|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
ASSESSING TRADE-OFFS IN MULTIPLE-OBJECTIVE TROPICAL FOREST
GEOFFREY MICHAEL LATE
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
Geoffrey Michael Blate
This document is dedicated to my father, who I wish I could have known, and to my
mother, who raised me with tremendous courage and unconditional love after his death.
It is also dedicated to my late mother-in-law, who very much wanted to celebrate its
completion with me.
I am deeply indebted to the many people who contributed to the successful
completion of this dissertation. Without their patience, support, and encouragement, it
would have been difficult to finish.
I especially thank my mentor and supervisory committee chair, Jack Putz for his
patience, guidance, encouragement, and critical reviews of my writing. Jack' s high
expectations of himself as a teacher, writer, and adviser inspired me to also strive for
excellence. In my early discussions with Jack about a suitable topic, he always
emphasized the philosophy in the PhD. In Jack' s view, scholarship, logic, and clarity
must be infused and brought to life through philosophical perspective. I thank Jack for
giving me the intellectual freedom and time to develop a dissertation proj ect that was
philosophically rich and that satisfied my desire to work on something with practical
application to tropical forest management. Finally, I thank Jack for his friendship and for
opening his home to me while I endeavored to complete my dissertation from a distance.
I would also like to thank my other supervisory committee members: Ben Bolker,
Kaoru Kitajima, and Dan Zarin. Their insights, encouragement, and patience were more
helpful to me than they know. I thank each of them for listening to ideas, reading the
updates I sent from Bolivia or Washington, D.C., and for their critical review of
dissertation drafts. Although Colin Chapman left my committee after saying farewell to
the University of Florida, I also thank him for his insights, perspectives, and
encouragement. Ultimately, the belief each of my committee members had in what I was
trying to accomplish made all the difference.
Although Paul Phillips was not officially part of my committee, he deserves
honorary status. I thank Paul for countless hours of programming and de-bugging help so
that I could eventually make SYMFOR do what I wanted it to do. I also am grateful for
his advice, patience, and encouragement during my many moments of programming
frustration, all of which helped me avoid despair. The printout of our trans-Atlantic
correspondence could easily constitute its own dissertation!
I also would like to thank Todd Fredericksen, Marielos Pefia-Claros and Lourens
Poorter for sharing their ecological insights and statistical advice. I especially thank
Todd and Marielos for ensuring that I always had logistical support and field assistance in
La Chonta. I thank each of them for their encouragement during each phase of my
proj ect, but especially during the analysis and writing of my dissertation. Finally, I
thank them for their companionship, friendship, sense of humor, and in the end, making
my experience in Bolivia enj oyable and profoundly memorable.
I also would like to thank Steve Mulkey for graciously participating in my defense
at the last minute, and for encouraging me from the moment I began my studies at the
University. I am grateful that he took time to share his ecological insights with me.
I am indebted to BOLFOR for supporting me during my time in Bolivia and to all
of the BOLFOR staff who welcomed me and provided various kinds of assistance. I
especially wish to thank the research team (Juan Carlos Licona, Joaquin Justiniano,
Alfredo Alarcon, Urbano Choque, Claudio, and Marisol Toledo) for their hard work and
dedication in collecting the permanent plot data, which I used to calibrate SYMFOR. I
also want to thank Daniel Nash for helping me improve my Spanish and for translating
my work, and Froilan Merlo for making sure I had a working computer. I also wish to
express my sincere appreciation to my principal field assistants (Eugenio Mercado, Pablo
Mercado, and Lucio Alvarez Negrete) without whose help I could not have conducted my
experiments. I wish to thank Astrid Nielsen for her invaluable assistance with the tedious
task of assessing fuel loads and for her help with the fire experiments. I also thank my
wife, Suj ata Ram, who braved the ticks, sand fleas, and mosquitoes to help me collect
canopy cover data. Finally, I wish to thank Tita Alvira; Sanne Arst; Fred; Tina; and Anja
Boltz; Eben Broadbent; Peter and Martha Cronkleton; Betty Flores; Nell Fredericksen;
Jose Carlos Herrera; Werner Krueger; Calvin and Kristen Ohlson-Kiehn; Claudia;
Juliana; and Juan Antonio Romero; and Pablo Solano for helping make my experience in
Bolivia productive, fun, and memorable.
As I was developing and completing this dissertation, many students and professors
at the University of Florida provided advice, criticism, and encouragement. I especially
thank Claudia Romero, Bil Grauel, Tita Alvira, Tova Spector, Stephanie Weinstein,
Kevin Gould, Hillary Cherry, Geoff Parks, Joseph Veldman, Morgan Varner, Susan
Moegenburg, Toben Galvin, Matt Palumbo, Scot Duncan, John Paul, April Randle, Nat
Seavy, Kelly Keefe, and Skya Rose.
I would also like to thank Anne Taylor in the Graduate School Editorial Office for
her thorough review of my dissertation, and for her timely responses to all my editorial
The process of earning my PhD was longer and more arduous than I anticipated at
the outset. My family and friends have often felt neglected and abandoned, but their
love, patience, and encouragement have always been a great source of emotional support
without which I could not have persevered. I sincerely thank them for their faith in me.
Finally, I wish to express my heartfelt gratitude to my wife, Suj ata Ram, who now
knows far more tropical biology than perhaps she could have imagined when she married
me. She has endured the long journey with grace, and her faith in me never faltered. She
gave me the freedom and space to pursue my dream, and for that I am grateful.
TABLE OF CONTENTS
ACKNOWLEDGMENT S ................. ................. iv.............
LI ST OF T ABLE S ................. ................. xi......... ...
LI ST OF FIGURE S .............. .................... xiv
AB STRAC T ................ .............. xvi
1 MODEST TRADE-OFFS BETWEEN TIMBER MANAGEMENT AND FIRE
SUSCEPTIBLITY IN A BOLIVIAN SEMI-DECIDUOUS FOREST ......................1
Introducti on ................. ...............1.................
Site Description .............. ...............3.....
M ethods .............. ...............5.....
D esign .................. .. ........ .... ... .... .. ... ..... ..........
Treatment Effects on Forest Structure and Vegetative Cover ............... .... ...........6
Treatment Effects on Fuel Loads .............. .... ...............7...
Vegetation Cover and Dry-Down Relationships ................. ........................9
Test Fires .............. ..... .... ..............1
Calculation of Fire-Prone Days ...._ ......_____ ......._ ............1
Re sults............__...... .. ..__ ......._ ... ...........1
Treatment Effects on Forest Structure ....__. ................. ........._.._.......1
Treatment Effects on Fuel Loads .............. ...............16....
Effect of Cover on Fuel Dry-Down Rates ....__. ................. ........._.._.. ..17
Fire Trials .............. ...............17....
Late dry season ............ ..... .._ ...............17...
Early dry season .............. .... ... ....... ..........1
Persistence of treatment impacts on fire susceptibility .............. .............19
Fire-Prone Days ................. ...............20.................
Discussion................. .. .. .. .. .......2
Treatment Effects on Forest Structure .................. ...............20........... ...
Treatment effects on forest structure and vegetative cover ............... .................20
Treatment Impacts on Fuel Loads .............. ...............22....
Vegetative Cover and Fuel Dry-Down Rates ........._.._.. ....._.._ ........._.....24
Fire Trials .............. .. ..... ....... .. .......2
Factors influencing fire susceptibility .............. .......... ...............2
Persistence of treatment impacts on fire susceptibility .............. .................27
Treatment impacts on number of fire-prone days .............. ....................28
Conclusions and Management Implications ......._ ......... ___ ........._ ......29
2 PROJECTING FOREST RESPONSES TO DIFFERENT SILVICULTURAL
TREATMENTS IN A SEMI-DECIDUOUS FOREST IN LOWLAND BOLIVIA..42
Introducti on ................. ...............42..._.__.......
M ethods .............. .. ...............45...
Site and Data Source .............. ...............46....
Study site ..........._..._ .. ..... ........ .. ........ .. ...........4
Experimental design and silvicultural treatments .............. ....................47
Model Description and Calibration .............. ...............48....
SY MFOR overview............... ...............48
Ecological model s .............. ...............48....
Management model s .............. ...............62....
M odel Evaluation .............. ...............63....
M odel tuning .............. ...............63....
Model testing .............. .. ....__ .... ......_ .. ..... .........6
Model sensitivity to parameters not calibrated from data ..........................65
Comparison of treatment impacts ............_...... ._ ... ......_........67
R e sults............. .. ........._ .... ...... ...............6
Model Testing: Comparing Test vs. Calibration Data............... ..................6
Growth rates .............. ...............68....
Recruitment .........._.... ...............68.__... ......
Forest structure comparison ............... ........ .. ...... .........6
Model Sensitivity to Parameters not Calibrated from Data............... .................70
Mortality of large trees: total tree density and basal area.............................70
Mortality of large trees: N and G by species group............... .................7
Mortality of large trees: maximum dbh............... ...............72..
Liana infestation............... ..............7
Damage related mortality ..............._ ...............72 ......... ....
S ilvi culture-related re critm ent ...._._._.. ..... ..__... ...._._._.........7
Summary .............. ..... .. ...............7
Comparison of Treatment Impacts .............. ...............74....
Harvest volumes .........._.... ......_ ._ ...............74.....
Effects on overall stand structure ................. ...............75......___. ...
Effects on species composition ............... ... ...... .. ........7
Silvicultural effects of the intensive management treatment ........._.._..........78
Discussion ................... .... ... ...... .... ........ .. .... .... ........7
Effectiveness of the Applied Management Treatments in Achieving STY ........79
Prospects for Achieving STY in La Chonta ................. .. ......... ... ............... 80
Management Treatment Impacts on Forest Structure and Composition .............83
Total tree density and basal area .............. ...............83....
Impacts at the species group level ....__ ......_____ ...... ......_........8
Reliability of the Model's Proj sections ............... ....... ..... _...... ....._.......8
Sensitivity of the Model to Parameters Based on Little or No Data .................87
3 TIMBER PRODUCTION, BIODIVERSITY CONSERVATION, CARBON
SEQUESTRATION, AND WILDFIRE TRADE-OFFS IN TROPICAL
FORESTS .............. ...............128....
Introducti on ............... .. .. ....... .......... ............12
Trade-offs in Tropical Forest Management ................. ......... ................1 29
Focus on Timber, Carbon, Fire, and Biodiversity ................. .. .....................133
The Silvicultural Challenge of Achieving STY in Tropical Forests .................13 5
A pproach............... .. ..............13
General Approach............... ...... ....... ........3
Specific Approach: the Case of La Chonta ........._.._.. ......._ ........._......139
Site description .........._.......... ..._.._ ...............139...
Management impacts on timber volumes ........._._ ...... .. ..............140
Management impacts on biodiversity .............. ...............143....
Management impacts on carb on sequestrati on ................. ............... .....1 47
Management impacts on fire susceptibility and intensity ..........................152
Application to other Tropical Forests ................ ...............154........... ...
Eastern Amazon. ................. .. ......... ...............155......
Southeast Asian dipterocarp forests .............. ...............160....
Summary of Trade-offs in Three Tropical Forests ................. ........................162
Policy Options to Minimize Trade-offs ................. .........___....... 164.........
A VARIABLES AND PARAMETERS USED IN CHAPTER 2 ............... .... .........._.177
B ANNUALLY UPDATING THE LIANA CLASS OF EXISTING TREES IN
SY M F OR ............ ..... ._ ...............179...
LI ST OF REFERENCE S ............ ..... ._ ..............1 1....
BIOGRAPHICAL SKETCH .............. ...............201....
LIST OF TABLES
1-1 Impacts of timber harvesting and silvicultural treatments in Block 2 of the
Long-term Silvicultural Research Proj ect in La Chonta..........._._. ........._._.....3 1
1-2. Percentage of each treatment plot in Block 2 of La Chonta observed in
corresponding habitat classes based on point sampling along four 400-450 m
transects in each treatment plot 6 months post-harvest............... ..............3
1-3 Summary of assessment of treatment impacts on forest structure showing means
a standard errors for estimates of vegetative cover (%) in 6 vertical strata, total
cover, maximum height, and horizontal distance to nearest gap .............................32
1-4 Mean quantity of woody debris encountered per meter of transect in La Chonta ...33
1-5 Mean densities and mass estimates for each size and decay class of woody
debris in La Chonta .............. ...............34....
1-6 Summary of 7-9 October 2001 (late dry season) test fires in La Chonta showing
means a standard error for measured variables for each of the three consecutive
days of the trial and the three-day averages ........._._.._ ....... ........_.._......35
1-7 Path coefficients, P-values, and proportion of variance explained by factors
influencing area burned in October 2001 test fires in La Chonta determined by
path analysis .............. ...............36....
1-8 Summary of August 2002 fire trial in La Chonta showing means, standard
errors, and ANOVA or Kruskal-Wallis results for total cover, vapor pressure
deficit (VPD), average litter depth, 10-h fuel moisture content, and area burned
of 4 m2 plOts by year post-logging .............. ...............37....
2-1 Species groups obtained in the cluster and discriminant analyses ................... ........89
2-2 Estimated parameter values for diameter increment in La Chonta obtained by
non-linear regression with growth modeled as a function of dbh and liana class
for each species group ................ ...............90................
2-3 Estimated ingrowth parameters and associated R2 ValUeS obtained from
weighted non-linear regression using Eq. 2-6 ................. ................ ......... .91
2-4 Estimated mortality parameters obtained from logistic regression using Eq. 2-7 ...91
2-5 Parameter values for the mortality constant of large, liana-free and liana-infested
trees of each species group in La Chonta ................. ...............92..............
2-6 Liana infestation probabilities used for assigning a liana class to new recruits........92
2-7 Assigned probability of simulated gaps becoming liana infested depending on
the liana class (L) of the falling tree and the proportion of neighbors with lianas...92
2-8 Dbh and total tree height allometric relationship parameter estimates and
associated R2 ValUeS for each species group .............. ...............93....
2-9 Parameter values used to calculate the crown-point (a) for each species group......93
2-10 Mean wood density values for each species group and the number of species on
which each mean is based .............. ...............93....
2-11 Utility group assignment for commercial and potentially commercial species in
La Chonta ................. ...............94_ ......
2-12 Proportional representation of each utility group in each species group in
La Chonta ........... ..... .. ...............95...
2-13 Scarification factor (Es) estimates for adjusting the probability of recruitment of
trees of each species group in scarified felling gaps .............. ....................9
2-14 Parameters and values tested in sensitivity analysis .............. ....................9
2-15 Per capital recruitment for each species group in the calibration and test datasets
from La Chonta .............. ...............96....
2-16 Mean tree density and basal area by species group and as a proportion (%) of the
totals in the calibration and test datasets .............. ...............97....
2-17 Sensitivity of tree density and basal area of each species group to changes in
supplemental mortality of large trees ................ ...............97........... ...
2-18 Simulated harvest results from the normal harvest treatment by species and
utility group at years 0, 30, and 60 ................. ...............98..............
2-19 Simulated harvest results from the intensive management treatment by species
and utility group at years 0, 30, and 60 .............. ...............98....
2-20 Harvest volumes obtained (total and by commercial class) in years 0, 30, and 60
from simulations of the normal harvest and intensive management treatments in
La Chonta ................ ..............99. ...............
2-21 Relative contribution of silviculture and new species harvested to the increase in
harvest volumes obtained with the simulated intensive management treatment
compared to the normal harvest treatment ................. ........._.._ ...... 99__.. ...
2-22 Silvicultural liberation of future crop trees (FCTs) from lianas and neighboring
competitors (by poison-girdling), and the proportion of felling gaps that were
scarified in each cutting cycle in simulations of the intensive management
treatment in La Chonta ........._._.._......_.. ...............100...
3-1 Some silvicultural impacts on different attributes of the community component
of biodiversity in La Chonta ................. ...............167........... ...
3-2 Factors used to estimate aboveground biomass components based on the
proportion of total aboveground biomass (TAGB) they comprise according to
several recent studies............... ...............169
3-3 Simulated management treatment effects on carbon flux over 60 years in
La Chonta ................ ...............170................
A-1 Variables, parameters, and the equations or tables in which they are used in the
SYMFOR model described in Chapter 2 .............. ...............177....
LIST OF FIGURES
1-1 Treatment impacts on litter depth during the rainy and dry seasons in La Chonta..38
1-2 Effect of vegetative cover on understory microclimate and fuel dry-down rates
in La Chonta during the early dry season of 2002 ......__ ......... __. ...............39
1-3 Path diagram for the effects of total cover, vapor pressure deficit, litter depth,
wind, 10-h fuel moisture content (10-h MC), and 1-h fuel moisture content
(1-h MC) on area burned (%) in the 7-9 October 2001 test fires in La Chonta.......40
1-4 The number of fire-prone days per month predicted for each management
treatment applied in La Chonta .....__.....___ ..........__ ...........4
2-1 SYMFOR model overview and flow .............. ...............101....
2-2 Dbh growth patterns for 8 species groups in La Chonta by liana class revealed
by non-linear regression analysis .....__.....___ .........._ .............0
2-3 Probabilities of ingrowth calibrated by non-linear regression as a function of
predicted growth rate for 10 cm recruits of the 8 species groups defined in
La Chonta ........._ ...... .. ...............107...
2-4 Probabilities of annual mortality calibrated by logistic regression as a function
of dbh and liana class for the 8 species groups defined in La Chonta .........._......11 1
2-5 Sensitivity of total tree density and basal area to changes in mortality rates of
large trees ........._ ...... .. ...............115...
2-6 Examples of changes in tree density and basal area for species groups that were
sensitive and insensitive to changes in mortality rates of large trees .........._........116
2-7 Harvest volumes obtained in years 0, 30, and 60 from the normal harvest and
intensive management treatments in La Chonta based on simulations of
15 1-ha plots with 15 repetitions each. ....._._._ .... ... .__ ......._._......17
2-8 Simulated management treatment impacts on tree density and basal area by
commercial class in La Chonta ........._._......._. ....___ ............1
2-9 Diameter growth rates by commercial class observed in 62-year simulations of
management treatments in La Chonta ......___ ............. ......__...........2
2-10 Changes in total tree density and basal area observed in 62-year simulations of
management treatments in La Chonta ......___ ............. ......__...........2
2-11 Species group-specific changes in tree density and basal area observed in
62-year simulations of management treatments in La Chonta .............. ..............123
2-12 Changes in the proportion of trees with lianas observed in 62-year simulations of
management treatments in La Chonta ....__ ......_____ ...... ..__ ...........2
3-1 Production possibilities frontiers showing several types of trade-offs resulting
from j oint production of timber, biodiversity, and carbon ................. ................. 171
3-2 Timber volumes obtained and change in total carbon pools in response to three
different management treatments applied in La Chonta .............. ...................172
3-3 Production possibilities for timber and carbon based on forest management
options in La Chonta simulated with SYMFOR for 60 years ............... .... ........._..173
3-4 Management treatment impacts on 1,000-h fuel loads and number of fire prone
days in La Chonta............... ...............174
3-5 A conceptual model showing two extreme pathways and corresponding trade-
offs resulting from a strategy to conserve tropical production forests through
sustainable timber production (STY) with or without payments for ecosystem
services (PES) .............. ...............175....
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
ASSESSING TRADE-OFFS IN MULTIPLE-OBJECTIVE TROPICAL FOREST
Geoffrey Michael Blate
Chair: Francis E. Putz
Major Department: Botany
While tropical forests continue to be cleared at alarming rates, the debate over how
best to conserve them often proceeds without a clear understanding of the trade-offs that
result from different management alternatives. Because timber is the most valuable
product of tropical forests, substantial effort has been directed to harmonizing timber
production and other goals, especially carbon sequestration and biodiversity
conservation. Unfortunately, the nature and magnitude of the trade-offs between timber
production and these objectives remain unclear and depend on numerous factors. By
elucidating the key biophysical factors that influence forest management trade-offs, I
aimed to better inform the quest to manage tropical forests for multiple benefits.
I assessed the trade-offs between timber production and fire susceptibility in a
seasonally dry forest in lowland Bolivia subj ected to four silvicultural treatments of
increasing intensity aimed at achieving sustained timber yields (STY). By quantifying
treatment effects on fuel loads, vegetative cover, dry-down rates of 10-h fuels, and fire
spread, I found that the treatments had little effect on fire susceptibility; this forest is
fire-prone for about 130 days per year, even in the absence of logging. Fire severity,
however, would likely be greater with intensive management due to increases in 1,000-h
Using a simulation model (SYMFOR) to proj ect the effects of the silvicultural
treatments on future timber yields, forest structure and composition, and biomass, I found
that none of the treatments came close to achieving STY, indicating that silviculture
would probably need to be intensified to secure STY. Neither forest structure nor species
composition changed appreciably over two cutting cycles (60 years) in any of the
By contrasting the Bolivian forest with eastern Amazon and Borneo forests, I found
that the trade-offs resulting from intensive silviculture to secure STY paled in comparison
to the loss of forests. If maintenance of productive forest for both timber and carbon with
a full complement of biodiversity is the goal, then fire (and not silviculture) is the
menace. Fire prevention must complement silvicultural treatments to achieve sustained
yields if tropical forests are to serve both production and conservation goals.
MODEST TRADE-OFFS BETWEEN TIMBER MANAGEMENT AND FIRE
SUSCEPTIBILITY IN A BOLIVIAN SEMI-DECIDUOUS FOREST
Forest managers have long recognized that not all forest uses or obj ectives can be
simultaneously maximized (Dana 1943, Toman and Ashton 1996). Despite this
realization, sustainable forest management (which promises continuous delivery and
maintenance of multiple goods, services, and processes) has become the predominant
management paradigm in tropical countries. Although research into the compatibility
among different obj ectives of forest management has advanced sub stantially in temperate
forests (Johnson et al. 2002, Stevens and Montgomery 2002), such research is scarce in
tropical forests. Elucidating management trade-offs between sustained timber production
and fire susceptibility is especially important considering the large proportion of tropical
forests designated for timber production, the potential for such forests to contribute to
conservation and development obj ectives, and the increasing prevalence of wildfire in
Despite increasing attention to the problem of fires in tropical forests (Goldammer
1990, Nepstad et al. 2001, Cochrane 2003) and the recognized role of logging in
exacerbating the fire problem (Woods 1989, Holdsworth and Uhl 1997, Nepstad et al.
1999a, Cochrane 2003) especially during El Nifio-related droughts (Siegert et al. 2001),
few studies have determined the extent to which increased fire susceptibility is an
inevitable consequence of intensifying management to achieve silvicultural obj ectives.
Holdsworth and Uhl (1997) showed that fire susceptibility (i.e., likelihood of fires
starting and spreading in a forest) decreases when logging impacts are reduced in eastern
Amazonian evergreen forests, where deep roots help most trees retain their foliage during
the marked dry season (Nepstad et al. 1995). Unfortunately, reducing logging damage
will not sustain timber yields, let alone achieve sustainable forest management, in
seasonally dry, semi-deciduous forests. These forests typically require post-harvest
silvicultural treatments (e.g., liana cutting, felling gap enlargement, and soil scarification)
to promote regeneration and growth of commercial tree species, many of which are shade
intolerant and lack adequate regeneration (Mostacedo and Fredericksen 1999,
Fredericksen and Mostacedo 2000, Fredericksen et al. 2003). Considering that seasonally
dry tropical forests encompass more area and are more affected by people than wet
forests (Mooney et al. 1995), it is important to understand whether the intensive
management regimes thought necessary to sustain commercial timber species and hence
prevent conversion to other land uses, might elevate fire susceptibility and inadvertently
promote forest conversion. My study was conducted in a region of the Bolivian Amazon
where forest management is hoped to provide an economically viable alternative to
conversion of forests to pasture land, which has increased dramatically in recent decades
(Steininger et al. 2001).
Studies throughout the tropics show that logging increases forest susceptibility to
fire (Kauffman and Uhl 1990, Siegert et al. 2001, Cochrane 2003) as well as fire severity
(i.e., fire behavior and its ecological effects; Kauffman 1991) because logging results in a
drier understory and increased fuel loads (Uhl and Kauffman 1990). Because additional
silvicultural treatments beyond harvesting probably exacerbate the factors that drive fire
susceptibility and influence fire severity, it seems reasonable to expect that intensively
managed forests would be more prone to wildfires of greater severity than undisturbed or
less-intensively managed forests. Moreover, because more radiation reaches the forest
understory in deciduous forests, they are more likely to be fire-prone than are wetter,
evergreen forests (Swaine 1992).
The purpose of my study was to elucidate the nature and extent of tradeoffs
between sustaining commercial timber yields and increasing fire susceptibility in a
semi-deciduous tropical forest, and to contrast these findings with published studies from
moister sites. I expected that more intensive management would
* Increase the proportion of the forest that is fire-prone on any given day during the
* Extend the number of fire-prone days during the dry season;
* Increase the potential for severe or catastrophic fires; and,
* Increase the number of months or years of elevated fire susceptibility above
background levels (i.e., compared to un-logged control areas).
My study was conducted in a 100,000 ha Forest Stewardship Council-certified
timber concession operated by Agroindustria Forestal La Chonta Ltda. in Guarayos
Forest Reserve (15045'S, 62060'W) in the Bolivian lowlands (200-400 m.a.s.1.).
According to the Holdridge classification system, the concession (hereafter La Chonta) is
covered by subtropical humid forest. Seasonally deciduous and semi-deciduous forests
like La Chonta provide about 45% of Bolivia' s timber, and encompass about 35% of its
designated forest management area (Superintendencia Forestal 2001). Biomass estimates
for the region are 73-190 Mg/ha (Dauber et al. 2000). Mean annual temperature is
~24.50C and mean annual rainfall is ~1,500 mm, 77% of which falls between November
and April. During the peak of the fire season (July-September) in average years, only
43 mm of rain falls monthly and understory vapor pressure deficits range from 0.6 to
0.8 kPa. Soils are moderately fertile inceptisols, but 10 to 15% of the area has black
anthrosols enriched by humans several hundred years ago (Paz 2003).
The fire history of the region is unknown, but evidence that people extensively
inhabited much of La Chonta and the presence of charcoal in the subsoil (Paz 2003) both
suggest that the forest was historically subj ected to fire. In 1995, an escaped fire burned
about 30% of the La Chonta concession (Pinard et al. 1999, Mostacedo et al. 2001, Gould
et al. 2002), killing 23% of the trees (dbh > 10 cm) and 75% of the lianas, and causing a
proliferation of herbaceous vines (Pinard et al. 1999). Commercial tree regeneration
remained scarce 5 years after the fire (Gould et al. 2002). Another fire in 1999 nearly
destroyed the town on La Chonta' s southwest border, but did not otherwise affect the
concession. These fires occurred between July and September, when people in the
fragmented matrix of fire-maintained, anthropogenic palm savannas and active
agricultural fields surrounding La Chonta set fires to clear woody vegetation.
La Chonta is situated in a transitional zone between wetter forests to the north and
drier forests to the south and southeast, and is dominated by canopy tree species
characteristic of humid forests including Ficus boliviana, Hura cr titans, and
Pseudolmedia laevis. Tree s ecies common in drier forests (g., Centrolobium
microchaete, Chorisia speciosa, and Cedrela fissilis) are also present in La Chonta.
Although only a few high-value timber species (e.g., Swietenia macrophylla and
C. fissilis) were harvested between 1980 and the mid-1990s, 10-12 tree s ecies were
harvested during my study (2001 to 2003). About 7-20 m3/ha of wood (3-5 trees/ha)
were harvested from annual management units that each encompassed ~2,300 ha.
Harvest activities are planned based on a 30-year cutting cycle and implemented in
accordance with Forest Stewardship Council certification standards and criteria.
Minimum diameter for felling set by law is 70 cm for H. cre itans and F. boliviana, and
50 cm for all other species. Approximately 20% of trees above the diameter limit are left
as seed trees.
My study was conducted in and near 27-ha permanent plots established as part of a
long-term silvicultural research project (LTSRP, IBIF 2004). The LTSRP applied four
treatments representing a range of management options and intensities: "control" = no
logging; "normal logging" = planned logging with no other silviculture; "improved
management" = "normal logging" with liberation of future-crop trees from vines and
overtopping non-commercial trees; and, "intensive management" = "improved
management" with double the harvest intensity, additional future-crop tree liberation, and
soil scarification in selected felling gaps. The improved and intensive treatments aimed
to promote the regeneration and growth of commercial timber species, most of which are
light-demanding (Mostacedo and Fredericksen 1999, Pariona et al. 2003). Treatments
were randomly applied to 27 ha plots in each block, which were situated in three different
harvest units (3 blocks x 4 treatments = 12 plots).
My study was conducted in and near Blocks 2 (harvested between May and July
2001) and 3 (harvested between February and July 2002). Because roads can act as fire
conduits (Dell 1970, Wilson 1979, Schwartz and Caro 2003), it is important to note that
although skid trails were included, roads did not traverse the treatment plots. Although
the forest appeared disturbed, signs of previous fire were only evident in a few small
patches of the intensive plot and no old stumps from previous logging entries were found
in any of the plots.
Treatment Effects on Forest Structure and Vegetative Cover
To compare harvest treatment impacts on forest structure, I compared the harvest
volumes and basal area removed, as well as the corresponding ground and crown area
disturbed by harvest operations. Residual tree densities and basal areas were also
compared, based on post-harvest censuses of the 3 harvest treatment plots and the control
plot. To estimate the impacts of each treatment on vegetative cover, I measured cover at
5 m intervals along four transects (spaced 75 m apart) per treatment previously
established in Block 2. The total number of sample points was 327 for the intensive
treatment, 363 for the improved treatment, 338 for the normal treatment, and 377 for the
control treatment. The transect lengths varied between 400 and 500 m depending on the
distance from the southern to the northern border of each treatment plot. All cover
estimates were made in early December 2001, by which time deciduous trees were in full
I estimated the percentage of vegetative cover in six vertical strata (0-1 m; 1-2 m;
2-4 m; 4-8 m; 8-16 m; and >16 m) by viewing upward through a clear grid of
twenty-five 3 cm x 3 cm squares (Mostacedo and Fredericksen 2000). I counted the
number of squares covered and half-covered with vegetation in each stratum. To increase
the accuracy of ocular estimates of vertical heights, a 14 m telescopic pole was used for
daily calibration. Where total cover is reported, it is shown as a percentage and is simply
the sum of the six percentages of cover (one per stratum).
At each sample point, I also estimated the horizontal distance to the nearest gap
(felling or natural) and classified each point by habitat type. I defined a gap as any area
> 10 m2 in which the hi hest ve etation was < 2 m tall (rokaw 1982 Habitat classes
* Felling gap;
* Felling gap edge (0-20 m from gap);
* Skid trail;
* Skid trail edge (0-20 m from edge);
* Natural gap; and,
* Natural gap edge (0-20 m from gap).
Means from each transect per treatment were used to test for treatment differences
in a one-way ANOVA; separate tests were conducted for each cover stratum, total cover,
and distance to gap. Because the 4 transects per treatment were all located in one block,
they do not constitute true replicates. Nevertheless, inferential statistics were used to
indicate whether cover differed by treatment (Oksanen 2001).
Treatment Effects on Fuel Loads
To assess treatment impacts on fuel loads, censuses were conducted 6 and
15 months post-harvest in Block 2 using the planar transect method (van Wagner 1968,
Brown 1974, Uhl and Kauffman 1990). In each census, 36-47 randomly oriented
transects were established starting at 50 m intervals in each treatment plot. Transects
consisted of vertical planes (extending from the ground to 2.5 m) of variable length
depending on the diameter of the woody debris, with planes 11 m long for 1,000-h fuels
(> 7.5 cm), 5 m long for 100-h fuels (2.5-7.5 cm), 2 m long for 10-h fuels (0.6-2.5 cm),
and 1 m long for 1-h fuels (< 0.6 cm). The fuel time-lag concept is based on observations
that as relative humidity changes, fuel moisture changes in an exponential fashion; and
that smaller diameter fuels gain or lose moisture faster than larger diameter fuels, because
of their higher surface area to volume ratio (Agee 1993).
Wood fragments of each size class were tallied if they crossed the sample plane.
Separate tallies for 1,000-h fuels were made according to three decay classes (sound,
intermediate, and rotten) described by Delaney et al. (1998). Calipers were used to
measure diameters of all 1,000-h fuels and a sample of smaller woody debris to obtain
mean diameters for those size classes. Leaf litter depth was measured at three points
along each transect.
To obtain fuel-mass estimates, samples of litter and woody debris were collected
from the transects. Litter samples (all 1-h fuels down to mineral soil) from quadrats
(20 x 20 cm) located at the beginning of each transect were oven-dried at 800C to
constant weight. The first three pieces of 1-, 10-, and 100-h fuels were collected from
each transect to determine average wood density for each size class. Random samples of
larger pieces in each decay class were obtained elsewhere in each treatment plot.
Wood densities were calculated from measurements of the fresh volume (by water
displacement) and the oven dry mass of each sample. Heartwood, sapwood, and bark
were included in cases where large sections of trunk measured. Masses for each size
class were computed on a tons per hectare basis as per methods described by Brown
(1974) using the combined data from all transects within each treatment. To compare
differences among treatments, the log-transformed values for each transect in each
treatment were used in an ANOVA. Although the transects in each treatment did not
constitute true replicates, the only way to obj ectively compare the effect of the treatments
on fuel loads was to use inferential statistics. Treatment impacts on leaf litter depth and
litter mass were compared in a similar manner.
Vegetation Cover and Dry-Down Relationships
I determined the number of days needed to dry 10-h fuels to 12% moisture content
under a range of canopy-cover conditions during the early and mid-dry season. Uhl and
Kauffman (1990) considered 12% to be the threshold moisture content below which
forest fuels could easily ignite. From 10-19 July 2001, I measured dry-down rates of 10-h
fuel sticks made from a local species (H_ crepitans), but otherwise identical to the
standard pine fuel sticks (four 35-cm x 1-cm diameter dowels connected in a plane by
small dowels and staples) used by other researchers (Uhl and Kauffman 1990,
Holdsworth and Uhl 1997).
I placed 10 fuel sticks along each of 5 variable-length transects (50-75 m) that
originated in the centers of logging gaps (50-250 m apart) and extended into undisturbed
patches of forest in the intensive treatment plot of Block 2. After soaking overnight in
water to simulate a substantial rainstorm, the fuel sticks were suspended ~25 cm above
the forest floor, in a stratified random manner to represent the range of cover conditions
present throughout the four management treatments; sticks within a transect were at least
5 m apart. I repeated this experiment in Block 3, in late May 2002, using standard pine
sticks to determine the number of days necessary for 10-h fuels to reach 12% moisture
content in the early dr season. For anal ses, the moisture contents of the H. cre itans
fuel sticks were adjusted to values for standard pine fuel sticks, based on regression of
average moisture contents of both species during controlled drying (R2 = 0.98).
Data collected included vegetative cover estimates, temperature, relative humidity,
and moisture content of the fuel sticks. I estimated cover in six vertical strata above each
fuel stick, using the grid method described above. I recorded temperature and relative
humidity at 2 hour intervals for 2 weeks in 8 of the sites, using Hobo@ (Onset Computer
Corporation, Bourne, MA) temperature / relative humidity data loggers. I calculated
vapor pressure deficit as a function of mean maximum temperature and mean minimum
relative humidity at 1200 h using standard conversions (Rosenberg et al. 1983). I
weighed the fuel sticks to the nearest 0. 1 g daily, between 1200 and 1400 h, until they
dried below 12% moisture content.
Data were analyzed using non-linear regression, with vegetative cover as the
independent variable and number of days for the fuel sticks to reach 12% moisture
content as the response variable. Linear regression was used to relate total cover to mean
minimum relative humidity and mean maximum temperature.
To further test the influence of treatment-induced changes in micro-environmental
conditions on the forest' s susceptibility to fire, I carried out a series of test fires in 4 m2
square plots located just outside of the LTSRP plots. In the first experiment, I set 99 test
fires during three consecutive days in early October 2001 (late dry season). Although
4 mm of rain fell 2 days before the trials, before that event no rain fell for 1 week. Test
fire plots were located across the full range of cover conditions found in the treatment
plots. All plots were within 25-300 m of the principal logging road and within 5-50 m of
primary skid trails, to facilitate water transport and for safety.
Using methods adapted from U.S. Forest Service fire research in Brazil (personal
communication, D. Sandberg and E. Alvarado, August 2001), I established 33 replicate
sites for conducting experimental fires. A replicate consisted of a center point in which I
placed a standard 10-h pine fuel stick and a litter sample in a nylon mesh bag for
measuring moisture content surrounded by three test fire plots each separated by 1-2 m.
I also placed a Hobo data logger at 6 of the sites to record temperature and humidity over
the range of cover conditions among all sites. Test fire replicates were at least 25 m apart
and located to avoid steep slopes, dense Heliconia patches, and major vine tangles.
Before attempting to burn the plots, I estimated cover in 6 vertical strata using the
grid method described above from the center of each of the 99 test fire plots. To control
for variation in the presence of 100-h and 1,000-h fuels, I removed woody debris
> 2.5 cm diameter from several test plots. This step also helped increase my confidence
that the results were due to microclimatic factors and the quantity and moisture content of
1-h and 10-h fuels. To assess the quantity of 1-h fuels, I measured litter depth in ten
locations (corners, perimeter midpoints, and two points near the plot center) to the nearest
mm. I also collected a 20 cm x 20 cm litter sample from the center of each site using the
method described above to assess 1-h fuel moisture content. Finally, the percentage of
live vegetation below 1 m height (including ferns as well as herbaceous and woody
vegetation) covering each sample plot was estimated.
To prevent fires from escaping beyond the plots, a 50 cm safety buffer was
established around their perimeter 2-3 hours before starting the test fires. Specifically,
all leaf litter and larger fuels were removed so that any fire reaching the plot edge would
encounter mineral soil. Thus, all test fires eventually self-extinguished within the plot' s
I started the test fires at 1200-1430 h each day by igniting 50 ml of diesel that was
dripped over a small area (400 cm2) in the center of each plot. As the fires were set, the
fuel sticks and litterbags were weighed, the temperature and relative humidity were
measured with a digital max/min thermohygrometer (Thermo-Hygro, Control Company,
Friendswood, Texas, USA), and wind speed (m/s) was estimated by measuring the
distance a feather dropped from 2 m flew and the time it took to land. Vapor pressure
deficit was calculated as described above. The amount of time flames were visible in
each plot was recorded. After all the plots burned, I measured the maximum distance
(0 to 90 cm) fire carried from the ignition zone in each plot and visually estimated the
percent of each plot that burned. I repeated this experiment in June and July 2002 (in an
area adj acent to Block 3 harvested 1-3 months previously) to quantify the extent to which
intensifying management extends the fire-prone season.
After transforming percentage data (arcsine square root), the effects of total cover,
litter depth, vapor pressure deficit, wind, and 10-h and 1-h fuel moisture content on the
percent area of the plots that burned were evaluated with path analysis using SAS version
9.0 (SAS Institute, Inc., Cary, North Carolina) and procedures described by (Mitchell
2001). Path analysis allows hypothesized causal relationships (both direct and indirect)
among independent and dependent variables to be tested via a series of multiple
regressions (Schemske and Horvitz 1988, Mitchell 2001).
To quantify the persistence of any elevation in fire susceptibility due to
intensification of management, I conducted an additional trial to compare the ability of
fire to carry in forest logged at 3 different times. This trial, conducted in August 2002,
was carried out near Block 2, near Block 3, and in an area logged in 1999. Block 3 had
just been logged and hence was termed the 0-year treatment. Block 2 was logged 1 year
previously (1-year) and the 1999 area was logged 3 years previously (3-year). Two
research teams burned 10 plots per day at 1200-1430 h in each treatment. In contrast to
the criteria described above for selecting plot locations, I placed plots in the most
severely disturbed sites (i.e., large felling gaps and primary skid trails) I could find in
each of the three areas. After arcsine square-root transformation for percentage data,
one-way ANOVA was used to compare the means of total cover, litter depth, 10-h fuel
moisture, vapor pressure deficit, and plot area burned among the 3 treatment blocks. In
the two cases with unequal variances among treatments, a Kruskal-Wallis test was used
(test statistic, H). Post-hoc comparisons (Tukey or Mann-Whitney U) were conducted to
specify which treatments differed.
Calculation of Fire-Prone Days
A simple model was used to calculate the number of days La Chonta is prone to
fire in response to each management treatment in an average year and a dry year. The
model is based on the assumption that the number of consecutive rainless days is a good
predictor of fire susceptibility. It also depends on the relationship between canopy cover
and the dry-down rate of 10-h fuels that was experimentally derived (see above).
The number of fire-prone days per month (F) was calculated as a function of fo,
fire-prone days for forest patches with very sparse vegetative cover (=open; total cover
<110%); fm, fire-prone days for patches with intermediate cover (=mid; total cover 110 to
210%); and fe, fire-prone days for patches with dense cover (=closed; total cover >210%).
F =fo*" Co +fm* Cm +f,* Cc
where Co, Cs, and Co represent the proportion of each treatment plot consisting of
"open," "mid," and "closed" vegetative cover, respectively. Values for C were obtained
from the assessment of treatment impacts on cover. The model weights the number of
fire-prone days for open, intermediate, and closed canopy conditions by the proportion of
the forest in each condition.
Fire-prone days for each cover class were calculated with the following procedure.
First, I examined the daily rainfall records for La Chonta, which were available for 1998
to 2002, and defined a rainfall event as being a one-day measurement of at least 5 mm of
precipitation, which included 98% of the rainfall records. Second, I counted the number
of consecutive rainless days required for 10-h fuels to dry sufficiently to ignite for open,
intermediate, and closed forest patches. Based on the results of the cover and dry-down
experiments (described above), I assumed that 10-h fuels become flammable after 3
consecutive rainless days in open forest patches, after 6 consecutive rainless days in
intermediate forest patches, and after 9 consecutive rainless days in closed forest patches.
For example, if in a particular month two rain events were separated by 10 consecutive
rainless days, open patches would be fire-prone for 7 days, intermediate patches for
4 days and closed patches for 1 day. The monthly tallies of fire-prone days (based on
consecutive rainless days) for open, intermediate and closed forest were used as values
for fo,fm, and fe, respectively. This equation was used to calculate F on a monthly basis
for each treatment for the 5-year period 1998-2002. The results for each treatment were
compared by visual inspection of box plots of these values and of summary statistics.
Because the inter-annual, within treatment variation greatly exceeded variation among
treatments, treatment effects were not compared with inferential statistics.
Treatment Effects on Forest Structure
Between 35 and 44% more trees were harvested per ha from the intensive treatment
plot than from the other two harvest treatment plots (Table 1-1). The corresponding basal
area and volume harvested from the intensive plot was also greater than the other
treatments, but the difference between the normal logging and intensive treatments was
smaller than expected due to the greater average diameter of harvest trees in the normal
treatment than in the intensive treatment plot (Table 1-1). The intensive treatment killed
41% more trees (> 10 cm dbh) than the i proved treatment and 33% more than the
normal treatment (Table 1-1).
Treatment impacts on forest structure were relatively modest. Six months
following treatment, the harvest treatment plots had only 10-25% more gaps and
building-phase forest (vegetation < 8 m) than the control plot. Similarly, the proportion
of the plots with mature forest (vegetation > 16 m) was 20% less in the harvest treatments
than in the control treatment, but only 30% of the control plot comprised mature forest.
Average horizontal distances to gaps were shorter in the intensive management than in
the improved harvesting and control treatments (F3, 12 = 6.3, P < 0.01), but were similar in
the intensive management and normal harvest treatments. The area disturbed by felling
and skid trails was similar among the harvest treatments (Table 1-2). The harvest
treatment plots retained less cover than the control treatment plot in the 8-16 m stratum
(F3, 12 = 8.2, P < 0.01), but cover in this stratum was similar among the harvest
treatments. When all cover strata were combined (maximum cover = 600%), only the
normal harvest and intensive management treatments retained less cover than the control
(F3, 12 = 7.7, P < 0.01); total cover was similar in the intensive management and normal
harvest treatments (Table 1-3). Only 6% of the intensive treatment plot comprised
patches with sparse cover (< 110%).
Treatment Effects on Fuel Loads
Based on 3,575 m (325 transects) inventoried for woody debris 6- and 15-months
post-harvest, the treatment plots differed most notably in the quantity of coarse woody
debris encountered. The intensive management plot contained twice the quantity of
1,000-h fuels as the normal and improved harvest plots and 25 times the quantity as the
control plot. In addition, about 60% more 100-h fuels were encountered in the intensive
management plot than in the control plot and about 30% more were tallied in the
intensive management plot compared to the normal and improved harvest plots (Table
1-4). Similarly, the variation in spatial distribution of fuels increased with treatment
intensity. Compared to the control treatment, the intensive management treatment
increased the variation in the quantity of 1-h fuels by 58%, 100-h fuels by 101%, and
1,000-h sound fuels by 121% (Table 1-4).
Because the results of the 6-month and 15-month post-harvest woody debris
assessments were similar (Fl, 317 = 1.3; P =0.27), the data from both were pooled. The
assessments revealed that (i) greater fuel loads were present in the harvest treatments than
in the control treatment (Table 1-5), and (ii) total mass of woody debris (all size and
decay classes) was greatest in the intensive treatment (Table 1-5), but differences among
harvest treatments were not significant. The greater mass of sound 1,000-h fuels in the
harvest (especially the intensive) treatments compared to the control treatment accounted
for most of the difference in mass of woody debris.
The harvest treatments reduced leaf litter depths (H = 31.0, P < 0.01, Figure 1-la)
relative to the control treatment six months after logging (during the early rainy season),
but among harvest treatments, litter depth did not differ (Figure 1-la). In contrast, litter
depth did not differ among any of the treatments at the beginning of the subsequent dry
season (15 months post-harvest; Figures 1-1b) although the trend was similar. Only the
intensive harvest treatment had lower litter mass 15 months post-harvest than the control.
Effect of Cover on Fuel Dry-Down Rates
Vegetative cover had the expected effects on both microclimate and fuel dry-down
rates. Maximum understory temperatures were lower and minimum relative humidities
remained higher with increasing vegetative cover during the dry season (Figure 1-2a).
Despite faster dry-down rates of fuels in sites with sparse cover (Fl, 46 = 16.10;
P = 0.0002), 10-h fuels dried down enough to ignite in all sites (Figure 1-2b).
Specifically, during the mid-dry season, 10-h fuels dried down to 12% moisture content
within 3 to 6 days after wetting in open sites (total cover < 110%) and within 7 to 10 days
in shaded sites (total cover > 210%; Figure 1-2b).
During the early dry season of 2002, vegetative cover similarly slowed dry-down
rates, but it was not possible to determine by how many days. A wetter than average
May caused the moisture content of 10-h fuels under most cover conditions to remain
well above 12% throughout the experiment. Ten-hour fuels dried down to 12% moisture
content (2-3 days after soaking) only in the most open sites.
Late dry season
During the early October 2001 fire trials, the area of each plot burned varied
substantially with some plots burning entirely and others not at all. Of the 99 plots tested,
only 21% burned completely (> 90% charred). However, fire carried to at least 1 plot
edge in 57% of the plots. In cases where fires did not reach the plot edge, only 15% of
the plot burned on average (range: 5-35%). Fires did not carry at all in 18% of the plots.
The plot area that burned differed among the three days of the 2001 trial (Table 1-6,
H = 9.2, d.f. = 2, P = 0.01). Because neither cover nor litter depth varied over the three
days, the variation in area burned among days was probably attributable to the cooler,
more humid, less windy, and cloudier weather on the third day compared to the first two
days of the trial (Table 1-6). Because the principal aim was to quantify the effects of key
factors on area burned across sites, the three-day averages per site were used in the path
analy si s.
Path analysis confirmed the expected effects of total cover, litter depth, vapor
pressure deficit, and 1-h fuel moisture content on percent area of test plots burned (Figure
1-3, Table 1-7). Despite the influence of these factors on area burned, a large proportion
of unexplained variance remained (Figure 1-3), which was consistent with the
observation that several plots in deep shade burned while some plots with sparse cover
and apparently ample fuel did not. The moisture content of 10-h fuels in all sites tested
was at or below the theoretical threshold for combustion (Table 1-6) and did not affect
area burned (Figure 1-3, Table 1-7). Wind speeds throughout the trial were low (Table
1-6) and also did not directly affect area burned (Figure 1-3, Table 1-7).
Early dry season
During the 11-12 June 2002 fire trials, no test plots burned. Despite 12 rainless
days, mean minimum relative humidity remained above 70% during the trials.
In July 2002, only 3 of the 30 plots tested burned completely, whereas less than
30% of the test area was charred in the remaining plots. Total cover was sparse (35 to
110%) in the three plots that burned, whereas in the plots that did not burn total cover
ranged from 30 to 415% (mean: 254% + 21). Analysis of paths for which sufficient data
existed indicated that 10-h fuel moisture increased with increasing total cover (partial
regression coefficient = 0.87, P <0.001, R2=0.75), and in turn area burned increased with
decreasing 10-h fuel moisture (partial regression coefficient = -0.79, P = 0.01, R2 =0.69).
Total cover did not affect litter depth and litter depth did not influence area burned.
Persistence of treatment impacts on fire susceptibility
In the August 2002 fire trials to compare the fire susceptibility of forest areas
logged 0, 1, and 3 years previously, only 10 of 94 plots burned. Nevertheless, in the
recently logged area (0-year), 8 of 34 plots burned compared to only 1 of 31 plots in the
area logged 1 year previously (1-year) and 1 of 30 plots in the area logged 3 years
When the data for the three days of the trial were averaged by time since logging,
the mean area of the test plots burned was nearly three times greater in the 0-year
treatment (27%) than in the 1-year (9%) and 3-year (10%) treatments indicating that plots
in the recently logged area were more flammable than plots in the older logging
treatments (F4, 45 = 41.5; P < 0.0001; Table 1-8).
Analysis of paths (by time since logging) for which sufficient data existed indicated
greater total cover increased 10-h fuel moisture and litter depth only in the 1-year
treatment. Litter depth, but not 10-h fuel moisture, in tumn influenced area burned but
only in the 3-yr treatment. Litter in the 0-year treatment was about twice as deep as in
the 1- and 3-year treatments (Table 1-8) and about the same as the litter depth measured 3
months post-harvest during the late dry season of the 2001 fire trials (3.3 + 0.9 cm;
n = 99). Fuel moisture did not influence area burned in any of the treatments. Although
fuels were drier in the 0-year treatment, 10-h fuel moisture was at or below the theoretical
ignition threshold in all treatments (Table 1-8).
Analysis of 5 years of rainfall records revealed that long (~20 days) rainless
periods are common in La Chonta. About 198 & 52 mm of rain falls each month during
the wet season, but only 60 & 3 1 mm of rain falls monthly during the 6 mo dry season and
only 36 & 20 mm of rain falls in La Chonta during each of the driest months (June to
Long rainless periods in all months strongly influenced the results obtained from
the model used to forecast fire-prone days. The number of fire-prone days ranged from 5
in February (the wettest month) to 26 in July and August for open sites (cover < 110%),
from 1 in February to 24 in August for intermediate sites (cover 111-210%), and from 1
in February to 22 in August for closed sites (cover > 210%). When these data were used
to forecast fire-prone days for each cover class, the number of fire-prone days in any
given month was similar among treatments and depended much more on the strength of
the dry season than on the treatments (Figure 1-4). For example, in an average year the
number of fire-prone days in August ranged from 19. 1 in the un-logged control plot to
19.5 in the intensive plot. In a drier than average year, all plots would be fire-prone for
3 1 days in August. These estimates of fire-prone days were based on the cover
conditions in the treatment plots 6 months post-harvest after which the small differences
among treatments further diminished.
Treatment Effects on Forest Structure
Impacts of the silvicultural experiment in La Chonta on forest structure were
modest and similar among treatments despite the greater number of trees and lianas killed
in the intensive management treatment. This similarity in experimental treatment impacts
was attributable in part to the smaller average diameter of harvested trees in the intensive
management plot compared to the normal and improved logging plots. In addition,
although the intensive management treatment doubled the harvest intensity relative to
what is typically applied in similar forests in Bolivia, it was mild when compared to other
logging operations in the tropics where harvest volumes are 2-10 times greater than those
observed in this study (Putz et al. 2001).
Although the intensive management treatment had more gaps than the normal or
improved logging treatments, the average distance to gaps differed by only a few meters
among the treatments. Considering that edge effects on understory microclimate can
penetrate at least 40-60 m (Kapos 1989, Didham and Lawton 1999, Cochrane and
Laurance 2002) and the fact that half the canopy trees in La Chonta lose their leaves
during the height of the dry season, differences among treatments are probably
insufficient to markedly affect fuel dry down rates. Among harvest treatments,
vegetative cover differed only in the mid-canopy strata, but the modest differences in
total cover need to be considered in reference to the relatively open canopy characteristic
of this semi-deciduous forest even before logging. In fact, the most striking result was
the high background level of disturbance in La Chonta: 30% of the control plot
comprised natural gaps or young forest.
Due to its open canopy, La Chonta is already very susceptible to fire even in the
absence of logging. Although elevated fuel loads and decreased canopy cover resulting
from mahogany extraction probably exacerbated the severe fires of 1995 (Pinard et al.
1999b, Mostacedo et al. 2001, Gould et al. 2002), large swaths of apparently undisturbed
forest also burned. Similar fires have occurred in undisturbed forests of the eastern
Amazon, but only after several years of drought (Uhl and Kauffman 1990, Nepstad et al.
1995, Nepstad et al. 1999b). In contrast to these forests, La Chonta receives 300 to
500 mm less rainfall annually and is semi-deciduous, factors that help explain its fire
susceptibility even without logging.
Treatment Impacts on Fuel Loads
The average fuel load (including woody debris and leaf litter) recorded in the
un-logged control area in La Chonta (~27.5 Mg/ha) is on the low end of values reported
for primary forests in the Amazon. For example, fuel loads twice as great were reported
from the eastern Amazon (Uhl and Kauffman 1990) and Venezuela (Kauffman et al.
1988, Delaney et al. 1998). The mass of 100-h fuels in La Chonta was 10 times lower
(1.6 vs. 16.8 Mg/ha) than what Cochrane et al. (1999) measured in un-logged forests in
the eastern Amazon. The relatively low mass of dead and down fuels in La Chonta is
probably attributable to the low tree biomass, which is an expected consequence of lower
rainfall (Murphy and Lugo 1986) and high liana density (Alvira et al. 2004). Similarly
low fuel loads were found in second growth forests in Brazil (Uhl and Kauffman 1990).
Relative to the control, fuel loads were doubled by the normal logging treatment
and tripled by the intensive treatment. These increases were mainly attributable to a
greater amount of sound 1,000-h fuels resulting from harvest damage and residues that
persist for at least 15 months post-treatment. The relative increase in fuel loads from
intensive management compared to the un-logged control was similar to that reported by
Uhl and Kauffman (1990). In their study, however, the magnitude of woody debris
inputs (150 m3/ha) was much greater than reported here. This difference might be
attributable to the size of trees harvested, harvest practices, and to the greater overall
volume (50 m3 VS. 18 m3) and number of harvested trees (8 vs. 5 ha l) in their study site
Although the harvest treatments increased woody fuel loads, they decreased leaf
litter depth (6 months post-treatment) and mass (15 months post-treatment) compared to
the control, with the lowest values in the intensive treatment plot. These results differ
from those reported by Uhl and Kauffman (1990); in their study site, the mass of fine
fuels was 30% greater in the logged vs. the control area. Site factors (e.g., greater leaf
area index and litter production in the eastern Amazon) probably explain most of this
difference, but the timing of the censuses might also be relevant. Uhl and Kauffman
(1990) estimated litter mass a few months post-logging, whereas in this study litter mass
was quantified 15 months post-logging, which corresponded to the early dry season when
leaf litter is typically sparse (personal observation).
In this study, it appears that the harvest impacts on fuel loads are more likely to
affect potential fire severity than fire susceptibility. The quantity, arrangement, and
moisture content of fine fuels, including litter, determine to a large extent the
susceptibility of a particular area of forest to fire (Stott 2000, Cochrane 2003). Once a
fire ignites, the arrangement, quantity, and moisture content of larger fuels influence
whether combustion is sustained and also determine the fire's severity (Stott 2000).
Although this study did not assess fire behavior, it is nevertheless reasonable to infer that
fire severity would be greatest in the treatments with the greatest quantity of fuels,
especially 1,000-h and larger fuels. The harvest treatments at La Chonta would increase
fire severity relative to the control, but the potential for severe fires would be greatest in
the intensive treatment because of the three-fold increase in 1,000-h fuels relative to the
normal treatment. The greater quantity of 1,000-h fuels in the intensive treatment plot
implies that if a fire were to occur, it would smolder longer and be harder to extinguish
(Cochrane 2003), and consequently cause higher tree mortality rates (Kauffman 1991)
than in the other treatment plots.
Vegetative Cover and Fuel Dry-Down Rates
As described for forested areas throughout the tropics (Walsh 1996), understory
vapor pressure deficits decreased with greater total cover in La Chonta. That the effect of
cover on vapor pressure deficit was strongest early in the dry season and weakest in the
mid-dry season indicates that once the forest dries, not even sites with dense cover retain
much moisture. Only about half of the canopy species and canopy trees retain their
leaves during the greatest period of water stress (i.e., July-August; Justiniano,
The semi-deciduous canopy of La Chonta differs from the evergreen forests of the
eastern Amazon, the canopies of which remain closed (thanks to deep rooting of most
tree species) despite equally long and severe droughts (Nepstad et al. 1995). Retention of
an evergreen canopy in the eastern Amazon is considered to be the key attribute that
provides fire immunity during normal years because it prevents desiccation of the
understory vegetation and surface fuels (Nepstad et al. 1995).
Ten-hour fuels in the most open sites dried below the moisture content threshold
within 3-6 days, a result similar to what Holdsworth and Uhl (1997) found in the eastern
Amazon. In contrast, 10-h fuels required only 7-10 rainless days to dry enough to ignite
even in the most shaded sites in La Chonta, whereas 10-h fuels remained above threshold
levels at which ignition could occur throughout the dry season in un-logged forests of the
eastern Amazon (Uhl and Kauffman 1990, Holdsworth and Uhl 1997) and Venezuela
(Uhl et al. 1988). These results suggest that the understory microclimate in semi-
deciduous forests like La Chonta apparently is much drier than the evergreen forests
elsewhere in the Amazon where similar fire studies have been conducted, and that open
sites in La Chonta can bum after only 3 rainless days, while most undisturbed sites can
burn after only 10 rainless days.
The extent to which experimental fires carried depended on vegetative cover, litter
depth, relative humidity, and fuel moisture content. These factors are well known to
influence forest flammability (Stott 2000), but they only explained 59% of the observed
variation. In the late dry season (October 2001) and only a few months after logging,
many plots in sites with sparse vegetation and apparently sufficient litter did not bum
completely, and conversely, some plots in densely shaded areas burned completely or
nearly so. This latter observation contrasts with results from undisturbed evergreen
forests in Amazonian Brazil (Uhl and Kauffman 1990, but see Holdsworth and Uhl
1997), Venezuela (Uhl et al. 1988, Kauffman and Uhl 1990), and Indonesia (Siegert et al.
2001, van Nieuwstadt et al. 2001), which remain fire resistant in the absence of severe
disturbances or El Nifio associated droughts. The deciduous nature and open canopy
structure of La Chonta apparently renders it fire-prone even in the absence of logging or
In the early dry season, at least at the management intensities tested in this study,
La Chonta is not very fire-prone. The failure of any plots to bum in June 2002 supports
this conclusion. Moreover, few if any ignition sources exist during the early dry season
because most small farmers and ranchers wait until July or August to set fires, and
lightning initiated fires are rare (Tutin et al. 1996, Stott 2000, Saarnak 2001). All of the
destructive fires of the past 4-5 years (since a fire monitoring system was established)
occurred in August or September (Politica y plan national de prevencion y control de
incendios forestales en Bolivia. 2002. Ministerio de desarrollo sostenible y planificacion,
La Paz, Reputblica de Bolivia). Finally, fine fuels are much less available in the early dry
season than they are toward the end of the dry season after half the canopy trees shed
Factors influencing fire susceptibility
The key factors explaining variation in area burned varied somewhat during the
dry season, but moisture content of fuels, litter depth, and vapor pressure deficit were
always important. The lower number of plots that burned on day 3 of the October 2001
fire trial when conditions were overcast and humid exemplified the influence of low
vapor pressure deficit, which resulted in higher moisture contents of the short time-lag
fuels. Low vapor pressure deficit during the July 2002 trial similarly appeared to render
most plots non-flammable.
Fuel moisture content, a critical determinant of forest flammability (Stott 2000),
influenced the area of plots burned to a different extent during the dry season.
Specifically, moisture content of 10-hour fuels was a better predictor of the area of plots
burned earlier in the dry season because the moisture content of 10-h fuels was about the
same regardless of cover in the late dry season. Cochrane's (2003) suggestion that 1-h
fuel moisture would be a more appropriate predictor of fire susceptibility than 10-h fuel
moisture was supported by the path analysis of the late dry season fire data.
Nevertheless, the use of 10-h fuel dry down rates in the fire-prone days model is
considered valid because 10-h fuel moisture did explain whether plots burned in the
early-dry season. The packed arrangement of 1-h fuels on the forest surface explains
why they may remain moister than 10-h fuels (Uhl et al. 1988), which were suspended
25 cm above the surface.
Differences in litter depth among plots also helped explain whether plots burned.
Fuel coverage and quantity are key determinants of whether fires will carry across the
forest floor (Stott 2000). Regardless of time during the dry season, few plots burned if
they had litter less than 2. 1 cm deep. It is conceivable but unlikely that sufficient litter
would accumulate by the end of the dry season in the areas logged 1 and 3 years
previously to render them more flammable because the senesced leaves of the pioneer
species occupying the disturbed areas tested were already on the ground.
As reported for other tropical forests (Cochrane 2003), the extent to which plots
burned decreased with increasing cover in La Chonta, but this effect was strongest in the
early dry season. Apparently, only sites in or very near large gaps are likely to be
fire-prone early in the dry season. To the extent that intensive management creates more
gaps, it will elevate fire susceptibility in the early dry season. This differential effect
should be smaller, however, in the late dry season because by then fuels are dry enough
to ignite regardless of cover. In summary, intensified management, at least to the extent
tested in La Chonta, did not appreciably elevate fire susceptibility compared to the
un-logged control or normal management practices, at least in the late dry season.
Persistence of treatment impacts on fire susceptibility
The modest increase in susceptibility to fire in the early dry season resulting from
intensive timber stand management in La Chonta seems to persist only for the year
immediately following treatment; almost none of the plots in severely disturbed sites that
were logged 1 and 3 years previously burned. In contrast, about one-fifth of the plots
burned in an area logged 2-4 months previously suggesting that the elevated fire
susceptibility associated with intensifying management in the early- to mid-dry season
would become indistinguishable from normal harvest practices within 1 year. In the
eastern Amazon, Holdsworth and Uhl (1997) found that re-growth in 3-year old logging
gaps slowed fuel dry down rates rendering such gaps fire resistant. In La Chonta, 10-h
fuels were drier in the area logged immediately prior to the fire trials than in the older
logging areas, but even fuels in areas logged 1 and 3 years previously were dry enough to
burn. Rather than fuel moisture, depth of fine fuels was the more important factor in the
few plots that burned in the areas logged 1 and 3 years previously.
Treatment impacts on number of fire-prone days
The most striking result from the model employed to estimate the number of days
the forest is fire-prone in response to each treatment was that the effects of inter-annual
variability in rainless periods greatly exceed any differences among treatments. For
example, only 22 mm of rain fell in 2 rain events from July to September 1999, which
resulted in 76 fire-prone days for closed forest patches according to the model. During
the same period in 2000, 215 mm of rain fell in 7 rain events, which resulted in only
47 fire-prone days. In an average year, the long rainless periods in the middle of the dry
season resulted in very minor differences (maximum of 4 days) in fire-prone days
between open, intermediate and closed sites. Considering the modest treatment impacts
on cover, the difference in fire-prone days among treatments regardless of month was
Among the many assumptions inherent in the model I used to predict the number of
fire-prone days, the most important are that 12% moisture content is a valid threshold on
which to base fire susceptibility and that the number of consecutive rainless days is the
best predictor of fire susceptibility. A more elaborate model might include fuel loads
(especially litter depth and arrangement) as well as vapor pressure deficit, both of which
helped explain the test fire results.
Conclusions and Management Implications
Four major conclusions can be drawn from the previous discussion. First,
un-logged seasonally deciduous forests like La Chonta appear to be very susceptible to
fire throughout much of the dry season. This finding contrasts with results reported for
intact evergreen forests of the eastern Amazon, which remain fire resistant except during
the driest years. Second, the range of timber harvest intensities assessed in this study had
a trivial impact on the forest' s susceptibility to fire, and this small impact diminished
rapidly with time. Therefore, there appear to be only very modest and short-lived
tradeoffs between doubling the management intensity in an effort to secure adequate
regeneration and growth for sustained timber yields and fire susceptibility in these
forests. Assuming the forest can be adequately protected from ignition, especially during
the year after harvesting, silvicultural activities at the intensities carried out in this study
should not be viewed as creating excessively threatening conditions for fire. These
conclusions must be considered, however, in the context of the substantial increase in
coarse woody debris resulting from intensive management. Should a fire occur, it would
be most severe where the forest is managed intensively because of the increased quantity
of 1,000-h fuels.
Based on the these conclusions, it is clear that better control of ignition sources to
prevent fires must complement silviculture treatments to achieve sustained yields if these
forests are to satisfy both production and conservation goals. To some extent, training
forest crews in fire prevention and control techniques will help reduce fire damages to
production forests. Vigilance against fires escaping from nearby pastures (or starting
within logging areas) should be concentrated in the same year as harvest activities, which
is relatively easy if roads are designed as fire breaks. Moreover, forest managers should
consider spatial arrangements of harvest units such that sufficiently wide buffers
(Holdsworth and Uhl  suggested 1 km) surround each unit at least for 1-3 years
Unfortunately, controlling ignition sources in the Amazon will require a
monumental shift in attitudes and practices and an equally large improvement in
governance so that the positive economic and biophysical feedbacks that are making the
Amazon Basin more fire-prone can be broken (Nepstad et al. 2001). A better
understanding of fire science across the range of forest types in the Amazon will provide
a sound basis for breaking these feedbacks (Cochrane 2003). Equally important is greater
emphasis on multidisciplinary research that elucidates cultural and social uses and
attitudes pertaining to fire (Roman-Cuesta et al. 2003), as well as an increase in
interagency and intergovernmental cooperation. Considering the rapidly changing land
use and climate in the region, unless greater priority is given to better defining and
resolving the multiple dimensions of the fire problem in different biophysical and
socioeconomic contexts, the kind of fire-prone forests found in La Chonta may become
more common throughout the entire Amazon Basin.
Table 1-1. Impacts of timber harvesting and silvicultural treatments in Block 2 of the
Long-term Silvicultural Research Proj ect in La Chonta.
Parameter Normal Improved Intensive
Plot area (ha) 25.9 27.8 25.0
Trees harvested per ha 3.3 2.8 5.0
Basal area (m2) harvested per ha 2.1 1.4 2.7
Volume (m3) harvested per ha 16.8 11.9 18.5
Mean (f se) dbh of harvested trees 84.3 f 3.5 77.7 f2.8 70.7 f 2. 1
Reduction in tree density (%)a 7.8 (11.8) 7.3 (11.1) 12.5 (17.9)
Source: unpublished data from the Bolivian Institute for Forestry Research (IBIF).
a Total percentage of trees harvested and killed (>10 cm dbh; based on differences in
census 1 and 2). The number in parentheses refers to the proportion of trees > 40 cm dbh
that was harvested.
Table 1-2. Percentage of each treatment plot in Block 2 of La Chonta observed in
corresponding habitat classes based on point sampling along four 400-450 m
transects in each treatment plot 6 months post-harvest.
Habitat Class Control Normal Improved Intensive
Undi sturbed 62.3 35.8 39.2 28.4
Skid trail 0 5.9 5.2 4.9
Skid trail edge 0 12.4 10.5 9.8
Felling gap 0 16.0 9.9 15.0
Felling gap edge 0 15.1 14.9 22.6
Total harvest disturbed 0 49.4 40.6 52.3
Natural gap 7.7 4.1 3.0 5.5
Natural gap edge 30.0 10.7 17.1 13.8
Total natural disturbed 37.7 14.8 20.1 19.3
Total 100.0 100.0 100.0 100.0
Table 1-3. Summary of assessment of treatment impacts on forest structure showing means a standard errors for estimates of
vegetative cover (%) in 6 vertical strata, total cover, maximum height, and horizontal distance to nearest gap.
Vegetative cover (%) Maximum Distance to
Treatment N 0-1 m 1-2 m 2-4 m 4-8 m 8-16 m >16 m Total height (m) gap (m)
Control 377 52.9 11.1 29.4 11.2 41.7 11.6 67.0 11.3 64.4 11.6a 29.0 11.6 284.4 13.3a 13.8 10.4 10.3 10.4a
Normal 338 47.6 11.4 27.2 11.3 35.4 11.6 55.7 11.5 54.8 11.5b 21.7 11.5 242.4 13.9b 12.4 10.4 7.4 10.3bc
Improved 363 52.4 11.2 28.8 11.4 39.9 11.8 57.9 11.6 58.8 11.6b 23.5 11.4 260.9 +4.0b 12.2 10.4 8.4 10.3ab
Intensive 327 50.5 11.4 32.7 11.4 35.0 11.6 54.3 11.6 46.2 11.6b 18.8 11.4 237.5 14.2b 11.5 10.3 6.3 10.3"
Notes: The maximum value for total cover = 600%. Estimates of all variables were made every 5 m (N = number of sample points)
along each of four 450 m-long transects in each treatment plot. Different letters indicate differences among treatments at P < 0.05
based on 1-way ANOVAs in which sample means for each transect in each treatment plot were compared (n=4 per plot).
Homogenous subsets determined with a Tukey's post-hoc test.
Table 1-4. Mean quantity of woody debris encountered per meter of transect in La Chonta.
Fuel size / decay class
1,000-h 1,000-h 1,000-h
Treatment N 1-h 10O-h 100O-h sound intermediate rotten
Control 79 17.00 (10.20) 5.00 (3.41) 0.27 (0.32) 0.00 (0.02) 0.02 (0.05) 0.04 (0.08)
Normal 77 20.21 (12.39) 4.74 (4.40) 0.48 (0.50) 0.06 (0.13) 0.05 (0.09) 0.04 (0.07)
Improved 82 24.27 (16.72) 4.40 (2.61) 0.45 (0.40) 0.05 (0.15) 0.03 (0.06) 0.03 (0.05)
Intensive 87 20.02 (16.09) 4. 10 (3.11) 0.62 (0.65) 0.10 (0.26) 0.05 (0.10) 0.02 (0.04)
Notes: Values are the mean number of fragments of dead and down woody material measured in two censuses (6 months and 15
months post-treatment). Numbers in parentheses are standard deviations. Fuel sizes classified by diameter according to the fuel
time-lag concept (see text): 1-h (<0.6 cm); 10-h (0.6-2.5 cm); 100-h (2.5-7.5 cm); and 1,000-h (> 7.5 cm). Fuel decay classes follow
Delaney et al. (1998). N = total number of transects from the two censuses. Transect lengths varied by size class as follows: 1-h: 1 m;
10-h: 2 m; 100-h: 5 m; and 1,000-h: 11 m.
Notes: Means (a standard error) based on averages from two censuses 6 and 15 months post-treatment. Fuel sizes classified by
diameter according to the fuel time lag concept (see text): 1-h (<0.6 cm); 10-h (0.6-2.5 cm); 100-h (2.5-7.5 cm); and 1,000-h
(>7.5 cm). Decay classes follow Delaney et al. (1998). Different letters next to treatments indicate significant differences based on
1-way ANOVA with masses loglo transformed (F3, 321 = 3.81; P = 0.01; N = 79 for control, 77 for normal harvest, 82 for improved
harvest, and 87 for intensive management). Values shown are based on actual data, not the back-transformed data.
Table 1-5. Mean densities and mass estimates for each size and decay class of woody debris in La Chonta.
0.34 & 0.01
0.35 & 0.03
0.30 + 0.01
0.66 & 0.06
0.59 & 0.06
0.47 & 0.09
1.4 & 0. 1
4.3 & 0.3
1.6 & 0.2
0.3 & 0.2
7.3 & 4.7
22.0 & 5.0
1.7 & 0.1
4.1 & 0.4
2.9 & 0.3
11.9 & 4.0
12.4 & 6.3
8.3 & 2.4
41.3 & 8.0
1.9 & 0.1
3.8 & 0.2
2.8 & 0.3
11.2 & 3.4
11.4 & 5.5
37.1 & 6.4
1.7 & 0.2
3.4 & 0.3
3.8 & 0.4
35.8 & 12.5
14.8 & 5.3
3.7 & 1.1
Fuel size/decay class
Table 1-6. Summary of 7-9 October 2001 (late dry season) test fires in La Chonta showing means a standard error for measured
variables for each of the three consecutive days of the trial and the three-day averages.
Max. Area Relative Wind Litter 10-h fuel
burn burn humidity Temp. VPD speed moisture moisture
(cm) (%) (%) (oC) (kPa) (m/s) (%) (%)
33 33 5 6 5 33 33 33
74.5 & 5.7 45.0 & 5.7 40.4 & 4.6 35.1 & 0.9 3.5 & 0.5 0.6 & 0.0 17.5 A 1.5 10.2 & 0.2
87.0 &3.9 55.6 &5.4 38.2 &2.5 35.5 &0.8 3.7 &0.5 0.6 & 0.0 14.9 & 1.1 10.1 & 0.2
67.1 & 3.2 34.1 & 5.2 50.4 & 1.6 30.3 & 0.7 2.3 & 0.4 0.4 & 0.0 18.5 A 1.4 12.0 + 0.2
76.2 &4.2 44.9 &4.3 43.0 &2.9 33.6 &0.6 3.1 & 0.5 0.5 & 0.0 16.9 & 1.1 10.8 & 0.2
199.8 & 10.8
200.3 A 12.6
206.3 A 12.4
202.2 & 6.8
33.6 & 1.6
33.1 & 0.9
Notes: Test fires were set at 1200-1430 h in 33 sites per day representing the range of cover conditions found throughout the 4
management treatments 2-3 months post-harvest. Max. burn = maximum linear distance from the center toward the edge of the
2 m x 2 m plot that burned; Area burn = percent surface area of each plot that burned; VPD = vapor pressure deficit; Total
cover = sum of percent vegetative cover estimates for six vertical strata (maximum = 600%). Data for relative humidity and
temperature are daily mean minimum relative humidity and mean maximum air temperature among 6 data loggers (at a height of
~25 cm) at sites across the cover range. The humidity sensor malfunctioned in one of the data loggers, hence N = 5. Vapor pressure
deficits calculated from temperatures and relative humidities.
Table 1-7. Path coefficients, P-values, and proportion of variance explained by factors
influencing area burned in October 2001 test fires in La Chonta determined by
Dependent variable Independent variable coefficient P R2
Litter depth Total cover 0.51 0.001 0.24
Vapor pressure deficit (VPD) Wind 0.51 0.001 0.45
Total cover -0.47 0.002
10-hr fuel moisture Total cover 0.38 0.023 0.39
VPD -0.35 0.034
1-hr fuel moisture Total cover 0.30 0.070 0.39
VPD -0.43 0.011
Burn area (%) VPD 0.40 0.043 0.59
Wind -0.26 0.092
Litter depth 0.66 0.000
10-hr fuel moisture -0.19 0.279
1-hr fuel moisture -0.43 0.014
Notes: The sample size for this fire trial was 33 test fires. See Figure
analy si s.
1-3 for the path
Table 1-8. Summary of August 2002 fire trial in La Chonta showing means, standard errors, and ANOVA or Kruskal-Wallis results
for total cover, vapor pressure deficit (VPD), average litter depth, 10-h fuel moisture content, and area burned of 4 m2 plOts
by year post-logging.
Measure 0-vr 1-vr 3 -vr F d.f. P H
Total cover (%) 119.9 & 16.4a 192.5 A 12.0b 204.3 & 8.7b 12.9 2, 47 0.001
VPD (kPa)2 5.4 & 0.2a 3.2~ O .1b 3.2 & 0.1b 88.1 2, 6 <0.001
Area burned (%/) 26.7 & 5.5a 8.7 & 3.3b 9.8 & 2.6b 7.2 2, 48 0.002
Litter depth (mm) 3.5 & 0.3a 1.4~ O .1b 1.5 & 0.1b <0.001 29.6
10-h fuel moisture (%) 9.0 + 0.4a 12.2 & 0.5b 12.5 & 0.4b <0.001 23.1
Notes: N=17 plots per treatment, except in the 0-yr treatment where N=15 for 10-h fuel moisture, and N=16 for total cover.
Percentages were arcsine square root transformed. Means for VPD are based on 3 data loggers per treatment for the 3 days of
the trial. Different letters indicate differences between treatments based on post-hoc comparisons with Tukey or
Mann-Whitney U tests.
Control Normal Improved Intensive Control Normal Improved Intensive
Figure 1-1. Treatment impacts on litter depth during the rainy and dry seasons in La Chonta. A) Rainy season, 6 months
post-harvest. B) Dry season, 15 months post-harvest. Boxes represent quartiles, whiskers the 10th and 90th percentiles,
and dots the 5th and 95th percentiles. The solid line shows the median. Different letters indicate significant differences.
Sample sizes are shown in parentheses.
b b b
100 200 300 400 500
May 2002; R2 = 0.87; P<0.001; y = 62.4 + 0.07x
June 2002; R2 = 0.77; P<0.01; y = 65.0 + 0.06x
Total Cover (%)
R2 = 0.26; P=0.002;
*y = 3.983 e0.002x
Total Cover (%)
Figure 1-2. Effect of vegetative cover on understory microclimate and fuel dry-down
rates in La Chonta during the early dry season of 2002. A) Mean minimum
relative humidity. B) Number of rainless days required for 10-h fuels to dry
down from saturation to 12% moisture. Total cover is the sum of percent
cover estimates for six vertical strata. Points in A represent the means of
midday relative humidity (%) measured with Hobo data loggers placed
under different cover conditions 1-4 months post-harvest in Block 3. Error
bars represent standard errors. Points in B represent the number of days
each 10-h fuel stick required to reach 12% moisture content in July 2001.
The R2 and P-values are based on regression analysis.
1-hrM 0.78 U1
U M ~Litter ooAe
Figure 1-3.Path diagram for the effects of total cover, vapor pressure deficit, litter depth, wind, 10-h fuel moisture content (10-h MC),
and 1-h fuel moisture content (1-h MC) on area burned (%) in the 7-9 October 2001 test fires in La Chonta. Arrows
illustrate hypothesized paths of causation; solid arrows indicate positive effects and broken arrows negative effects.
Arrows with adjacent values indicate significant effects at P <0.05. Magnitude of the values indicates strength of the
effects illustrated. Unmeasured factors affecting the variables are represented by "U". See Table 1-7 for P- and R2-ValUeS.
I I I I
I I I I
I I I I
I I I I
l i Il
Jan Feb Mar Apr
i l l i l l i
May Jun Jul Aug Sep Oct Nov
Figure 1-4.The number of fire-prone days per month predicted for each management treatment applied in La Chonta based on a model
of consecutive rainless days and the proportion of each treatment plot consisting of "open" (<1 10% cover), "intermediate"
(110-210% cover), and "closed" (>210% cover) forest patches (see text for details). Boxes represent quartiles, whiskers
the 10th and 90th percentiles, and dots the 5th and 95th percentiles. The solid line shows the median.
I i I
I I : I
I I I I
( I I
I I : I
( I I I
( I I
I I : I
I I I I
I I : I
I I r
01 I II g R i i
,I I ,,
,, I I
1,1 T (I I
I I II II I
I I I ( I
I' II( II I
,I ,,,, I
,, I I
,I I I
g g( i OQI I
PROJECTING FOREST RESPONSES TO DIFFERENT SILVICULTURAL
TREATMENTS IN A SEMI-DECIDUOUS FOREST IN LOWLAND BOLIVIA
Although tropical forests are valued for their rich biodiversity (Frumhoff and Losos
1998, Fearnside 1999a) and the myriad goods and services they provide (Johnson and
Cabarle 1993), Einancial analyses based on available markets typically find that the
timber production value of tropical forests is paramount (Pearce et al. 1999). Moreover,
considering that most tropical governments view their forests as engines for sustainable
development and given that protected areas alone are insufficient to conserve all of the
tropics' biodiversity, the sustainable management of production forests has been
promoted as a means to achieve both development and conservation goals (Dickenson et
al. 1996, Chazdon 1998, Poore et al. 1999, Whitmore 1999). Unfortunately, managing
tropical forests sustainably, or even just for sustained timber yields (STY), remains an
elusive challenge for many reasons not the least of which is the limited information about
the silvicultural requirements of commercial species used to set policies and management
guidelines (Mostacedo and Fredericksen 1999).
Forest management in Bolivia has advanced substantially in the past decade (Nittler
and Nash 1999) and an increasing number of forestry enterprises are seeking Forest
Stewardship Council (FSC) certification (Blate et al. 2001, Fredericksen et al. 2003).
Despite impressive advances, the Bolivian forest sector must contend with poor
infrastructure, high transport costs, and limited market access, all of which constrain
investments in training, field supervision, and silviculture.
In semi-deciduous forests, which account for 45% of Bolivia's timber production,
foresters face a host of management problems that if left unresolved may undermine the
aspirations of the forest sector and conservationists alike. First, application of
reduced-impact logging practices alone is not a viable strategy for sustaining timber
yields of the principally light-demanding commercial species because they regenerate
poorly without intensive silvicultural treatments (Mostacedo and Fredericksen 1999,
Fredericksen and Putz 2003). Second, these forests harbor lianas at densities that are
among the highest recorded in the tropics (Perez-Salicrup et al. 2001, Alvira et al. 2004),
which further inhibits commercial species regeneration and growth (Whigham 1984,
Clark and Clark 1990, Putz 1991, Schnitzer et al. 2000, Gerwing 2001, Alvira et al. 2004,
Grauel and Putz 2004). Third, most of these forests are highly susceptible to fire during
the long and severe dry season, and because fires frequently escape from the agricultural
matrix in which most remaining forests are embedded (Chapter 1). Finally, the current
silvicultural rules were developed as first approximations because of limited knowledge
of the population biology and silvicultural requirements of the species being managed
(Fredericksen et al. 2003). Thus, questions remain about how to achieve STY.
In 2001, a silvicultural experiment was implemented to determine if the
regeneration and growth problems could be mitigated to secure STY (IBIF 2004). The
only way to obtain conclusive results of the experiment is to wait for 20-30 years, but
rapid land use changes and heavy pressure on forests in general motivated an alternative
approach that would provide a sufficient basis for effecting early changes to forest
policies and on-the-ground management practices. The main alternative to waiting
decades is to proj ect the likely outcomes of the experiment via simulation modeling. The
principal assumption with this approach is that the medium- to long-term consequences
of the management treatments can be proj ected with reasonable confidence.
It is difficult to model the dynamics of tropical forests because of their high species
diversity, uneven-aged structure, and spatially heterogeneous nature (Vanclay 1994).
Spatially explicit, individual-based models are most appropriate for capturing this
complexity especially when the goal is to evaluate the effects of silvicultural treatments
(Gourlet-Fleury and Houllier 2000, Phillips et al. 2003), which are applied to individual
trees or specific ground areas (e.g., felling gaps). These models track individual trees
over time either by simulating physiological processes (e.g., carbon budgets) for each tree
in the stand (Ditzer et al. 2000, Huth and Ditzer 2000, Kammesheidt et al. 2001,
Chambers et al. 2004), or by simulating the fundamental ecological processes of growth,
mortality, recruitment, and competition (Pacala et al. 1996, Gourlet-Fleury and Houllier
2000, Phillips et al. 2003, Phillips et al. 2004b, van Ulft 2004, Arets 2005). The
physiological process models require detailed site- and species-specific information about
photosynthesis, respiration, and decomposition, which are unavailable in Bolivia. In
contrast, models that simulate ecological processes rely on permanent plot data, which
were available. I chose to use the SYMFOR modeling platform (Phillips et al. 2003,
Phillips et al. 2004b) because it provides a framework to simulate these processes on a
spatially explicit, individual tree basis. In addition, the ability to use, modify, and add
management scenarios (van Gardingen et al. 2003) allows comparisons among the likely
outcomes of the different silvicultural treatments and for assessing trade-offs between
timber production and other forest values (e.g., carbon sequestration).
Using SYMFOR, I aimed to address the following research questions in this
* How effective were the applied silvicultural treatments in sustaining timber yields?
* How do forest structure and composition of change in response to different
* How do future timber harvest volumes differ in response to each management
* How sensitive is the model to parameters for which there are little or no data?
* How reliable are the model's proj sections?
An additional goal of the model developed in this chapter was to provide a basis for
elucidating the trade-offs between timber production and carbon sequestration, which is a
maj or focus of Chapter 3.
Because the model was based on data spanning only 3 years, I limited my
proj sections to 62 years with 3 harvest entries beginning in year 0 and a 30-year cutting
cycle. Although the limited data for model calibration injects considerable uncertainty in
the estimates, the study provides a first approximation of the likely impacts and
effectiveness of the silvicultural treatments applied.
My main purpose in this chapter was to answer practical management questions
rather than to develop a new or more sophisticated model. Thus, I only briefly describe
the version of SYMFOR used; additional details are available in Phillips et al. (2002,
2003, 2004b) and at www. symfor.org.
Site and Data Source
The data to calibrate the model used in this study came from a 100,000 ha timber
concession operated by Agroindustria Forestal La Chonta Ltda. in Guarayos Forest
Reserve (15045'S, 62060'W) in the Bolivian lowlands (200-400 m.a.s.1.). Although the
Holdridge classification system classifies the concession (hereafter La Chonta) as
subtropical humid forest, La Chonta is situated in a transitional zone between wetter
forests to the north and drier forests to the south and southeast. The dominant species
include Ficus boliviana, Hura cre itans, and Pseudolmedia laevis, but s ecies common in
drier forests are also present (e.g., Centrolobium microchaete, Chorisia speciosa, and
Aspidosperma rnd). Mean annual temperature is ~24.50C and mean annual rainfall is
~1 500 mm, 77% of which falls between November and A ril. For stems with dbh >
10 cm the mean density is 35818 trees/ha and mean basal area is 21.110.7 m2/ha (IBIF
2004). The average canopy height for trees with dbh 2 40 cm is 28.010.2 m with
emergent trees reaching heights of 50 m (IBIF 2004). The forest is extremely fire prone
for about 130 days/year (Chapter 1); in 1995, 30% of the concession burned (Pinard et al.
1999), and a 1999 fire destroyed the town adjacent to the concession. Nevertheless, there
were no signs of recent fires in my study area. Soils are moderately fertile inceptisols
and 10-15% of the area has black anthrosols (Paz 2003).
La Chonta typically harvests 10-12 tree species and 7-20 m3/ha (3-5 trees/ha).
Harvest activities are planned based on a 30-year cutting cycle and implemented in
accordance with Forest Stewardship Council principles and criteria. The minimum
diameter for felling MDF) set by law is 70 cm for H. cre itans and F. boliviana, and
50 cm for all other species. Twenty percent of trees > MDF are left as seed trees.
Experimental design and silvicultural treatments
A silvicultural experiment in La Chonta was established in 2001 as part of a
multi-site long-term silvicultural research project (LTSRP, IBIF 2004). In La Chonta, the
LTSRP applied four management treatments of increasing intensity: "control" = no
logging; "normal logging" = standard planned logging according to the stipulations of the
Bolivian forestry law with no other silvicultural interventions; "improved management" =
"normal logging" with liberation of future crop trees from vines and overtopping non-
commercial trees; and, "intensive management" = "improved management" with double
the harvest intensity (as measured by number of trees harvested), additional future crop
tree liberation, and soil scariaication in selected felling gaps. The improved and intensive
treatments aimed to promote the regeneration and growth of commercial timber species,
most of which are light-demanding (Mostacedo and Fredericksen 1999, Pariona et al.
2003). Evidence for the potential benefit of soil scarifieation on commercial regeneration
was particularly strong: Fredericksen and Pariona (2002) found that Schizolobium
amazonicum densities were nearly 10 times higher and height growth was twice as great
in scarified vs. unscarified felling gaps.
Treatments were replicated in three blocks, each situated in a different annual
harvest unit, and were randomly applied to 27 ha plots in each block (3 blocks x 4
treatments = 12 plots. (A map of La Chonta and block layout is available at:
http://www. ibifbolivia. org.bo/ESP/estaciones_de_campo/la~chonta. htm). All trees 1
40 cm dbh were identified, tagged, measured, and mapped in these large plots. Smaller
trees were censused in sub-plots nested within each large plot; trees 20 cm dbh were
measured in 13-ha sub lots and trees > 10 cm dbh were measured in four 1-ha plots.
Data to calibrate the model came from the first 3 annual censuses in blocks 2 and 3 (2001
to 2003). Data collected over the same period in block 1 were used for model testing. In
simulations, I used 15 of the 32 available 1-ha plots from blocks 2 and 3.
Model Description and Calibration
SYMFOR is a strategic modeling framework that allows managers to simulate and
compare different forest management scenarios (Phillips et al. 2003, 2004b). SYMFOR' s
ecological model consists of linked sub-models for competition, growth, mortality, and
recruitment that act on individual trees in annual time steps to simulate the spatial and
temporal dynamics of the forest (see Figure 2-1 for model overview and flow). The
SYMFOR model I used assumes a homogeneous abiotic environment. In this study, the
effects of liana infestation were included as a modifier of growth, mortality, and
recruitment. I used the La Chonta data to calibrate these models (see Appendix A for a
list of the parameters used in this version of SYMFOR and their symbols). Separate
management models simulated treatments (i.e., harvest or kill trees, cut lianas, and
scarify soil) applied in the LTSRP in La Chonta (IBIF 2004). Although La Chonta is
quite fire prone, I did not explicitly include fire risk or the ecological effects of fire in the
model to keep it tractable and focused on the likely outcomes of the silvicultural
Species grouping. The purpose of aggregating species into ecological groups was
to retain as much of the variation in forest dynamics as possible to allow realistic
simulations while creating a tractable model by assuming that all members of a group
behave similarly in terms of growth, recruitment, mortality and response to competition.
I aggregated the 153 non-palm tree species (>10 cm dbh) found in La Chonta into
8 ecological groups in 3 steps using the same approach as Phillips et al. (2002). First,
species with at least 25 valid growth records were grouped using cluster analysis (SAS
9.0) based on the 75th percentile of the growth rate (for each species), the maximum
height of the species, and the proportion of juvenile trees (10-20 cm dbh) growing in
high light. Second, using the same variables as those used in the cluster analysis, I
performed a discriminant analysis to assign the remaining (rare) species to one of the 7
groups obtained from the cluster analysis.
The third step involved peer-review of the groups by forest ecologists familiar with
the species in La Chonta. This step led us to shift 6 species between groups and also to
split the shade-tolerant canopy group into fast- and slow-growing species. The final
species grouping is shown in Table 2-1. I excluded palms, which comprise only 2.5% of
all stems in La Chonta, due to lack of data to adequately model their growth, mortality,
Utility groups. SYMFOR uses utility groups, a second kind of species grouping,
to distinguish which species within a species group (SG) are commercial, potentially
commercial, and non-commercial. Although utility groups are used during the simulation
of management treatments, they are assigned by the user in the input data and
subsequently assigned to new trees during recruitment (see Recruitment and Management
Models for further details).
Competition. Competition among trees for resources (light, water, nutrients, and
space) was represented by a competition index (Cr) following the approach of Phillips et
al. (2003, 2004b). The index for each individual is based on the dbh of, and distance to,
larger diameter neighbors that were assumed to be taller:
Ct = -, [2-1]
where D, is the dbh (in cm) of the focal tree, D, is the dbh of a neighbor (within a radius
of 20 m of tree i), D, D,, and d, is the distance (in m) between a focal tree i and
neighbor tree j. Cr was calculated for each tree in the original data set using Eq. 2-1;
values for Cr ranged from 0.0001 to 20.98 with a mean of 2.48 (n=27,079). Cr tends to
decline with dbh because big trees are less likely to be overtopped. Although different
species experience and respond to competition differently, no data were available to
calibrate a species-group specific model as was done by Arets (2005) for a version of
SYMFOR for Guyanese forests. Thus, for this model, I assumed that the relationship
between Ct and dbh would be the same for all species groups. I modeled the dependence
of Cr on dbh using non-linear regression (SAS 9.0) as a negative exponential function:
Ct = a, e(ay-D) [2-2]
The estimated parameter values were: ao = 28.0 and al = -0. 115; the associated R2
was 0.82. A dbh-independent competition index (C) was obtained--by subtracting Ct
from Ct--and used to model growth (see below). By using C, which is independent of
dbh, both dbh and competition could be included in the growth model.
Growth. Growth (i.e., dbh increment) was modeled for each species group using
non-linear regression (SAS 9.0) as a function of each tree's dbh, its competition index,
and whether or not it carried lianas:
I= [D(poL +1)- (p, + p, e(p3D)] I4C +p5, [2-3]
where I = estimated annual dbh increment (in cm/yr), po- ps are estimated parameters for
each species group, C is the diameter-independent competition index, and L is the liana
infestation class (0=no lianas; 1 = lianas). Equation 2-3 is similar to growth functions
used in previous SYMFOR models (e.g., Phillips 2004b).
The justification for including lianas, despite lack of data on their dynamics, was
that 60% of the trees in the forest were liana-infested. In addition, these trees grew
substantially slower than liana-free trees: mean growth in mm/yr for trees with lianas was
3.010.0 (n=14,872) vs. 4.910.1 (n=9,797) for liana-free trees.
The parameters were calibrated from the averages of valid growth records between
census year 0 and census year 2. 'Valid growth records' were those within 4 standard
deviations of the mean growth rate for the species after removal of extreme outliers (e.g.,
> 7.5 cm/yr or < -1 cm/yr). This process of identifying valid growth records for analysis
is similar to that used by Condit et al. (2004). After selecting only those trees with valid
growth records, the final calibration dataset (n=19,634) still contained some individuals
(n = 1,554) with negative growth. Negative growth observations result from uncertainties
in the dbh measurement process and should be included in the dataset because they're
equally as valid as overly positive growth observations (Condit et al. 2004). Thus, all
individuals in the final calibration dataset were included in the regression model above.
(Eq. 2-3), which yielded positive growth values for all diameters (Figure 2-2).
Considering the large variation in the growth data (Figure 2-2) and the variation in
competition experienced by different trees, it was no surprise that the fitted model only
explained 4.8 to 11.3% of the overall variance (Table 2-2). These R2 ValUeS are similar to
those reported for growth models in other tropical forests (e.g., Gourlet-Fleury and
Houlier 2000, Phillips et al. 2004b, Arets 2005). The fitted model explains a higher
proportion of the variance (range: 26 to 86%, Table 2-2) in the observed mean growth
rate for each dbh class. This second R2 was computed as follows:
R = ,[2-4]
where for each dbh class and species group, I, is the predicted growth rate, Io is the
observed growth rate, and I;, is the overall mean growth rate for each species group.
Mean growth rates predicted by the model for liana-free trees ranged from
23 mm/yr for understory species to 92 mm/yr for intermediate-lived pioneers. Predicted
mean growth rates for liana-infested trees were, on average, 41% lower than for liana-
free trees, which was similar to the 43% average slower growth observed in the data.
The calculated regressions yielded parameter values that caused growth rates to
continue increasing with dbh for the two shade-tolerant species groups (Table 2-2, Figure
2-2) due to a relative lack of observations at large dbh. This result meant that species
would attain unrealistically large sizes. To ensure that tree diameters remained within the
range observed in the data, I added a simple function to make growth constant for the
largest trees (D > D95, i.e., greater than the 95th percentile of the dbh distribution) in each
species group by liana class (L):
As5 = p6L p7, [2-5]
where 195 is the mean dbh increment (cm/yr); and ps and p7 are parameters calibrated for
each species group depending on the mean growth rate of each species group at D95.
Recruitment. Lacking data to model regeneration, I chose to model recruitment
(i.e., ingrowth) using the same approach implemented in most previous SYMFOR models
(Phillips et al. 2003, 2004b, but see van Ulft 2004). Because the LTSRP in La Chonta
had no records for trees smaller than 10 cm dbh, I modeled recruitment into the 10 cm
dbh class. Although ontogenetic shifts in tree growth rates are likely, I had to assume
that the growth rate of trees at 10 cm dbh was similar to the growth rate for smaller trees
because I lacked growth data for smaller trees.
The recruitment model simulates ingrowth of new trees into the 10 cm diameter
class annually as a function of the predicted growth rate (i.e., Eq. 2-3) of a 10 cm tree of
each species group in each 10 m x 10 m grid square in the simulated stand. SYMFOR
calculates the predicted growth rates for potential new recruits (=10 cm dbh) by species
group based on the average competition index for each corresponding grid square. The
predicted growth rates, in turn, determine the probability of ingrowth.
I considered several ingrowth probability models, but chose an exponential
function because it was the only model that did not cause extirpation of any of the species
groups in 60-year test simulations. I assumed that no species group would be completely
eliminated from La Chonta during such a short period of time. A second justification for
using this model was that functions with similar form were used in previous SYMFOR
models (Phillips et al. 2003, van Ulft 2004, Arets 2005). I calibrated the selected
function with a weighted non-linear regression fitted to the ingrowth data (IBIF 2004):
F, = z wm),[2-6]
where for each species group j, F, is the probability of recruitment, ii and it are estimated
parameters, and Is, is the mean growth rate of 11 growth rate classes (corresponding to
different levels of competition) for each species group. The weighting factor was the
number of grid squares with recruits.
The fitted model explained 23.4 to 91.3% of the variance for 6 of the 8 species
groups (Table 2-3), but explained only 12.2% of the variance for the intermediate-lived
pioneers and only 0.9% of the variance for the colonizing pioneers (Table 2-3, Figure
2-3). The recruitment probabilities predicted by the model for most species groups
ranged from 0.1% per year (at high competition for dry forest species) to 7.5% per year
(at low competition for slow-growing shade-tolerant species). Recruitment rates for the
dry forest and light-demanding species were an order of magnitude lower than rates for
the other species groups.
In this and most previous versions of SY1VFOR, it is assumed that disturbances
(treefall gaps, skid trails, scarified areas, and liana tangles) kill advanced regeneration in
the affected grid squares. Recruitment only occurs in these grid squares once the age of
the disturbance equals the time required for a newly established seedling of each species
group to reach 10 cm dbh. Obviously, many factors influence this 'ingrowth time' (T,),
and ideally T, would be estimated from actual data from all stages of regeneration.
Lacking these data, I estimated T, based on the maximum growth rates of juvenile trees
(dbh < 20 cm) in the dataset (Table 2-3).
To simplify the model, recruits were assumed to be liana free. Upon recruitment,
the annual probability that a tree will become liana infested was based on its species
group and the liana load in its neighborhood (see section on lianas below).
SY1VFOR assigns a utility group to each new recruit to indicate whether it is
commercial, potentially commercial, or non-commercial, and also whether its dbh must
be 50 cm or 70 cm before it may be harvested. For each species group, the probability a
recruit will be assigned a particular utility group equals the proportional representation of
that utility group in the input data (see Management Models and Tables 2-13 and 2-14 for
Natural mortality. Previous SYMFOR models have calibrated mortality based on
the assumption that un-logged forest is at dynamic equilibrium (Phillips et al. 2003).
This approach assumes that the number of tree deaths necessary to maintain a stable dbh
distribution equals the mortality rate for each dbh class (Phillips et al. 2003, 2004b).
Given the structure and composition of the forest in La Chonta, it is unlikely that it is at
dynamic equilibrium and therefore modeling mortality to mimic that state seems
unjustified. Instead, I modeled mortality as a function of each tree's dbh, species group,
and liana class. Because there was no reliable way to distinguish 'natural' mortality from
logging related mortality, I restricted the dataset to the control treatment. Of the 6,618
trees in this restricted dataset, 273 died 'naturally' over a period of 1.9 years yielding an
annual mortality rate of 2.2%. Although using only a subset of the database further
reduced the limited data available for calibration of a mortality model, using the data
from the logging treatments probably would have biased 'natural' mortality rates upward.
I fit a logistic regression (SAS 9.0) to the data with tree status (live or dead) as the
dependent variable and species group, dbh, and liana class (L) as predictors. Of the
possible interactions among these factors, I found that only dbh x liana class was
significant. Thus, the resulting model for the linear predictor of the logistic regression
I Mi nt nzaD nas3L m4D L, [ 2-7]
where nr -n?4 are parameters estimated by the regression (Table 2-4). I obtained the
annual probability of mortality (M)1 by transforming My from the logit scale and
accounting for the time interval between censuses as suggested by Alder (1995):
M= e1- MMi [2-8]
where t = 1.9 years (i.e., the mean interval between the first and third censuses for the
The goodness of fit of a logistic regression model is indicated by the residual
deviance, degrees of freedom, and scaled residual deviance (which equals the deviance
divided by degrees of freedom). The residual deviance was 2, 193 for 6,607 degrees of
freedom yielding a scaled residual deviance of 0.33 for the fitted model.
The average annual mortality rates predicted by the model for liana-free trees
ranged from 1.9% for slow-growing shade-tolerant species to 7.8% for colonizing
pioneers, and for liana-infested trees ranged from 3.1 to 13.1% for the same species
groups respectively (Figure 2-4). Liana-infested trees had predicted mortality rates 61%
(range: 30-105%) higher than liana-free trees across all species groups and diameters
In previous SYMFOR models, it was necessary to introduce a supplemental
mortality function to ensure that large trees would eventually die. I introduced a similar
function to calculate the annual mortality (M') for trees of each species group and liana
class where D > D95:
M'=" n?5 ; [2-9]
where for each species group i, D95, is 95th percentile of the dbh distribution, D;;;m is the
maximum observed dbh, mi, is the ratio of the wood density of the colonizing pioneers to
the wood density of each of the other species groups, and naps, is the probability of
mortality at D951. The parameters for m95, were determined by
naps = nzeL n z-, [2-10]
where mn and nz- are parameters (Table 2-5) calculated from the mean value obtained
from Eq. 2-8 for trees of each species group and liana class with D > D95. The parameter
naS is a user-defined value that I varied from naps, to 1 during model evaluation. Annual
mortality was constant (i.e., it equaled nass) with no = naps. The amount of additional
mortality beyond nass was a function of naS; M~' increased as no approached 1. Thus, for
simplicity later in this paper, I refer to naS as the 'supplemental mortality parameter'. In
initial simulations, it was apparent that the colonizing pioneers especially were growing
to unrealistic sizes. The purpose of the wood density ratio parameter (0i, ) was to ensure
that colonizing pioneers and other species groups with low wood densities would be more
strongly affected by this supplemental mortality function than species groups with higher
Damage mortality. I lacked data to determine what proportion of tree deaths were
due to natural tree falls or falling branches vs. Other causes of mortality (e.g., disease),
which meant that the parameter estimates for Equations 2-7 to 2-10 included both causes
of mortality. Because SYMFOR accounts for damage-related mortality separately (see
Phillips et al. 2003 for details), it was necessary to adjust the value ofM~ and M~' in
Equations 2-8 and 2-9 downward so that they represented only non-damage caused
mortality. Specifically, I multiplied M~(or M)1 by (1-md), where md is a parameter that I
varied during model evaluation from 0.03 to 0.20.
In SYMFOR, trees can either die standing or falling as determined by the
parameter Pf (the probability of falling). If they die standing, SYMFOR assumes that
they do not kill other trees. If they fall when they die, they may kill trees in their path
with probability Pd. The damage adjustment parameter, md, is the product of these
probabilities (Pf .Pd). I lacked data to calibrate these parameters; hence, their values
were best estimates subsequently tested in the sensitivity analysis (see below).
Liana class and assigning a liana class to recruits. Lianas are common in La
Chonta and a significant challenge for managers (Alvira et al. 2004). Lacking adequate
data on liana dynamics, however, the simulation of lianas had to be approximated from
Hield observations, the literature, and logic. Thus, lianas are simulated either as attributes
of trees by a liana class variable (0=no lianas; 1=1ianas), or as infestations of tree fall or
felling gaps (i.e., liana tangles), which delay recruitment for 9 to 62 years depending on
the species group.
The liana class, L, of existing trees in the stand was obtained from Hield data. I
assumed that L never changes from 1 to 0 except by liana cutting (a silvicultural
treatment), whereas trees without lianas (whether existing at the beginning of the
simulation or new recruits) can become infested with probability = 0.0475 annually until
60% of the trees in the stand are liana infested. I estimated the annual probability of
infestation and proportional limit from data available on liana class in La Chonta (IBIF,
unpublished data; see Appendix B for details).
All recruits enter the stand without lianas, but depending on their species group and
neighborhood (i.e., within 20 m of the recruit), they may become liana infested. The
probability that a recruit might become liana infested is defined by a single parameter that
varies for each species group according to the proportion of liana-infested neighbors.
The parameter values were obtained from the proportion of liana-infested juvenile trees
(D < 15 cm) in the data for each species group in each of four neighborhood infestation
classes (Table 2-6).
Liana tangles. The many liana infested gaps and other areas of La Chonta without
advanced tree regeneration constitute a maj or challenge for managers because they
reduce the future timber value of the forest stand. In general, the recruitment of
commercial trees is delayed in these low productivity areas due to high liana densities
(Schnitzer et al. 2000) as well as monodominant patches of clonal herbs (e.g., Heliconia
spp.) or treelets, especially chocolatillo (Erythrochiton fallax). Lacking data on the
dynamics of any of these areas of arrested succession, I assumed that they would all delay
tree recruitment in a similar way, and hence considered them all to be 'liana tangles' in
which a tree could not recruit until the age of the liana tangle equaled the ingrowth time
(T,) estimated for each species group (Table 2-3). At that time, the basic recruitment
rules described above applied.
I spatially represented these liana tangles in SY1VFOR through their specific
association with tree fall gaps whether inferred from the input data or created during
simulations. To create liana tangles from the input data, I instructed SY1VFOR to
randomly sample 10 m x 10 m areas without trees > 10 cm dbh. Based on the
observation that half of the 1-year old felling gaps I evaluated had 50% or more liana
cover (GB unpublished data, Alvira et al. 2004), SYMFOR then randomly creates a liana
tangle in half of those areas. Finally, I assumed that liana tangles would remain as such
for up to 20 years, and therefore assigned an age (1-20 years) to each liana tangle from
an even-aged distribution.
SYMFOR creates liana tangles in simulated treefall or felling gaps based on the
liana class of the falling tree and the proportion of liana-infested neighbors (i.e., trees
within 20 m of the falling tree). Without reliable data to estimate parameters for this
process, I assigned the probability of a gap becoming liana infested based on a series of
conditions (Table 2-7).
Other functions. SYMFOR uses several other functions to calculate key attributes
of trees during the simulations, including height, volume, and biomass.
Tree height is calculated based on the allometric relationship with dbh described by
H= ~L'hmax -e-1D [2-11]
where h;;;a, hi, and hz are parameters estimated by non-linear regression for each species
group. The data used to estimate these parameters comprised dbh and height
measurements for 1,174 trees of 59 species in La Chonta (Poorter et al. In press). Height
was measured with a clinometer (height > 8 m) or with an extension pole (height < 8 m).
Although most trees in this dataset were not liana infested, I used the main LTSRP
database to compare heights for trees with and without lianas. I found no significant
difference, and hence decided to use the estimated parameter values (Table 2-8) for all
trees regardless of liana class. SYMFOR uses tree height to calculate volume (see below)
as well as the size of gaps caused by falling or felled trees.
Commercial timber volume (V) is calculated from a tree's basal area (G, calculated
from dbh), a form factor (f), and its merchantable height, which is estimated by a
parameter called "crown-point" (c,) in other SYMFOR models (Phillips et al. 2003):
V= G -cp f [2-12]
The form factor, J was assumed to be 0.65 for all species groups and was taken from an
analysis of a large database of trees in the La Chonta region (Dauber et al. 2000). The
crown-point for each species group was calculated as:
cp =H -a, [2-13]
where H (height) is calculated as described above, and a is a species group specific
parameter that represents the ratio of merchantable height to total tree height. The value
of a for each species group was approximated using the mean log length from harvested
trees (Table 2-9).
The biomass function in SYMFOR calculates the aboveground biomass (B) of each
tree (kg) annually based on its dbh (D), wood density (p), and total height (H) using a
function calibrated by Chave et al. (2005) for moist tropical forest stands:
B = 0.0509 pD2H [2-14]
The values for p (Table 2-10) were obtained both from La Chonta (L. Poorter,
unpublished data) and the literature (Chudnoff 1984).
The necromass function in SYMFOR calculates the annual change in mass of dead
trees, which is an estimate of the mass of woody debris. A dead tree's biomass is
reduced annually by a decay function (Kd) aCCOrding to its dbh, mean wood density ( p)
for each species group, and parameters estimated by Chambers et al. (2000) for trees in
the southwest Amazon:
Kd = (1.104 0.670p 0. 163 log(D)) [2-15]
The simulated harvests in SYMFOR target individual trees that meet selection
criteria including size, stem quality, and commercial status. A utility group variable
defines the commercial status of each tree within each species group. The input data
contain the initial utility group information. Each tree was assigned to 1 of 4 utility
groups depending on actual practice in La Chonta: valuable species typically harvested
in the normal treatment with minimum dbh for felling limit (MDF) = 50 cm; 2=valuable
species with MDF = 70 cm; 3=species not typically harvested by La Chonta, but
considered commercial or potentially commercial (and therefore harvested only in the
intensive management treatment, MDF = 50 cm); and 4= non-commercial trees (Table
2-11). During the simulations, new recruits are assigned to a utility group based on the
proportional composition of utility groups within each species group in the input data
I adapted existing SYMFOR management models and functions (see Phillips et al.
2003, 2004b), or created new ones, to mimic the silvicultural treatments (i.e., cutting
cycle, selective felling, skidding, poison-girdling, liana cutting, and soil scarification)
applied in La Chonta. For example, depending on the size and stem quality criteria for
each utility group, SYMFOR simulates liana cutting by changing the liana class of
infested trees from 1 to 0. Depending on dbh and proximity to future crop trees,
SYMFOR also kills (via poison-girdling) non-commercial competitors (utility group 4).
I created a new model to simulate the soil scarifieation treatment in La Chonta,
which cleared vegetation down to mineral soil in selected felling gaps at the time of
tree-fall. Gaps were considered eligible for scarifieation if:
* No advanced regeneration of commercial species was present in the gap; and,
* At least 1 mature (dbh > 30 cm) commercial individual was present within 20 m of
If these criteria are met, the scarifieation function kills all non-commercial trees in the
The effect of scarifieation on recruitment is simulated in the same way as liana
tangles, skid trails, and other kinds of damage: once the age of the scarified gap equals
the ingrowth time required for each species group (Table 2-3), trees of that group may
recruit. Evidence from La Chonta (Fredericksen and Pariona 2002, M. Pefia-Claros
unpublished data), however, indicates that scarification enhances the regeneration of
some species and slows the regeneration of others. To account for this differential effect,
the probability of recruitment function (Eq. 2-6) is multiplied by a species group specific
parameter (Es; Table 2-13). Depending on the species group, Es increases, decreases, or
does not change the probability of recruitment. The values for Es are the mean relative
densities of each species group in La Chonta, which were calculated by M. Pefia-Claros
(unpublished data) as the ratio of mean seedling density in the scarified gaps (n=81) vs.
the non-scarified gaps (n=78).
Previous SYMFOR models were evaluated for internal and logical consistency
using methods similar to those described by Vanclay (1994). Most of these models
(except van Ulft 2004 and Arets 2005) assumed that the forests they were simulating
were at dynamic equilibrium. In most cases, the calibrated parameters in the functions
for growth, mortality, and recruitment had to be adjusted so that the simulated unlogged
forests behaved as if they were at this hypothetical steady state. I decided against this
approach because it is unlikely that La Chonta is at dynamic equilibrium, and because it
will be impossible to test assumption from field data for many years. In addition, even
apparently undisturbed Amazonian forests have become more dynamic (cf. Arets
2005)--with increases in biomass and turnover rates--in recent decades (Baker et al.
2004, Lewis et al. 2004a, Lewis et al. 2004b, Phillips et al. 2004a).
Although I did not tune the model to behave as if it were at dynamic equilibrium, I
nevertheless adjusted parameters that were either not calibrated from data (i.e., the
probability a dying tree will fall; the probability a falling dead tree will kill a neighbor;
the minimum dbh below which a dying tree will remain standing; and the liana
infestation rates) or that had a high degree of uncertainty (i.e., the supplemental mortality
for large trees; the ingrowth time; and the effect of scarification on recruitment rates). I
estimated values for these parameters by evaluating the response of several key
variables--total number of trees (N) and basal area (G), N and G by species group, and
proportion of Nthat were liana-infested (NZ)--during 60-yr simulations repeated 10 times
with input data from 15 of the 32 1-ha plots. I chose values that were consistent with the
most reasonable expectations considering the short simulation period: no species group
extinctions; liana-infested trees account for 54-60% of total N; and, tree density, basal
area, and maximum dbh for each species group remain within the range of variation
observed in the data.
Lacking long-term, independent data against which the simulation results could be
compared, meaningful model testing was limited to a statistical comparison of the data
used to calibrate the model (LTSRP blocks 2 and 3) with an independent dataset from the
same forest (LTSRP block 1). Although less rigorous a test of a model than validation of
the model's proj sections with time-series data, this approach nevertheless tests whether the
data used to calibrate the model are idiosyncratic or representative of the forest the model
aims to simulate. If the calibration data do represent the forest (i.e., the test and
calibration datasets are similar), then it can be concluded that the model will behave as
well for test data from different areas of the same forest as the model behaves with the
Thus, I tested for differences between the calibration and test datasets in terms of:
* Growth rates (mean and 95th percentile of the distribution, or p95) by species group
with a Wilcoxon 2-sample test;
* Number of new trees by species group as a proportion of initial tree densities using
Poisson regression with a log link function; and,
* The proportion of total N and total basal area by species group, and the proportion
of all trees that were liana-infested using general linear models after
arcsine-square-root transforming the proportions.
I obtained means and / or proportions for these variables for each of the 1-ha plots in the
experimental blocks (n=32 for calibration dataset; n=16 for test dataset).
Model sensitivity to parameters not calibrated from data
Assessing the sensitivity of the model's behavior to changes in the values of
parameters estimated from the calibration data was beyond the scope of this chapter.
However, I wanted to determine how sensitive the model was to the variables for which
the calibration was either highly uncertain (due to limited data) or simply based on best
estimates (see Table 2-14). The tested parameters were in functions that determined:
* Supplemental mortality of large trees;
* Liana infestation rate;
* Damage related mortality; and,
* Recruitment after silvicultural treatments.
The damage parameters included the minimum dbh of a falling dead tree, the probability
that a dead tree will fall, and the probability that a falling dead tree will kill neighbors in
Because I could not determine the uncertainty associated with the tested
parameters, I simply increased or decreased (by 50% of the baseline) each value during
sensitivity analyses (Table 2-14). I also ran simulations to test the effect of additional
deviations in the values of the parameters for supplemental mortality (mS) and liana
infestation rate (Table 2-14).
Rather than set the supplemental mortality parameter (mS) to the same value for all
species groups in the baseline scenario, I set it higher for the colonizing pioneers to
ensure that they would die before reaching unrealistic diameters. My justification for this
decision was the observation that colonizing pioneers especially grew to unrealistic sizes
in preliminary simulations with values of nas lower than 0.2. In contrast, the diameters of
trees in the other species groups appeared to stay within the range observed in the data in
60-year simulations with naS = 0.05. Thus, I used these values for naS in the baseline
scenario of the sensitivity analysis.
Overall, I ran 17 simulations with the control (no logging) treatment for the
parameters listed in Table 2-14. I ran an additional 6 simulations for the two recruitment
parameters (T, and Es) that could be influenced by silviculture; I used the intensive
management treatment for these simulations. The simulations for liana infestation rate
and the two recruitment parameters were run for 60 years. All other simulations were run
for 120 years. The input data for the simulations consisted of 15 of the 32 available 1-ha
plots from the calibration dataset. I repeated simulations of each of the plots 10 times.
I evaluated the model's sensitivity in terms of the response variables mentioned
above: total N and G, N and G by species group, and Nz. For liana infestation rate, I also
examined the number of years required to reach the stand liana threshold (60%). I
calculated sensitivity (S) according to the formula provided by Vanclay (1994):
S = ,[2-16]
where 8v/v is the relative change in the result variable and 89/ p is the relative change
in the parameter value.
Comparison of treatment impacts
I compared the management treatment impacts using the same response variables
mentioned above: total N and G as well as N and G by species group, and Nz. Using the
parameter set indicated in Tables 2-1 to 2-14, I ran simulations of each management
treatment for 62 years, which was long enough to evaluate the impacts of two cutting
cycles and to estimate potential yield from a third harvest entry. I used 15 of the 32
available 1-ha plots from the calibration dataset for input and repeated simulations for
each plot 15 times.
Model Testing: Comparing Test vs. Calibration Data
Growth rates (both means and p95) WeTO Similar between the calibration and test
datasets for all species groups except the dry forest species. For dry forest species, mean
growth was higher in the calibration dataset than in the test dataset (Z15, 31 = -2.1;
P = 0.04). The 95th percentile of the growth rate distribution, however, did not differ for
this species group between the datasets.
The Poisson regression indicated that the calibration and test datasets differed in
per capital recruitment (X1 = 9.5; P = 0.002). The per capital recruitment of colonizing
pioneers (SG7), in particular, was greater in the calibration than in the test dataset (X1 =
21.5; P < 0.001). In contrast, per capital recruitment was on average 59% greater (range:
4-116%) in the test dataset than in the calibration dataset for all other species groups
(Table 2-15). The explanation for these differences is not immediately obvious but could
be attributable to various factors including, environmental differences among the
experimental blocks, differences in the way the management treatments were applied in
each block, and unusually high or low recruitment rates for the years sampled in one of
These results led me to run two test simulations, one with the recruitment
parameter values doubled for all species groups except the colonizing pioneers, and the
second with the recruitment rate halved for colonizing pioneers and left unchanged for all
other species groups. The outcome of the first test was unrealistic: total N and total G
increased by 40-50% over 60 years. The outcome of the second test, in contrast, seemed
reasonable: total N and total G increased only slightly over 60 years. Moreover, the
density and basal area of the colonizing pioneers remained fairly constant over 60 years,
whereas this species group doubled in density and basal area when the unadjusted (i.e.,
parameterized) recruitment rate was used in test simulations. This latter behavior of the
model was counterintuitive and inconsistent with successional theory, which predicts that
in the absence of disturbance, short-lived colonizing pioneers should not increase in
relative dominance. Thus, I used the adjusted parameter value for colonizing pioneers in
Forest structure comparison
Tree density, basal area, and dbh distribution. The calibration and test datasets
appeared similar upon visual inspection of their dbh distributions, but they differed in
terms of mean dbh and mean basal area. Specifically, the mean dbh in the calibration
dataset (n = 32) was 1 cm greater than in the test dataset (n = 16): 22.310.15 vs.
21.310. 17 cm (t46 = 2.9; P = 0.006; Satterthwaite method for unequal variances). The
95th percentile of the dbh distribution was also higher in the calibration (51.3 cm) than in
the test (45.6 cm) dataset. Consequently, basal area was greater in the calibration than in
the test dataset (21.110.7 vs. 19. 110.5 m2/ha; t46 = 1.94; P = 0.02; Satterthwaite method
for une ual variances Densities (#/ha) of trees 10 cm dbh in the two datasets were not
significantly different (calibration: 358.1f8.2; test: 384.2f9.1; t46 = -1.97; P = 0.06).
Although the mean dbh and basal area of the two datasets statistically differed, I
concluded that they were biologically similar because of the small magnitude of these
differences and because the overall tree densities were similar. This conclusion was
further supported by the following two tests.
Species group composition. Overall, the calibration and test datasets did not
differ in terms of species composition (F1,366 = 0.02; P = 0.9), although the understory
and long-lived pioneer species groups were more abundant in the calibration dataset than
in the test dataset (F7,366 = 176.4; P = 0.001; Table 2-16). Similarly, the proportion of
total basal area was similar in the two datasets (F1 = 0.34; P = 0.56).
Liana infestation. The proportion of trees that were liana infested when the plots
were established (i.e., census 1) was similar in the calibration and test datasets. The
proportion of all trees without lianas in the test dataset was 48% and in the calibration
dataset was 44% at census 1. At census 3, those proportions dropped to 43% (test) and
Summary. Despite differences between the test and calibration datasets in per
capital recruitment and some differences in forest structure, the results of the other tests
suggest that the datasets are similar. Thus, I concluded that the calibrated model would
perform equally well using any input data from La Chonta or similar forests.
Model Sensitivity to Parameters not Calibrated from Data
Mortality of large trees: total tree density and basal area
Total tree density (total N) was relatively insensitive (S = 0.01) to changes in the
supplemental mortality parameter (mS) within the range of values tested (mS = m9s to 0. 1
for non-colonizing pioneers and m9s to 0.3 for colonizing pioneers). Total N decreased,
as expected, with higher values of ms, but the change relative to the baseline was small
(Figure 2-5a). When ms was doubled for all species groups (except the colonizing
pioneers) total N differed by <1% after 120 years between the baseline (35913 trees/ha)
and test scenarios (35712 trees/ha). Without supplemental mortality (i.e., mS = m95), total
N increased by only 3% (S = 0.68).
Total basal area (G) was somewhat more sensitive to changes in mS than total N
(Figure 2-5b). Total G decreased by 1.6 m2/ha (7%; S = 0.14) from 24.010.2 in the
baseline to 22.210.2 m2/ha when mS was doubled (compare baseline vs. s2 and s5 in
Figure 2-5b). When mS = m9s, total G increased by 10% (S = 1.97, Figure 2-5b).
Mortality of large trees: N and G by species group
Tree density was unaffected when mS was doubled, and also remained similar to the
baseline for most species groups with mS = m95 (Table 2-17). This response variable was
most sensitive to setting mS = m95 in the colonizing pioneers (SG7), the long-lived
pioneers (SG2), and slow-growing shade-tolerant species (SG3). Tree density for the
colonizing pioneers increased by 12% (S=0.12), for the long-lived pioneers decreased by
8% (S=0.15), and for the slow-growing shade-tolerant species increased by 7% (S=0.15;
As expected, basal area was somewhat more sensitive to changes in mS for most
species groups, although for many groups the changes were quite modest (Table 2-17).
Doubling the value of mS caused less than a 10% change in basal area for all but three
species groups (Figure 2-6b): intermediate-lived pioneers (decreased 23%), colonizing
pioneers (increased 11%), and fast-growing shade-tolerants (decreased 18%).
The effect on G was greater for most species groups when mS = m95. The most
sensitive group was the fast-growing shade-tolerant species, which increased in basal area
by 26% with mS = m95. The intermediate-lived pioneers and colonizing pioneers both
increased in basal area by 20% and the dry forest species group increased by 19%.
Mortality of large trees: maximum dbh
Within the range of values of mS tested (m95 to 0. 1), maximum dbh at year 60 of the
simulation was virtually identical for the non-colonizing pioneer species groups. Thus, I
decided to keep the value of mS at 0.05 for the other simulations.
The calibrated value of liana change probability, which controls the annual rate of
liana infestation of liana-free trees, was 0.0475 (see Appendix B for calculation). None
of the tested response variables was sensitive to changes of up to 100% in the value of
this parameter. The time required for the proportion of trees with lianas to reach the
threshold value of 60% (from the mean value of 54% in the input data) depended on the
value of this parameter. That threshold was reached within 10 years with the baseline
value and within about 60 years with a value of 0.02. Because the impacts on tree
density and basal area were negligible, I used the baseline value (from Appendix B) in
Damage related mortality
Total tree density and basal area were relatively insensitive to changes of 50% in
the parameters controlling damage-related mortality. After 120 years, depending on
which parameter was altered, the deviation from the baseline (N= 360; G = 24 m2/ha)
was only 1-3% for total tree density (i.e., 10 trees) and 1-4% for total basal area
(i.e., 1 m2/ha). Tree density and basal area were also relatively insensitive to changes in
damage-related mortality at the species group level, though these response variables
increased or decreased by 6-9% for some species groups depending on which parameter
Tree density was sensitive to changes in these parameters principally for
slow-growing shade-tolerant species. Reducing the probability that a tree will fall when
it dies (Pf) by 50% caused an 8% increase in the number of trees per ha (from 9513 to
10314). Conversely, doubling Pf caused an 8% decrease in the number of individuals of
this species group.
Basal area was somewhat more sensitive to changes in damage-related parameters
than tree density for most species groups. Again, the slow-growing shade-tolerant
species group was the most sensitive. Reducing the probability that a dying tree will fall
by 50% caused a 7% increase in basal area in this species group (from 5.210.5 to
5.610.5 m2/ha) whereas doubling that probability caused a 9% decrease.
Total tree density and basal area were insensitive at year 60 to 50% changes in the
values of two species group specific parameters--ingrowth time (T,) and scarify effect
(Es)-that affect post-silviculture recruitment rates. These response variables were
similarly insensitive at the species group level with one exception. When the value of T,
of each species group was decreased by 50%, the basal area of the dry forest species
group decreased by 10%. When the value of T, for each species group was increased by
50%, basal area of the dry forest species group decreased by 15%. This species group
exhibited the poorest recruitment in the dataset.
In general, the model did not appear to be particularly sensitive to any of the
parameters that were calibrated with little or no data. The most sensitive response
variable tested was basal area, although it did not change by more than 26% under any of
the scenarios tested. Of the parameters tested, the model was most sensitive to the value
for supplemental mortality.
Comparison of Treatment Impacts
Harvest volumes, regardless of management treatment, were quite low by tropical
standards even in the first harvest entry. In the normal harvest treatment, only
2.410.4 trees/ha were felled in the first cutting cycle for a mean total volume of
10.311.7 m3/ha. In the intensive management treatment, 3.510.6 trees/ha were felled for
a mean total volume of 14.712.4 m3/ha. These volumes are similar to the actual harvest
volumes obtained in La Chonta: on average 2.3 trees/ha and 10.4 m3/ha were harvested
from the normal harvest treatment and 4.0 trees/ha and 14.4 m3/ha were harvested from
the intensive management treatment (IBIF, unpublished data).
Both the number of trees felled and volume harvested declined in the simulated 2nd
and 3rd harvest entries (Tables 2-18 to 2-21, Figure 2-7) regardless of management
treatment. In the normal harvest treatment, relative to the first cutting cycle, timber
yields were only 33% in the second harvest entry and 14% in the third harvest entry. In
the intensive management treatment, yields relative to the first harvest were 47% and
32% for the second and third harvest entries, respectively. While falling far short of
STY, the intensive management treatment yielded 105% more timber in the second
harvest entry (t28 = -3.51, P = 0.0015) and 231% more timber in the third harvest entry
(t28 = -4.70, P < 0.001) than did the normal harvest treatment (Tables 2-20 and 2-21,
Figure 2-7). These differences were attributable to both the greater number of species
harvested in the intensive management treatment (Tables 2-20 and 2-21) and the
silviculture applied in the intensive treatment. The applied silviculture apparently
increased commercial species' growth rates and recruitment compared to the other
treatments (Table 2-21, Figures 2-8 and 2-9). The potentially commercial species
comprised about 6% of the harvest volume in the first cutting cycle, and 15% and 28%,
respectively in the second and third cutting cycles of the intensive management treatment
Effects on overall stand structure
Total tree density. The simulated management treatment effects on tree density
(number of trees > 10 cm dbh / ha) were small, consistent with expectations, and similar
in magnitude and direction to short-term effects observed in the field. Tree density
decreased by 2.8% (from 350115 to 34015) over 60 years in the absence of harvesting or
additional silviculture (Figure 2-10a). After two harvest cycles (=60 years), tree densities
recovered to the same levels in response to the normal harvesting regime (32714 trees)
and the intensive management treatment (32614 trees) (Figure 2-10a). After the first
cutting cycle, tree density was reduced by 7.5% in the normal harvest treatment and 8.5%
in the intensive management. The post-harvest silviculture in the intensive management
treatment further reduced tree density by 1.5% (Figure 2-10a). Regardless of
management intensity, after each of the first two cutting cycles, tree densities recovered
to within 3-4% of those observed in the control treatment within 30 years (Figure 2-10a).
Tree densities recovered faster in response to the intensive management treatment than in
response to the normal harvest treatment implying better recruitment or lower mortality
or both (Figure 2-10a).
Total basal area. The effects of the management treatments on basal area were
slightly greater than the effects on tree density, which is logical considering that the
largest trees were harvested. Plot basal area increased by 8.1% (from 20.211.0 to
21.910.3) over 60 years in the absence of logging (Figure 2-10b). After the first cutting
cycle, basal area was reduced by 15.4% (13.4% from felling and 2.0% from additional
silviculture) in the intensive management treatment and by 11.1% in the normal harvest
treatment. The second harvest reduced basal area by 9.8% (including silviculture) in the
intensive management treatment and by 4.2% in the normal harvest treatment. By year
60 (after 2 cutting cycles) total basal area recovered to the same level as at year 0 for both
harvest treatments, but was 8% less than in the control treatment (Figure 2-10b).
Although more basal area was removed in the intensive management treatment than in
the normal harvest treatment, the faster growth rates for commercial species afforded by
the applied silviculture (Figure 2-9) apparently contributed to faster basal area recovery
Effects on species composition
Tree density. In the absence of logging or other silvicultural treatments, tree
densities did not fluctuate dramatically for most species groups (Figure 2-11a). The
greatest changes during the 60-year simulation were observed in the long-lived pioneers
(29% increase), intermediate-lived pioneers (16% increase) and dry forest species
(14% decrease). The change in abundance of other species groups over the 60-year
period was less than 10%.
The harvest treatments amplified the patterns observed in the control plots during
the 60-year period for most species groups. The most extreme examples of this
observation were in the understory and slow-growing shade-tolerant canopy species. In
both cases, the reduction in tree density observed in the control treatment was doubled in
the normal harvest treatment and nearly tripled in the intensive management treatment.
Most other changes in the patterns observed in the control treatment were more moderate.
The main differences observed with intensive management were greater increases in the
long-lived pioneers and greater decreases in the slow-growing shade-tolerant canopy
species (Figure 2-11a).
Basal area. Changes in basal area at the species-group level during the 60-year
simulation period were greater than those observed for tree density in the control as well
as in the harvest treatments (Figure 2-11b). In the control treatment, the greatest
increases in basal area were in the fast-growing shade-tolerant species (105% increase)
and intermediate-lived pioneers (83% increase). The basal area of the slow-growing
shade-tolerant and understory species also increased by 23% and 11%, respectively. In
contrast, the basal area of the light-demanding species (many of which are commercial)
decreased by 3 5% and the basal area of the dry forest species (some of which are
commercial) decreased by 18% over 60 years in the control treatment. Both of these
species groups exhibited poor recruitment and slow growth in La Chonta.
The impact of the harvest treatments on basal area varied for each species group.
As was the case in the control treatment, the greatest increases in basal area for both the
normal harvest and intensive management treatments occurred in the fast-growing
shade-tolerant species and intermediate-lived pioneers. The percentage increase in basal
area, however, was less for both species groups in the intensive management treatment
than in the normal harvest treatment (Figure 2-11Ib). Basal area of the intermediate-lived
pioneers increased by 93% in the normal harvest treatment and by only 80% in the
The silviculture applied in the intensive management treatment appeared to benefit
the principal commercial timber species groups. The intensive management treatment
increased the basal area of the long-lived pioneers by 3% over 60 years whereas basal
area of this species group decreased by 6% in the normal treatment. Also, the decrease in
basal area of the light demanding species caused by intensive management was similar to
the decrease observed in the control treatment, whereas the basal area of this species
group decreased a further 6% in the normal harvest treatment (Figure 2-11b).
Silvicultural effects of the intensive management treatment
The intensive management treatment simulated three silvicultural treatments: liana
cutting on future crop trees (FCTs), killing competitors near FCTs, and soil scarification
in felling gaps. In the simulations, lianas were removed from 23-26 crop and FCTs per
ha, and 6 FCT were released from neighboring competitors (Table 2-22). Depending on
the cutting cycle, 37-41% of felling gaps were scarified (Table 2-22).
Liana cutting and liana infestation. The liana cutting treatment applied
principally to FCTs in the intensive management treatment decreased the proportion of
trees with lianas (Figure 2-12), which, together with the liberation and scarification
treatments (Table 2-22), benefited the commercial tree species. During the first cutting
cycle, the liana cutting treatment decreased the proportion of all trees that were
liana-infested by 20% (from 54% to 43%) and also decreased the proportion of
liana-infested basal area by 42% (from 64% to 37%).
Effects on growth rates. Mean growth rates (pooled across species groups,
repetitions, plots, and simulation years) were about the same regardless of treatment.
However, growth rates were twice as great for liana-free trees as they were for liana-
infested trees (50.310.7 mm/yr vs. 26.610.1 mm/yr in the control treatment). The
liana-cutting treatment apparently increased growth rates of the commercial species
sufficiently to allow faster post-harvest basal area recovery of these species than in the
normal harvest treatment (Figures 2-8 to 2-10). Immediately after the harvest and
additional silviculture, commercial species grew 15-16% faster in the intensive
management treatment than in the normal harvest treatment and 19-23% faster than in
the control treatment. This effect persisted for about 10 years for the commercial species,
after which time growth rates among the management treatments were more similar.
Effectiveness of the Applied Management Treatments in Achieving STY
The simulated silviculture applied in the intensive management treatment clearly
benefited the commercial species as demonstrated by faster growth rates, better
recruitment, and faster post-harvest basal area recovery relative to the normal harvest
treatment. Consequently, compared to the normal harvest treatment timber yields
obtained under intensive management were 105% greater in the 2nd harvest and
23 1% greater in the 3rd cutting cycle. Although neither of the harvest treatments applied
in La Chonta came close to STY (at best the recoverable volume was only 47%), the
intensive management treatment came closer. The timber yields obtained in the 2nd and
3rd harvests as a percentage of the volume harvested in the first cutting cycle were 15 to
20% greater in the intensive management treatment than in the normal harvest treatment.
These proj sections are similar to estimates of recoverable volumes (i.e., 28%)
reported by Dauber et al. (2005) for a second harvest in forests like La Chonta under a
management regime similar to the normal harvest treatment applied in this study except
using a 20-year cutting cycle. Under their optimum scenario (which assumes growth
rates of free-to-grow, liana-free trees of good form), however, recoverable volumes
approached 90% in the second cut assuming a 30-year cutting cycle. The intensive
management treatment results move toward this estimate, but do not equal it because not
all future crop trees were free to grow (i.e., only 6 trees/ha liberated) or remained
liana-free for an entire cutting cycle, which means maximum growth rates and low
mortality rates did not necessarily apply in all cases.
The proj sections from this study are also consistent with estimates of potential
volume recovery from other Neotropical forests. In simulations of different management
regimes in humid forests in Venezuela, Kammesheidt et al. (2001) estimated that 60-year
cutting cycles would be required to achieve sustained yields of 30-60 m3/ha depending
on the minimum felling diameter and whether conventional or reduced-impact logging
were applied. In French Guiana, Gourlet-Fleury et al. (2005) estimated that only 60% of
the initial stock would recover after logging with a DMC of 60 cm and a cutting cycle of
40 years. In contrast, in logged humid forests in the Eastern Amazon, Silva et al. (1995)
found that the total basal area was about 75% of that in a comparable unlogged forest
13 years after logging with no additional silviculture. However, they also reported that
annual commercial volume increments were only 0.8-1.8 m3 ha-l yr- depending on
whether new commercial species were included. Since their study was conducted after
the removal of 75 m3/ha, the implication is that cutting cycles would need to be about
75 years to achieve sustained yields of the commercial species at the time of the first cut.
Prospects for Achieving STY in La Chonta
One of the reasons that the intensive management treatment came closer to
achieving STY than the normal harvest treatment is because additional (mostly
shade-tolerant) species were eligible for harvest. Although the silviculture applied in the
intensive treatment yielded obvious benefits--the volume of commercial species
harvested in the 2nd cut was 74% greater under intensive management than under the
normal harvest treatment--it is also important to recognize that including the potentially
commercial species elevated the total yields in all three cutting cycles. In addition, the
proportion of the total volume harvested comprising potentially commercial species
(under intensive management) increased from only 6% in the first cut to 15% in the
2nd cut and 28% in the 3rd cut suggesting that additional silviculture will be needed to
promote the growth and recruitment of the more light-demanding commercial species.
Thus, although the applied silviculture helped move management toward STY,
considerable work remains to achieve it in these forests if cutting cycles are to remain as
short as they are at present or unless markets are found for the potentially commercial
Considering the relatively slow growth rates and poor recruitment of the
commercial species and the short cutting cycle, it is no surprise that recoverable volumes
in the second and third harvests were so much lower than in the first. The first harvest of
previously unlogged tropical forests includes a subsidy from nature and the second
harvest typically comprises trees that were too small to be cut at the time of the first
harvest (Dawkins and Philip 1998). But, this point assumes that there is an adequate
stocking of commercial trees that will reach the minimum diameter for felling (MDF)
within the cutting cycle period.
Dauber et al. (2005) indicated that of the lowland forest types in Bolivia, those in
the La Chonta region have the best stocking of future crop trees capable of reaching the
DMC within 30 years under optimal conditions. Under such conditions, they found that
90% of the volume harvested in the first cut could be recovered in the second. The data
from La Chonta, however, are less optimistic.
In La Chonta, before logging, commercial species comprised only 15% of all trees
(and 1 1% of the total basal area) capable of attaining the MDF (50 cm) within 30 years
(i.e., dbh > 20 cm) under optimal conditions. Furthermore, they comprised only 9% of
recruits (dbh < 20 cm), which under similarly ideal conditions could be considered future
crop trees for the 3rd cut. If potentially commercial species are included, those
percentages increase to 50% of potential FCTs (and 75% of basal area) for the 2nd cut
and 32% of trees (and 36% of basal area) for the 3rd cut. These percentages improved
only slightly under intensive management.
An obvious research question that emerges from this study is the extent to which
more intensive management strategies would come closer to achieving STY. The model
described in this study could be used to evaluate the effectiveness of various alternatives.
Although the simulation of plausible alternative strategies constitutes the subj ect of a
separate study, it is worth noting a few possibilities. In addition to testing how long
cutting cycles ought to be to achieve STY under the current federally mandated
guidelines (i.e., for seed trees and MDF), different aspects of the intensive management
treatment could be tested alone and at different intensities to determine their potential
benefits. For example, the liana cutting treatment could be applied every 15 years instead
of at each harvest entry, or it could be applied to more trees, or both. Similar experiments
could be simulated for the poison-girdling and scarification treatments. Finally, it would
be worthwhile to simulate management regimes that alter the MDF, as was recently done
by Arets (2005) for Guyanese forests.
Management Treatment Impacts on Forest Structure and Composition
Total tree density and basal area
In the absence of logging or additional silviculture, total tree density slightly
decreased and total basal area slightly increased over 60 years, which leads to the
inference that the basal area increase is mainly attributable to growth of existing trees.
The fact that tree density did not substantially increase is surprising at first glance
considering that the tree density (> 10 cm dbh) in La Chonta is at the lower end of the
range (371-768 trees / ha) found in other neotropical forests by Dewalt and Chave
(2004). On the other hand, the fact that 73% of trees (> 10 cm dbh) are liana infested
(Alvira et al. 2004) probably contributes to relatively high annual mortality rates (about
3% per year) compared to the 2% annual mortality rates commonly reported for tropical
forests (Phillips et al. 2004a). In addition, this result could be explained by La Chonta's
relatively open structure and the fact that up to 30% of the area is in gap or building
phase (Chapter 1). Although we would expect succession to proceed in these areas over
the long term, it is conceivable that, over the short- or medium-term, the high liana
densities along with dense patches of giant herbs (e.g., Heliconia spp.) and understory
trees (e.g., chocolatillo) that have probably slowed succession to date, would continue to
do so in the absence of any management interventions. One can imagine that as
succession eventually proceeds in existing patches dominated by herbs, lianas, and
treelets to trees capable of reaching the canopy, new open patches form. The observed
basal area increase occurred mostly in the long-lived pioneers, which in the field appear
robust and would probably continue growing over the 60-year period simulated, and
among the shade-tolerant trees that dominate the sub-canopy.