Thus, in
order to estimate annual rates of selective logging, the results of the inter-annual analysis
(Chapter 4) performed for two case studies in the states of Mato Grosso and Rondénia
were extrapolated Amazon-wide.

The first assumption was that regions (Landsat scenes) showing an average of
more than 100 sz yr" of total logging (appendix E.2), were more intensive in terms of
forest impacts. Field studies showed that these types of logging operations were
generally planned in advance, prioritized extraction of more lucrative export quality trees,

and used heavy machinery and systematic field operations. Therefore, the results from

100

the Mato Grosso inter-annual study case were used to estimate annual logging rates from
total detected logging that was evident in the scenes with this type of extraction activity.

Selective logging in the remaining Landsat scenes had less than 100 sz yr" of
logging activity, which appeared to be more opportunistic than systematic. Field research
in these areas showed that loggers harvested fewer commercial trees and avoided
extensive construction of roads to support their operations by making use of the relatively
dense infrastructure of local colonization areas. Therefore, regions (Landsat scenes)
showing an average of less than 100 sz yr" of logging activity were assumed to be
represented by the Rondonia case study results.

Two different methods were used to estimate annual selective logging rates from
the total logging area that was evident in the satellite imagery. The first method simply
used average relations between detected logging and annual logging from the case
studies. Annual rates of new selective logging (increment) for 1992, 1996, and 1999,
were estimated using the average annual increment percentages (calculated as total area
of forest showing logging activity for the first time divided by the total area of forest
wherein logging was detected) observed in the Mato Grosso and Rondenia study cases,
39.5% and 91.42% respectively.

The second method made use of the observable trend in the changing relationship
between detected logging and actual new logging activity. This approach accounted for
the fact that, in the areas with intense logging activity, the average persistence time for
previously logged forest evident in subsequent imagery-based logging detection
increased over the period of the study (1992-1999). The reasons for this change are not

certain but likely result from increased volumes of timber extraction over time as both

101

smaller trees and new species became marketable (Uhl et al. 1997) and further, post-
logging degradation of forests by fire (Cochrane et al. 1999). The rate changes were
calculated using separate equations for each case study region (see chapter 4). For the
Mato Grosso case study the increment of new logging areas was 60%, 48%, and 38% for
1992, 1996, and 1999, respectively. Slightly increasing rates were found for the Rondenia
case study, 82%, 87%, 91% for 1992, 1996, and 1999, respectively.

The differences between the total area detected and that which was new logging

areas was quite large in the Mato Grosso region (60.5%) but much smaller for Rondenia

(8.58%). These residual areas were considered to mostly be areas of previous logging
that, due to severe forest damage, persisted as detectable logging on satellite imagery.
However, it is noteworthy that, a small portion of the former logging was actually
secondary logging of previously logged sites (7% and 0.4% of the total detected for Mato
Grosso and Rondenia, respectively).

Assurrring that the equations presented in the Chapter 4 can be extrapolated basin-
wide, the total increment of new selective logging areas in 1992 was, at least 3,689 sz.
By 1996 and 1999, the annual selective logging rate expanded to 5,107 sz, and 11,638
sz, respectively. The remaining detected forests with logging that were not new to the
actual year of the imagery grew from 2,197 sz in 1992, to 5,031 sz in 1996, and
14,962 sz in 1999. The total areas detected and annual rates of logging calculated by

the two described methods are presented in table 5.6.

102

Table 5.6. Total detected and increment of new logging areas Amazon-wide

 

Amazon-wide logging area (Km2)

 

 

 

 

 

Year Total Increment of new logging
detected Based on equations (*) Based on annual average (*)
1992 5,979.80 3,689 2,547
1996 10,064.10 5,107 4,361
1999 26,085.40 1 1,638 1 1,889 I

 

* Note: Extrapolated fiom the Mato Grosso and Rondonia case studies (see chapter 4)

 

Although for 1999 the estimates of logging increment are almost the same, the

estimated areas for 1992 and 1996 have a more significant difference. It seems that the

equations can better represent the increment of selective logging basin-wide, since there

is a natural reduction of raw material (timber), which forces loggers to increase intensity

and revisit logged forests over time.

New logging areas increased 138% and 227% in the period between 1992 and

1996, and between 1996 and 1999, respectively. This supports the reports by Uhl et al.

(1997) who observed that loggers are advancing more and more into undisturbed forests

in search of new raw materials, as well re-logging some forests for second tier economic

species.

Based on the results of the case study in the State of Mato Grosso, presented in

the Chapter 4, formerly logged areas are increasingly being re-logged. An average of

2.9%, 8.6%, and 10.3% of the total logged forest where revisited in the period of 1993-

1996, 1996-1999, and 1999-2002, respectively. This shows that an increasing scarcity of

raw materials is forcing the timber market and loggers to adjust to the natural resource

103

availability, revisiting logged forests searching for previously non-commercial sized or
species of trees.

Despite the methodological limitations discussed above, the logging estimates
from this study (1996, 10,064 km2 overall logging detected and 5,107 km2 of new logging
areas estimated) are well in the range of other published estimates for 1996, 1,561 km2
(Krug 2000) and 10,000-15,000 km2 (Nepstad et al. 1999). In any case, the rate of growth
in the area of detected logging, from approximately 5,980 km2 in 1992 to over 26,085
km2 in 1999 is substantial. Even comparisons of only the most visible canopy degradation
within the satellite images show 3,936, 6,524, and 17,600 sz respectively for 1992,
1996, and 1999. This indicates a 66% increase between 1992 and 1996 and a 347%
increase between 1992 and 1999. In total, it was estimated that at least 37,465 km2 of
forest were logged from 1992 to 1999 of which approximately 26,085 km2 was still

extant on the landscape in 1999.

5.5.4. Final considerations

Factors that could help explain the apparent large increase in logging rates during
the 1996-1999 time period as compared with 1992 to 1996 period include the legislative,
resource and climate/disturbance explanations.

One hypothesis is that the changes in apparent logging rates are the result of the
logging industry’s response to a 1996 Executive Order by the Brazilian President. In
1996, Brazil’s Forest Code was changed to stipulate that 80%, not the previously
legislated 50%, of forests on individual properties must be protected. This action

prohibited landowners from deforesting more than 20% of their landholdings. With less

104

 

 

logs reaching the sawmills from newly deforested areas, loggers may be expanding their
presence in standing forests to acquire the timber necessary to maintain their production
levels. This would help to explain the sudden increase in the amount of selectively logged
forests.

A second hypothesis is related to forest production and productivity. According to
IBGE (2003), the round wood production in the Brazilian Amazon did not increase
during 1997 to 1999, being approximately estimated of 20, 16, and 16 millions of cubic
meters per year, respectively in 1997, 1998, and 1999. If the round wood production did
not increase, it may be that forest productivity was drastically lowered after 1996. This
would be a function of resource utilization where loggers utilize highly productive forests
first and only use more marginal forest once high production areas decline. The fact that
80% of all of the increase in logging areas is concentrated in just 9 of the 38 images
indicating logging activity supports this hypothesis.

A third hypothesis is that climate events, such as the1997-1998 “El Nino”, could
have increased the degradation in logged forests through drought and fire, thereby
enhancing the detection of selectively logged forests due to their more heavily degraded
canopies.

All three hypotheses may partially explain the observed changes in detectable
logging. Actual growth in the logging industry across the Amazon Basin should not be
discounted either. In any case, the area impacted by selective logging is rapidly

increasing making it a land use of growing importance in Amazonian forests.

105

Wan-5W —

5.6. References

ALVARADO, E. and SANDBERG, D V. 2001. Logging in Tropical Forest: Literature
review on Ecological Impacts. University of Washington. USDA Forest Service.
Internet site www.fs.fed.us/pnw/fera.

COCHRANE, M. A., ALENCAR, A., SCHULZE, M.D. SOUSA jr, C.M., NEPSTAD,
D.C. LEFEBVRE, P., and DAVIDSON, E. 1999. Positive feedbacks in the fire
dynamic of closed canopy tropical forests. Science. 284, 1832-1835.

FEARNSIDE, P.M., TARDIN, A. T., and FILHO, L.G.M. 1990. Deforestation in
Brazilian Amazonia. PR/SCT, Instituto de Pesquisas Espaciais, Instituto Nacional ..
de Pesquisas da Amazonia.

FEARNSIDE, PM. 1997. Greenhouse gases from deforestation in Brazilian Amazonia:
net committed emissions. Climate Change. 35. 321-360.

 

F RUMHOF F , P. C. 1995. Conserving Wildlife in Tropical Forests Managed for Timber
to provide a more viable complement to protected areas. BioScience. 45. 456-464.

HOLDSWORTH, A. R and Uhl, C. 1997. Fire in Amazonian Selectively Logged Rain
Forest and The Potential for Fire Reduction. Ecological Aplications. 7, 713-725.

HOUGHTON, RA. 1996. Terrestrial sources and sinks of Carbon inferred from
Terrestrial data. Tellus. 48B, 420-432

HOUGHTON, R. A. 1997. Terrestrial Carbon Storage: global lessons for Amazonian
Research. Ciencia Cultura. 49. 58-72.

IBGE. 2003 - Sistema IBGE de Recuperacao Automatica. Extracao Vegetal. Internet site
IBGE — Instituto Brasileiro de Geografia e Estatistica [Internet]. 2003 - Sistema IBGE
de Recuperacao Automatica. Extracao Vegetal. Rio de Janeiro, RJ. Available from
http://www.ibge.gov.br/

J ANECZEK, DJ. 1999. Detection and Measurement of Amazon Tropical Forest
Logging using Remote Sensing Data. A Master’s Thesis presented to Michigan
State University, Department of Geography.

KRUG, T. 2000. Deforestation and Fire in the Brazilian Amazonia: a historical
perspective. In. LBA First Scientific Conference. Belem. Para. June 26-30, 2000.

MARTINI, A. M. Z., ROSA, N. A., UHL, C. 1994. An attempt to predict which

Amazonian tree species may be threatened by logging activities. Environmental
Conservation. 21, 152-161.

106

NEPSTAD D.C., VERISSIMO A., ALENCAR A., NOBRE C., LIMA E., LEFEBVRE
P., SCHLESINGER P., POTTER C., MOUTINHO P., MENDOZA E.,
COCHRANE M.A., BROOKS V. 1999. Large Scale impoverishment of
Amazonian forests by logging and fire. Nature. 398. 505-508.

SOUZA, C. Jr. and BARRETO., 2000. An alternative approach for detecting and
monitoring selectively logged forests in the Amazon. International Journal of
Remote Sensing, 21, 173-179.

STONE, T. A. and LEFEBVRE, P. 1998. Using multi-temporal satellite data to evaluate
selective logging in Para, Brazil. International Journal of Remote Sensing, 19,
2517-2526.

SKOLE, D. and TUCKER, C. 1993.Tropical deforestation and habitat fragmentation in
the Amazon: satellite data from 1978 to 1988, Science 260, 1905-1910.

UHL, C. and VIEIRA, I.C.G. 1989. Ecological impacts of selective logging in the
Brazilian Amazon: A case study from the Paragominas region of the state of Para.
Biotropica. 21, 98-106.

UHL, C., VERISSIMO, A. MATTOS, M.M., BRANDINO, Z., and VIEIRA, I.CG. 1991.
Social, economic, and ecological consequences of selective logging in an Amazon
frontier: the case of Tailandia. Forest, Ecology, and Management. 46, 243-273.

UHL, C., BARRETO, P., VERISSIMO, A., VIDAL, E., AMARAL, P., BARROS, A.C.,
SOUZA, C. JOHNS, J ., GERWING, J. 1997. Natural Resources Management in the
Brazilian Amazon: An integrated research approach. BioScience. 47, 160-168.

VERISSIMO, A., P. BARRETO, M. MATT OS, R. TARIFA, AND C. UHL. 1992.
Logging impacts and prospects for sustainable forest management in an old
Amazonian frontier: the case of Paragominas. Forest, Ecology, and
Management. 55, 169-199.

VERISSIMO, A., BARRETO, P., TARIFA, R., and UHL, C. 1995. Extraction of a high
value resource from Amazonia: The case of mahogany. Forest Ecology and
Management. 72, 39-60.

VERISSIMO, A., SOUSA, C., STONE, S., and UHL, C. 1998. Zoning of timber
extraction in the Brazilian Amazon. Conservation Biology. 12, 128-136.

107

CHAPTER VI

Concluding Remarks

6.1.Seating the current study in the context of global change research

The impact of logging in tropical forests is considered significant in terms of
forest degradation and fire susceptibility (Stone and Lefebvre 1998, Nepstad et al. 1999,
Souza and Barreto 2000). F eamside (1997) estimated that in the Brazilian Amazon, the
impact of selective logging in undisturbed forest creates a net carbon flux equal to
approximately 4-7% of the annual carbon release from deforestation.

In spite of the important contribution of selective logging in global change, the
majority of selective logging has not been detected by remote sensing techniques and,
therefore, the total area of forest damaged by human activities has been underestimated.
Consequently, the current carbon flux estimates for the Amazon region may be low
because logging areas have not been accounted for (Nepstad et al. 1999).

A8 presented in the Chapters 1 and 5, Nepstad et al. (1999), J aneczek (1999), and
Krug (2000) have provided Amazon-wide estimates of selective logging impacts. Based
on a survey data of sawmill production, Nepstad et al. (1999) estimated that 10,000-
15,000 krnzyr‘l of undisturbed forests were selectively logged in the period of 1996-
1997. Based on visual interpretation of Landsat imagery, Krug (2000) estimated as little
as 1,561 kmzyr" in the given period. Based on visual interpretation and an automated
method of Landsat image analysis, Janeczek (1999) estimated a total of 5,406 km2 of

forests were selectively logged over 1991-1992.

108

In this research, I presented new estimates through a multi-temporal assessment of
selective logging Amazon-wide for 1992, 1996, and 1999, based on remotely sensed data
and field studies. The results show that the total active areas of selective logging in the
Brazilian Amazon have significantly increased from 3,689 to 5,107 and from 5,107 to
11,638 sz in 1992-1996 and 1996-1999 periods, respectively. It represents an
increment of roughly 38% and 128% in the given periods, respectively.

Furthermore, based on visual interpretation and field observations, I observed
that most of the forests heavily impacted by selective logging were previously
misclassified as deforested areas using an unsupervised classification of Landsat imagery.
The total of 341, 586, and 2,706 sz of forests showing highly damaged canopy and soil
exposure on Landsat imagery were reclassified as selective logging for 1992, 1996, and
1999, respectively. It is likely that logging intensity and forest fire occurrence were
higher in 1999 than in the previous years of this analysis.

It was also observed that in 1999 loggers revisited logged forest more often than
in 1992, increasing the rate of revisiting from 1.2% to about 10% of the total logged area
detected in 1999. It shows that loggers are now re-visiting former logging areas
searching for previously non-commercial tree species that presently are market demanded
due a possible raw material scarcity in the Amazon timber centers. That said, it is
expected that selective logging is more likely to expand into new frontiers as well as into
protected areas, where raw material for timber industries is still abundant.

Finally, it is noted that the methodology used in this study is considered

conservative since it may omit very-low intensity logging areas and logging in riparian

109

forests. Nevertheless, the data presented here represent the first multi-temporal

assessment of the amount and locations of selective logging in the Brazilian Amazon.

6.2. Research questions revisited

o Is selective logging a significant disturbance in terms of area increment of total

forest cleared or damaged?

By applying visual interpretation and automated methods presented in this research,
selective logging was shown to be rapidly increasing in the Brazilian Amazon. A total of
5,980 him2 of forests were selectively logged by 1992, 10,064 Km2 by 1996, and 26,085
sz by 1999. Assuming that the results of the multi-annual analyzes in two case studies,
in the State of Mato Grosso and Rondonia, represent a general trend and therefore can be
extrapolated basin-wide (see Chapter 4 and 5 for further details), it is reasonable to
estimate that the total area of undisturbed forests selectively logged in 1992, 1996, and
1999 was roughly 3,689, 5,107, and 11,638 sz, respectively. I also estimate that
revisited logging areas are increasing from 1.7% to more than 10% from 1992 to 2002,
respectively. In any case, the areas impacted by selective logging are rapidly increasing,

making it a land use of growing importance in the Brazilian Amazon.

. Can we efficiently measure it basin-wide as a form of degradation in addition to

outright deforestation?

The automated texture analysis and visual interpretation combined to form an efficient

methodological approach for estimating the area affected by selective logging in the

110

 

Amazon region. Based on the findings of the field studies and remote sensing analyzes
conducted in research for the entire Brazilian Amazon, the most impacted selective
logging areas were surely detected and mapped. Even areas of reduced impact logging
could be estimated. However, this methodology is considered conservative, since, in

different circumstances, logging areas could not be detected.

0 Is logging causing more deforestation than would otherwise occur? ..

Based on the basin-wide results of this research, 13% (796 sz) of logged

 

forests in 1992 were deforested by 1996, plus 15% (896 sz) by 1999. An additional
11% (1,074 sz) of logged forests in 1996 were deforested by 1999. Based on the study
case in the State of Mato Grosso, around of 1,900 sz of logged forests were deforested
between 1992-2002, which represents about 17% of the total (11,449 sz) logged forest
in that given period. In the State of Rondénia case study, 15% (72.6 sz) of logged
forests between 1989-2001 were deforested in this given period. Although deforestation
does not appear to be following selective logging in the period of analysis, it is more
likely to occur in the oldest logging areas. Moreover, as discussed in Chapter 1, there is a
synergism between selective logging and deforestation and further consequences on
natural forest are expected from it. In this regard, forest fire seems to be the most eminent
risk enhanced by selective logging in the Brazilian Amazon as result of its increased
impact in area and intensity (re-visited areas) of forest degradation. Other indirect
impacts also have to be considered, such as the financial support for agricultural systems,

access opening to new frontier, etc.

111

6.3. Significance and implications of this study

This study is the first multi-temporal assessment of the amount and locations of
selective logging in the Brazilian Amazon. The results presented here are fundamental to
evaluating the consequences of land cover change in terms of global climate change and
carbon cycle studies. These studies have yet to take into account the amount of carbon
released in the atmosphere from the large-scale selective logging. Complementarily, this
research provides useful information to promote sustainable management of tropical
forests.

In terms of land use and land cover change, this study extends the current multi-
annual Amazon-wide image classification techniques to include selectively logged forests
in addition to the more frequently used thematic classes (forest, deforestation, regrowth,

cerrado, cloud, cloud shadow, and water).

6.4. Opportunities for further studies

Although selective logging could be detected using the methodology presented in
this study, the intensity of forest damage caused by selective timber harvesting has not
yet been efficiently measured using remote sensing techniques. New approaches using
finer spatial resolution and hyperspectral satellite imagery as well as other alternative
approaches should be developed in order to improve the assessment of selective logging
impacts.

Dr. J iaguo Qi, from Michigan State University, has been developing
methodological approaches to evaluate forest fragmentation using remotely sensed data.

These ongoing studies are based on fractional percentage of green vegetation technique

112

 

that may be used to estimate forest degradation. However, this method may be tested in
different study sites before application to the entire Brazilian Amazon. The combination
of visual interpretation, automated methods, and fractional coverage may be a
methodology approach to firlly support the research on global changes.

In this study, I showed a 10 year trend of logging in the Amazon, but the long-
term impacts of selective logging are unknown. The process of selective logging and its
synergism with tropical deforestation is poorly understood in the Brazilian Amazon. :

Thus, future research is needed to fillly understand the process of land use and land cover

 

change in which selective logging is inserted. Finally, the development of scenarios for
the timber sector could support public policies to achieve a sustainable development in

the region.

6.5. References

F EARNSIDE, P.M., TARDIN, A. T., and FILHO, L.G.M. 1990. Deforestation in
Brazilian Amazonia. PR/SCT, Instituto de Pesquisas Espaciais, Instituto Nacional de
Pesquisas da Amazonia.

FEARNSIDE, PM. 1997. Greenhouse gases from deforestation in Brazilian Amazonia:
net committed emissions. Climate Change. 35, 321-360.

FEARNSIDE, P. M. 1999a. Biodiversity as an environmental service in Brazil’s
Amazonian forests: risks, value, and conservation. Environmental Conservation 26,
305-321.

FEARNSIDE, P. M. 1999b. Green Gases from Deforestation in Brazilian Amazenia: Net
Committed Emissions. Climate Change 35, 321-360.

HOUGHTON, R. A. 1997. Terrestrial Carbon Storage: global lessons for Amazonian
Research. Ciencia e Cultura. 49, 58-72.

HOUGHTON, R.A., SKOLE, D. L., NOBRE, C. A., HACKLER, J. L., LAWRENCE,
K.T., CHOMENTOWSKI, W.H. 2000. Annual fluxes of carbon from deforestation
and regrowth in the Brazilian Amazon. Nature 403, 301-304.

113

INPE- NATIONAL INSTITUTE FOR SPACE RESEARCH. 2003. Monitoring of the
Brazilian Amazon by Satellite. Sio Jose dos Campos, SP, Brazil. Available fi'om
http://www.inpe.br/

KRUG, T. 2000. Deforestation and Fire in the Brazilian Amazonia: a historical
perspective. In. LBA First Scientific Conference. Belem. Para Brazil June 26-30,
2000.

Q1, J. 2003. Personal communication and based on manuscript: Interpretation of Spectral
Vegetation Indices and Their Relationship with Biophysical Variables. Michigan
State University. East Lansing, MI. USA. In preparation. L

LAURENCE, W. F ., COCHRANE, M. A., BERGEN, S., FEARNSIDE, P.M.,
DELAMONICA, P., BARBER, C., D’ANGELO, S., FERNANDES, T. 2001. The
Future of the Brazilian Amazon. Science. 291, 438-439.

 

NEPSTAD D.C., VERISSHVIO A., ALENCAR A., NOBRE C., LIMA E., LEFEBVRE
P., SCHLESINGER P., POTTER C., MOUTINHO P., MENDOZA E., COCHRANE
M.A., BROOKS V. 1999. Large Scale impoverishment of Amazonian forests by
logging and fire. Nature. 398, 505-508.

NEPSTAD, D., McGRATH, D., ALENCAR, A., BARROS, A.C., CARVALHO, G.,
SANTILLI, M., DIAZ, M.C.V. 2002. Frontier Governance in Amazonia. Science’s
Compass. 295, 629-631.

SOUZA, C. Jr. and BARRETO A., 2000. An alternative approach for detecting and
monitoring selectively logged forests in the Amazon. International Journal of
Remote Sensing, 21. 173-179.

STONE, T. A. and LEFEBVRE, P. 1998. Using multi-temporal satellite data to evaluate
selective logging in Para, Brazil. International Journal of Remote Sensing, 19.
2517-2526.

SKOLE, D. and TUCKER, C. 1993.Tropical deforestation and habitat fragmentation in
the Amazon: satellite data from 1978 to 1988. Science. 260. 1905-1910.

114

 

APPENDICES

115

 

 

Appendix A.1. Timber production in Brazil’s Amazonian states - 1990 to 2001

116

3.. 552a.

Amoomvmoe ”ooh—om

 

 

 

 

55©.owo.~ mmw.mmm 3348.: £3.58de HHS—Em mg. Sim “42.53.34 metacvd mo5.cmm.m 38.
5.3.3 oofimm cmm.5om emmwvoxz m_m.m~5.m 5.1.5? 0.3me 5cm. 5 www.mvm Son
wmmfim 81mm m _ Q53 Sm. L 5.3 cmo.ooc.m Rad? www.mow 0 33cm Sodom ooom
ommfio cemdm v3.35 omo.mmm._ _ 3m6$t5 3”.on 35.85 35.3 $4ch 32
Nance owfla— www.mem wmv. 3;.N— O5w.o5m.m $343 Nmo.~w5 55o.m5 mmmdom moi
314.8 0%.: ooodwm m5m.mvc.m_ amo.53.m $0.59. 81m m5 mem.5m 5ww.m _ N 552
550.8 momd— ooodwm mm m.ww5.5m midodv ~59on www.mmo cm5.m5 Se.» 2.. 052
owm.5vm - mm _ .59.; 555.05.? O55.cmm.v O5 33o; 80.0mm 31% m mam. _ mm 32
~5v.55m 85.2 mmodm: w5o.wmm.3 S 133V manage; ~ 3.03 mmodmm mm5.m5m ema—
5ov.5mv - 03.3 m; 85.52.34 _5m.om5.m mmwdw 3 mmmfiov 30.63 395m m 82
O55d ~ m omo.5m 5oo.§5.v 35.35. _ m 5oo.5oc.m 3 ~. 3 3 fl 5.5.3 53.2 m 3 L.Gwm moi
omm.mmv 536m mom.5moJ Kodcmdm S5.¢5w.m mm H. _ mm mmwdfi m3.mmm mm5.vom 53
22684 58.2 03.804 $560de omodme 0863 mid 3.5m 35.5mm meow. Sm om?

2:580... «88.8% «Eoecom «can 88.5 8.32 3:532 Began? «mag Bo<
A mEv aerospace woo? venom Hmo>

 

117

 

Appendix 8.]. Study site location at Sinop municipality, State of Mato Grosso, Brazil.

118

 

 

       

:— 55.5..
mu
m 23.50:! gw 8m 0 g
§§§§ 8a 0 OON
\ v - - ...... .33 «5 3.3.3 .35 I
\- j .- - -- f-.- 55.5 an»; 2.55 n. W
K., r. .........
\ M
l. J
. v
_ . .
A stun—m ...\
.
p as _,
. tom f
.\_ wNN .
[If 5... W
J] }‘ le
H N
f a, ill I I III I I ‘0 I I
Ommohmv 052 3 33m
I=N<mm z

 

 

119

Appendix B.2. Satellite images used in the Mato Grosso case study.

120

 

 

 

Satellite / Scene Band Spatial Acquisition
Sensor resolution (m) date
Ikonos PO_44060 1, 2, 3, 4 4 x 4 04/30/2000
Ikonos PO_44060 Panchromatic 1 x 1 04/30/2000
Landsat ETM+ WRS_226/068 1, 2, 3, 4, 5 30 x 30 06/18/2000
Landsat ETM+ WRS_226/068 1, 2, 3, 4, 5 30 x 30 08/19/1999
Landsat TM WRS_226/068 1, 2, 3, 4, 5 30 x 30 06/05/1998
Landsat TM WRS_226/068 1, 2, 3, 4, 5 30 x 30 08/05/1997
Landsat TM WRS_226/068 1, 2, 3, 4, 5 30 x 30 07/01/1996
Landsat TM WRS_226/068 1, 2, 3, 4, 5 30 x 30 06/29/1995
Landsat TM WRS_226/068 1, 2, 3, 4, 5 30 x 30 07/12/1994
Landsat TM WRS_226/068 1, 2, 3, 4, 5 30 x 30 08/26/1993
Landsat TM WRS_226/068 1, 2, 3, 4, 5 30 x 30 05/19/1992

 

Appendix 8.2

121

 

Appendix B.3. Logged forests detected using Landsat and Ikonos images and different

remote sensing techniques (automated and visual interpretation).

122

mm 55...?

non—3:53 8.38985 33$ 28 omen: 8:8: 8:25:33 a wEm:

8&9: mm? 8.3 £5. .86 52% ofi 5.23 388 832 we $2: mm 38.5.3 33 8:38: meow; mo :38 2; ”802

.5893: .8984 main 5.88% wEwmo. 268—8 Sow $538.. 628 wao. .«o 988 :oBmMEO .m

.bowmg “82.3 $5.: 838% wEwwE 03823 Bob wet—:8.— Hmoeom wa2 ho macs coamgou .N

.56me “8.353 98 8:8: wEm: tongue “mono.“ Romme— mo 83 38 2: 5233 3:30 ._

 

 

 

“.9; 50.5% _ 0N 3.0m 9mm 8&3 fig emdmfi 33$ + E O5m Sham
3.2 3.3m ~ me mm€m flow 8ch 5.3 V5; #2 13m? + E ovm Sham
Q5: 9.3: m6 365 fig 3.3m 5.3 3.3 2 32> + 8 3m swam
5.0: 2.. 82 M; 362 v.2 8de NS 9.3: 33$ + E om: Stem
mi: $.82 o. 2 9.63 5.3 $9: Yaw 8.2 Z 353 + E o2 swam
of: $.83 m5 ~03 v.3 0:8 Ono m5d5: E ©5m swam
3w: v03: Nd— m ~69 cam ammo». mdw m~.5m: E ovm Sham
wdc— modwm _ w.m fl O56: 9mm 033m New on. 53 8 2m Sham
mam $.82 <3 3. Sm ow. 2.0mm adm $6.53 E of Hotsm
fiww 59m _ Z oem .5.on 3; :6: 3.5 enema E o2 Sham
3.5 @330 0.0.5. $53 0. 9.2 v.m5 vodma Eco 32>
.x. 3.; .x. 6 8.260 e. S sewageoo .x. 8 .555
$2.5m H3286
Amoafiooé 8?. 03:38.5

 

123

Appendix C. 1. Study site location at Cujubim municipality in the State of Rondénia,

Brazil.

124

 

5 552...

 

H

  

 

 

      

Ev. coop com o 03

:o~aE<
$48 .33 2t 87.5.. =35 I
. coup-5‘ .mmog 9:50 P... _
.a fll .l .....

//i||.. .m
_.
fi U
_. _.._
,. ..
_ fw l/‘I\.\l\
an in .512 xv
_, i t 1:7...IL.
I x ., ..\
.. Scrum . a i _. 8
._ _. 31.2.5. ,. .8
dr/tl/ V. a. ,., lit-ill... llllllx
x ._ ll 1..

 

«285$ he saw . , . . <

 

 

125

Appendix C.2. Forest inventory map (logged tree locations, and transects, patio and road
locations in the field)

126

 

 

 

 

 

name 9&3 m

moon vomwoq E
38889. g

mona owSoum D
szom‘H

 

 

 

8M

N

8 52.33

  

 

 

127

Appendix C.3. Textural algorithm image and buffer radius of 450 meters around log

landings captured using automated method. Path/row 232/066

128

 

 

O 1 2 3 4 5 6 Kilometers

 

LEGEND
- Study elte border

I Buffer radius (460 m)

9 Log landings

 

 

 

Appendix C.3.

129

Appendix C.4. Selective logging areas estimated using fieldwork measurements in the
study site at Manoa ranch, Cujubim municipality, State of Rondonia,

Brazil.

130

 

 

Buffer

 

 

Total detected Total buffer Omission Commission

(m) Trees Ha % Ha % Ha % Ha %

150 1147 557.80 27.72 557.83 22.67 1,903.06 77.33 - 0.00
180 1134 720.47 27.40 720.47 29.28 1,740.39 70.72 - 0.00
210 1736 880.19 41.95 880.19 35.77 1,580.67 64.23 - 0.00
240 2005 1,034.97 48.45 1,034.97 42.06 1,425.89 57.94 - 0.00
270 2250 1,189.71 54.37 1,189.71 48.35 1,271.14 51.65 - 0.00
300 2514 1,342.58 60.75 1,342.58 54.56 1,118.28 45.44 - 0.00
330 2777 1,495.16 67.11 1,495.16 60.76 965.70 39.24 - 0.00
360 2979 1,645.64 71.99 1,645.64 66.87 815.22 33.13 - 0.00
390 3175 1,793.82 76.73 1,793.82 72.89 667.04 27.11 - 0.00
420 3378 1,941.09 81.63 1,941.09 78.88 519.77 21.12 - 0.00
450 3549 1,954.19 85.77 2,083.16 79.41 506.67 20.59 - 0.00
480 3745 2,057.42 90.50 2,220.49 83.61 403.44 16.39 - 0.00
510 3829 2,135.36 92.53 2,335.26 86.77 325.50 13.23 - 0.00
540 3905 2,198.75 94.37 2,438.47 89.35 262.11 10.65 - 0.00
570 3963 2,253.76 95.77 2,537.06 91.58 207.10 8.42 76.20 3.10
600 4008 2,302.23 96.86 2,632.52 93.55 158.62 6.45 171.67 6.98
630 4068 2,344.17 98.31 2,724.98 95.26 116.68 4.74 264.12 10.73
660 4092 2,373.47 27.72 2,816.82 96.45 87.39 3.55 355.96 14.46
690 4112 2,393.02 27.40 2,908.42 97.24 67.84 2.76 447.56 18.19
720 4122 2,407.93 41.95 2,999.56 97.85 52.93 2.15 538.71 21.89

 

Appendix C.4 Note: Total of logged area = 2,460.86 hectares (100%) from fieldwork.

131

. -‘._fl.r

 

Appendix C.5. Selective logging areas estimated using Textural algorithm and Landsat
ETM+, band 5. Study case in the State of Rondonia, Brazil. Path/row

232/066 acquired in 08/05/2002.

132

 

 

 

 

Buffer Trees Total detected Total Buffer Omission Commission
(m) Ha % Ha % Ha % Ha %
150 1897 1,065.40 45.84 1,070.72 43.29 1,395.46 56.71 - 0.00
180 2165 1,212.67 52.32 1,222.28 49.28 1,248.19 50.72 - 0.00
210 2653 1,355.91 64.11 1,372.87 55.10 1,104.94 44.90 - 0.00
240 2858 1,489.37 69.07 1,518.07 60.52 971.49 39.48 - 0.00
270 3068 1,615.44 74.14 1,659.52 65.65 845.42 34.35 - 0.00
300 3230 1,731.03 78.06 1,794.42 70.34 729.83 29.66 - 0.00
330 3387 1,837.75 81.85 1,925.21 74.68 623.11 25.32 - 0.00
360 3512 1,937.18 84.87 2,052.38 78.72 523.68 21.28 - 0.00
390 3631 2,030.09 87.75 2,175.69 82.50 430.77 17.50 - 0.00
420 3728 2,108.23 90.09 2,287.83 85.67 352.63 14.33 - 0.00
450 3808 2,167.93 92.03 2,3 89.42 88.10 292.93 1 1.90 - 0.00
480 3875 2,218.55 93.64 2,487.33 90.15 242.31 9.85 26.47 1.08
510 3939 2,259.90 95.19 2,581.18 91.83 200.96 8.17 120.32 4.89
540 3975 2,294.64 96.06 2,673 .58 93.25 166.22 6.75 212.72 8.64
570 4014 2,322.59 97.00 2,764.41 94.38 138.27 5.62 303.55 12.34
600 4043 2,347.09 97.70 2,855.20 95.38 1 13.77 4.62 394.34 16.02
630 4080 2,369.33 98.60 2,945.28 96.28 91.52 3.72 484.43 19.69
660 4099 2,389.06 99.06 3,036.16 97.08 71.79 2.92 575.30 23.38
690 4109 2,405.98 99.30 3,127.17 97.77 54.87 2.23 666.31 27.08
720 4115 2,418.89 99.44 3,218.12 98.29 41.97 1.71 757.26 30.77

Appendix C.5 Note: Total logged forest = 2,460.86 hectares (100%) fiom fieldwork.

133

 

Appendix D.1. Landsat scenes used in this analysis (Path/row 232/062 and 226/068).

134

 

3 58...?

 

 

 

 

.538 84 e 8' .1fi..£s..§..§8.:. 3D.

 

 

 

20% in... ”.420

   

 

53:3 .30.. 2... 233.8 :55 4E.

 

88:2 .83 £85 WU

 

 

135

 

Appendix D.2. Landsat (TM and ETM+) images (path/row 232/067) used for the inter-

annual analysis in the State of Rondonia, Brazil

136

Path/Row 232/067

 

 

 

Year Date Sensor Cloud/Shadow Smoke
1989 August 08, 1989 TM 0% 6%
1990 December 2, 1990 TM 0% 0%
1991 June 01, 1991 TM 8% 0%
1992 June 22, 1992 TM 0% 0%
1993 October 07, 1993 TM 0% 3%
1994 June 04, 1994 TM 1% 0%
1995 July 25, 1995 TM 2% 1%
1996 June 25, 1996 TM 0% 0%
1997 June 28, 1997 TM 0% 0%
1998 July 17, 1998 TM 0% 0%
1999 December 16, 1999 ETM+ 3% 0%
2000 June 28, 2000 ETM+ 0% 0%
2001 August 11, 2001 ETM+ 6% 1%
Appendix D.2.

137

Appendix D.3. Landsat (TM and ETM+) images used for inter and intra-annual analysis

in the State of Mato Grosso, Brazil.

138

Path/Row 226/068 - Inter-annual analysis

 

Year Date

Sensor Cloud/Shadow Smoke

 

1992 May 19, 1992
1993 August 26, 1993
1994 July 12, 1994
1995 June 29, 1995
1996 July 01, 1996
1997 August 5, 1997
1998 June 5, 1998
1999 August 19, 1999
2000 June 18, 2000
2001 August 8, 2001

2002 April 21, 2002

TM

TM

TM

TM

TM

TM

TM

ETM+

ETM+

ETM+

ETM+

0%

0%

0%

8%

0%

0%

1%

2%

0%

0%

0%

0%

6%

0%

0%

0%

3%

0%

1%

0%

0%

0%

 

Intra-annual analysis

 

2001 May 4, 2001
August 8, 2001
October 27, 2001

2002 April 21, 2002

ETM+

ETM+

ETM+

ETM+

3%

0%

6%

0%

0%

0%

1%

0%

 

Appendix D.3

139

i ”In hJ‘" '

Appendix D.4. Study case results of deforestation and selective logging for 2001 in the

State of Mato Grosso.

140

 

 

Logging Deforestation Increment (Km2) Deforestation of

 

 

 

 

(date) logged forests

May August October April (2001 to 2002)
2001 2001 2001 2002 (Km2)

May 2001 - 9.37 3.43 10.81 23.60

August 2001 - - 0.34 5.83 6.17

October 2001 - - — 6.38 6.38

April 2002 - - - - -

Tm” Defms‘ed 9.37 3.77 23.02 36.15

Appendix DA

141

 

Appendix E.1. Landsat imagery used for the basin-wide logging analysis

142

 

 

Path/Row

Acquisition Date

 

 

1992 1996 1999
221/063 July 11, 1992 May 30, 1997 May 14, 2000
222/062 July 24, 1991 July 5, 1996 July 14, 1999
222/063 July 24, 1991 June 3, 1996 August 23, 1999
222/064 (*) (*) August 23, 1999
223/062 August 16, 1991 July 12, 1996 July 23, 1999
223/063 May 28, 1991 July 12, 1996 July 13, 1999
223/064 August 10, 1992 June 10, 1996 July 13, 1999
223/065 June 2, 1993 June 10, 1996 July 23, 1999
223/066 July 25, 1992 May 9, 1996 July 13, 1999
224/062 (*) (*) October 24, 1999
224/063 June 22, 1992 July 19, 1996 October 8, 1999
224/065 July 16, 1992 July 19, 1996 October 8, 1999
224/066 July 16, 1992 July 3, 1996 October 8, 1999
224/067 July 16, 1992 July 3, 1996 October 8, 1999
225/067 (*) (*) August 12, 1999
225/068 (*) (*) August 12, 1999
225/069 June 21, 1992 July 1, 1996 August 12, 1999
226/063 July 20, 1991 June 18, 1997 August 3, 1999
226/066 (*) (*) August 3, 1999
226/067 May 19, 1992 July 1, 1996 August 19, 1999
226/068 May 19, 1992 July 1, 1996 August 19, 1999
226/069 May 19, 1992 July 1, 1996 August 19, 1999
227/062 (*) (*) August 10, 1999
227/065 (*) (*) August 10, 1999
227/067 August 6, 1992 July 27, 1996 August 10, 1999
227/068 August 6, 1992 June 6, 1996 August 10, 1999
227/069 July 5, 1992 June 6, 1996 August 10, 1999
227/070 July 5, 1992 July 8, 1996 August 10, 1999
228/068 June 18, 1992 June 13, 1996 August 17, 1999
228/069 June 18, 1992 June 13, 1996 August 17, 1999
229/067 July 3, 1992 July 22, 1996 August 8, 1999
229/068 July 11, 1992 July 22, 1996 October 11, 1999
229/069 June 25, 1992 July 6, 1996 August 8, 1999
229/070 July 11, 1992 June 23, 1996 August 8, 1999
230/068 July 10, 1992 October 17, 1996 August 15, 1999
230/069 May 15, 1992 October 17, 1996 July 3, 1999
232/067 June 22, 1992 June 25, 1996 June 28, 2000
002/067 August 30, 1992 Augist 1, 1996 August 2, 1999

 

 

* Not mapped in 1992 and 1996
Appendix E]

143

 

Appendix E.2. Basin-wide results of selective logging detection for 1992, 1996, and

1999.

 

144

 

Logging Area Detected (Km2) Average (1996+1999/7)

 

 

 

 

Path/Row 1992 1996 1999 (Km2)
221/063 46.22 14.29 39.88 7.74
222/062 515.76 893.71 1,096.75 284.35
222/063 644.31 704.48 2,728.60 490.44
222/064(*) - - 20.24 6.75
223/062 684.67 2,018.67 3,554.09 796.1 1
223/063 515.25 1,553.23 2,928.84 640.30
223/064 27.43 26.41 223.63 35.72
223/065 26.60 15.93 60.50 10.92
223/066 73.80 20.92 39.34 8.61
224/062(*) - - 691.87 230.62
224/063 7.64 20.17 40.47 8.66
224/065 47.17 60.47 11 1.03 24.50
224/066 295.93 193.09 743.27 133.77
224/067 23.58 74.33 322.70 56.72
225/067 - - 15.53 5.18
225/068 - - 93.68 31.23
225/069 1.69 17.75 62.94 11.53
226/063 5.02 55.19 8.17 9.05
226/066(*) - - 35.61 11.87
226/067 17.53 26.37 368.89 56.47
226/068 1,245.05 1,756.77 4,502.92 894.24
226/069 873.82 1,157.57 3,318.36 639.42
227/062(*) - - 442.55 147.52
227/065(*) - - 111.26 37.09
227/067 4.41 13.08 161.06 24.88
227/068 303.67 325.41 1,279.07 229.21
227/069 366.50 448.17 866.38 187.79
227/070 2.35 0.63 13.96 2.08
228/068 140.13 206.31 725.38 133.10
228/069 20.46 85.94 382.69 66.95
229/067 5.69 72.27 244.69 45.28
229/068 6.66 34.93 438.46 67.63
229/069 4.99 6.50 17.71 3.46
229/070 23.82 28.94 55.75 12.10
230/068 12.21 81.52 103.38 26.41
230/069 19.29 93.01 86.85 25.69
232/067 10.32 28.53 124.98 21.93
002/067 7.84 29.52 23.94 7.64
5,979.80 10,064.10 26,085.40
Appendix E.2

* Not mapped in 1992 and 1996

145

 

w? gin... . _ _...
a “.r

 

Appendix E.3. Basin-wide spatial distribution of selective logging detected in 1992.

146

 

 

M.M— xenonn<

 

 

 

 

 

     

 

 

 
    

 

wanes. 8413 . _
mac—Q arcane; .,........,.. 990E0=¥ OONV 80 o 000
39.23.... 3%qu . Jaw «\fi «‘2 flu own own kw“ Jun Jun flow «\nu we: 33...
oszm;
0390 2o:
:3 I
as I Ecovcom

WCZCOOOP

 

no: 11
see 1
on: 1
mac 1
3° 1
no: 1

«no.1
96¢.

EOE = 2: «aw—Ea.—

Nmow ZON<E< 4<0m4 w..=N<m_m

147

 

Appendix E.4. Basin—wide spatial distribution of selective logging detected in 1996.

148

 

 

 

 

 

     

 

  
 

 

 

  

 

Lilli-Ll
4....— 5:83
.28 as. u, ......1. _ .
20... 9.5qu . 39050:! Dog. 80 o 000
2:28 8:3 ,. 5 «Na 8“ 4w“ mun emu km” an“ 9% 8w «8 «8 .58.:
.1. \ z 1 1 z 1 \ N
0233
ommEG 05.2
291 .
Sol .. .. ..1 .. scanner

 

as I wczcaooh.

no: I

5886 __ 2:. 38:3

woe. ZON<E< 4<Om4 m..=N<mm

149

 

 

Appendix E.5. Basin-wide spatial distribution of selective logging detected in 1999.

150

 

 

 

 

x28 8.. . H.

ago—a weave.-

BEEEQ «aw—ES x.

oszm;

y

A .

 

113.

x

 

 

so 1

~09 I

moo I.

3:1

no: 1

N09 1

E326 = 0:... «aw—Ea.—

mmo w

mascaoOF

mm.— 5.83.

8m

999:2? 00$. 80 o

FNN NNN ”Nu #NN anN emu th wNN «Nu on“ Now we: .56.":

xxxx

     

 

xxx

 

0320 Bus.

 

«Eat—.3.

 

ZON<E< 4<0m4 m.1=N<~._m

151

1111131'112111111111111111

93