§ uiifkidfl

a! 1.347
c .

‘

 

 

 

 

 

 

 

 

I)
i 3.3.1

zfi. tag
£1.21! .4. a I.
..

.2. .
3...“:

IIHL

1.
2mm”.
. I
gnu.
5 ‘
.th...
..

x‘c
uxllru .4541
.0;
war L...

a
i . , J
v. 11..
.0.- .fl"! it.

. 1-
£6.29 3»
i. I
. .fiuflthu

v3.54:
5 .2 x 9
a. l}. u.

x .3. l5

. .3191...“
.I I 3

‘ 516- 2.4%

 

This is to certify that the
dissertation entitled

THE EFFECTS OF FOREST HARVESTING
ON GIANT PANDA HABITAT USE
IN WOLONG NATURE RESERVE, CHINA

presented by

Scott L. Bearer

has been accepted towards fulfillment
of the requirements for the

PhD. degree in Fisheries and Wildlife

 

 

M

Major Professor’s Signature

/&-/é /fl,f’

Date

MSU is an Affirmative Action/Equal Opportunity Institution

LIBRARY
Michigan State
lJniversity'

 

 

 

 

_.—.-.—o-o-n-.--¢-n-o—.-.-.-.—-—-—--—.-.-n-n-u-.---.-o-o-o---p-.-.—.-.-—o—-----—o--.-.:-

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

 

DATE DUE

DATE DUE

DATE DUE

 

tIWNUIS‘

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/CIRC/Date0ue.indd-p.1

THE EFFECTS OF FOREST HARVESTING
ON GIANT PANDA HABITAT USE
IN WOLONG NATURE RESERVE, CHINA
By

Scott L. Bearer

A DISSERTATION

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

DOCTOR OF PHILOSOPHY
Center for Systems Integration and Sustainability
Department of Fisheries and Wildlife

2005

ABSTRACT

THE EFFECTS OF FOREST HARVESTING
ON GIANT PANDA HABITAT USE
IN WOLONG NATURE RESERVE, CHINA
By

Scott L. Bearer

Forest cover around the world has decreased by over 0.2% annually in the 19905.
Much of this decrease is due to forest harvesting. Decreasing forestland can have
significant consequences for species worldwide because forest ecosystems are a major
reservoir for global biodiversity, supporting 50-70% of the world’s terrestrial species. To
minimize impacts on biodiversity and to increase forest cover, many countries around the
world have planted vast areas of artificially regenerated forest and initiated bans on
logging natural forests. However, many questions remain unanswered concerning the
impacts of forest harvesting and plantation reforestation on wildlife.

To address some of the important issues related to logging and reforestation
impacts on wildlife, I investigated use of habitats by the giant panda in Wolong Nature
Reserve, China. 1 established 913 plots (443 primary and 470 secondary) that measured
30 m2 in forested, non-forested, and reforested areas throughout Wolong and monitored
panda use via feces presence. Within each plot, I recorded forest and geographic
characteristics and also surveyed bamboo stem characteristics in four 1 m2 subplots. I
implemented an adaptive cluster sampling strategy, which sampled I, 5, or 8 plots
depending on whether feces were found. Several statistical and modeling approaches

were used to describe results, including the Horvitz-Thompson estimator along with

autologistic regression to account for the spatial autocorrelation inherent in the adaptive
cluster dataset. I also developed a model that used field data and applied it spatially to
quantify the loss of information when field data were scaled up to the landscape level.

Recent harvests (0-10 years) that were <10 ha had some panda use while larger
recently harvested areas (10-100 ha) had no use. In older harvested forest (30-100 years),
panda use did not differ between small or large. harvests and was similar to that in old
growth forest. My results agreed with previous research that found bamboo and slope to
be primary factors in describing panda habitat, but I found other characteristics were
important in identifying panda habitat as forests mature. Populations of bamboo, the
primary food of the giant panda, have also been affected by harvesting. Reforested areas
that were established in Wolong did not provide suitable habitats for pandas. Reforested
areas did, however, offer the potential to supply an alternative fuelwood source, which
could limit impacts on natural forest.

This dissertation collectively describes how forest harvesting (i.e., timber
harvesting and fuelwood collection) and reforestation activities influence the habitats of
one of the world’s most endangered species, the giant panda. Results are useful to
describe potential impacts of logging activities to other mammalian species worldwide
and may also help project potential consequences of logging bans currently being
implemented in several countries. Additionally, I implement a novel application of a
technique for gathering and analyzing field data of rare species using adaptive cluster
sampling while accounting for spatial autocorrelation. Results from this technique can

provide a useful comparison of this method to simple random sampling.

To my wife
And our families
Especially my parents

For their endless support and encouragement

“I am ashamed ofmy emptiness, " said the Word to
the Work.
“I know how poor I am when I see you. ” said the
Work to the Word.

Rabindranath T agore- Stray Birds

iv

ACKNOWLEDGEMENTS

I would first like to give my sincerest thanks to my advisor Dr. Jack Liu, who
continually encouraged me to achieve something greater. Thanks to the members of my
graduate committee, Dr. David Lusch, Dr. Mike Walters, and Dr. Scott Winterstein for
their support and guidance over the years. I also thank Dr. Connie Page for statistical
support with adaptive cluster sampling and other analyses. J. Egeler, K. Borland, A.
Kortenhoven and R. Howe assisted with support and collection of field data. Thanks to
my colleagues in the Liu Lab, including E. Laurent, C. Lepczyk, A. Morzillo, and N.
Peterson, who were always on hand to provide suitable distractions. I must also give
thanks to those who survived the trips to China with me, including: L. An, M. Linderman,
G. He, X. Chen, V. Hull, W. Liu, 0. Moy, and L. Cheng. And, of course, my forever
patient and helpful wife, Yingyueh Lai. To all of you, may we together eternally dream
of rice, yakjerky and sachimas (maybe with three raisinsl). This research could not have
been accomplished without the brilliant assistance of Z. Ouyang of the Research Center
for Eco-Environmental Sciences at the Chinese Academy of Sciences. Logistical support
was provided by H. Zhang, P. Wang, M. Liu, Y. Tan, and J. Yang. Finally, thanks to my
field crew, especially .1. Huang and S. Zhou, as well as X. Yang and L. Fan, whom I spent
many exhausting days and lots of crazy nights with out in the mountains of Wolong.
Thanks for getting me 10 feet from that panda. For you guys, do ganbei!

This research was made possible by grants from the National Science Foundation,

National Institutes of Health, and the St. Louis Zoo. Thank you.

TABLE OF CONTENTS

 

 

 

 

 

 

 

 

 

 

LIST OF TABLES - viii
LIST OF FIGURES x
CHAPTER 1 1
BACKGROUND AND RESEARCH SUMMARY -- 1
Background ............................................................................................................. 2
Study Area .............................................................................................................. 7
The Forests and Bamboos of Wolong ..................................................................... 9
Giant Panda Habitat Use ....................................................................................... 13
Research Summary ............................................................................................... 13
CHAPTER 2 17
EFFECTS OF TIMBER HARVESTING AND FUELWOOD COLLECTION ON
GIANT PANDA HABITAT USE 17
Introduction ........................................................................................................... l 8
Methods ................................................................................................................. 23
Field Sampling Method: ........................................................................... 23
Statistical Analysis: ................................................................................... 26
Results ................................................................................................................... 28
Discussion ............................................................................................................. 40
CHAPTER 3 47
BAMBOO CHARACTERISTICS AFTER FOREST HARVESTING
IN WOLON G NATURE RESERVE 47
Introduction ........................................................................................................... 48
Methods ................................................................................................................. 51
Statistical Analysis: ................................................................................... 53
Results ................................................................................................................... 54
Discussion ............................................................................................................. 62
CHAPTER 4 - 66
USE OF BAMBOO PATCHES BY THE GIANT PANDA AFTER FOREST
HARVESTING 66
Introduction ........................................................................................................... 67
Methods ................................................................................................................. 69
Field Methods: ............................................. 69
Adaptive Cluster Analysis: ....................................................................... 71
Statistical Analysis: ................................................................................... 73
Autologistic Regression: ........................................................................... 74
Electivity Index: ........................................................................................ 75
Results ................................................................................................................... 77
Discussion ............................................................................................................. 83
CHAPTER 5 _ 90

 

vi

IS REFORESTATION EQUIVALENT TO RESTORATION OF PANDA

 

 

 

 

 

 

HABITATS: A CASE STUDY IN WOLONG NATURE RESERVE 90
Introduction ........................................................................................................... 9 1
Methods ................................................................................................................. 95

Potential of Reforestation Areas to Create Habitat: .................................. 95

Field Survey Methods: .................................................................. 95

Statistical Analysis: ....................................................................... 96

Conservation Potential of Reforestation Areas: ........................................ 97
Results ................................................................................................................. 100
Potential of Reforestation Areas to Create Habitat: ................................ 100
Conservation Potential of Reforestation Areas: ...................................... 104
Conservation Potential of Larix potaninii var. macrocarpa: ................... I 1 1
Discussion ........................................................................................................... 1 14

CHAPTER 6 - 118

SCALING UP FROM SURVEY SAMPLES TO THE LANDSCAPE LEVEL:

GIANT PANDAS ACROSS HETEROGENEOUS LANDSCAPES 118
Introduction ......................................................................................................... l 19
Methods ............................................................................................................... 121

Study Area .............................................................................................. 121
Field Sampling Methods for Data Collection ......................................... 123
Landscape Data ....................................................................................... 126
Modeling Procedure ................................................................................ 126
Statistical Analysis .................................................................................. 130
Results ................................................................................................................. I31
Logistic Regression Models .................................................................... I31
Probability Map of Panda Occurrence .................................................... 136
Discussion ........................................................................................................... 139

CHAPTER 7 144

SUMMARY AND CONCLUSIONS -- - -- 144

APPENDICES 151
Appendix 1. Number of sample plots, proportion with feces, and mean values
(+SE) of forest and geographic characteristics of plots. ..................................... 152

Appendix 2A. Number of sample plots, proportion with feces, and mean values
(+SE) of forest and geographic characteristics for arrow bamboo (Bashaniafabri).
............................................................................................................................. 153
Appendix 28. Number of sample plots, proportion with feces, and mean values
(+SE) of forest and geographic characteristics for umbrella bamboo (F argesia
robusta). .............................................................................................................. 154
Appendix 2C. Number of sample plots, proportion with feces, and mean values
(+SE) of forest and geographic characteristics for yushania bamboo (Yushania

chungii). .............................................................................................................. 155

Appendix 2D. Number of sample plots. proportion with feces, and mean values

(+SE) of forest and geographic characteristics for areas with no bamboo. ........ 156
LITERATURE CITED ...... -- 157

 

Vii

LIST OF TABLES

Table 1.1. Elevational gradient and distribution of forest types, bamboo species. and
panda habitat within Wolong Nature Reserve, Sichuan, China. ............................... I 1

Table 2.1. Univariate ANOVA results for forest and geographic characteristics measured
in primary plots only and neighborhoods (primary + secondary plots).
Neighborhood calculations were based on Horvitz-Thompson means of all plots in
that neighborhood. .................................................................................................... 29

Table 2.2. Significance of forest and geographic characteristics in logistic and
autologistic models developed from primary plots only and Horvitz-Thompson
models developed from all neighborhood plots (primary and secondary plots). ...... 35

Table 3.1. Mean age, patch and harvest size characteristics of three bamboo species and
areas with no bamboo in Wolong Nature Reserve, China. ....................................... 55

Table 3.2. Mean bamboo characteristics (+SE) over time since last timber harvest or
fuelwood collection activity ...................................................................................... 58

Table 3.3. Multivariate loess model results showing most significant forest and
geographic variables that relate bamboo stem characteristics. Variables are listed in
alphabetic order: Age = forest age since last harvest, Basal Area = forest basal
area ( I 0 factor prism), BBC over = bamboo cover (%), Elevation = MSL elevation,
Slope = slope (degrees), Size = forest patch size (ha), T otOverCover = total
overstory cover (%), TotalShrubCover = total shrub cover (%), and T otUnderCover
= total understory cover (%). ................................................................................... 60

Table 4.1. Univariate logistic regression models of forest and geographic characteristics
on primary plots only and Horvitz-Thompson estimators of all neighborhood plots
(primary and secondary). The variable ecological aspect is based on the

recharacterization of aspect into 21 ecologically significant units according to
Parker (1 982). ........................................................................................................... 81

Table 4.2. Multivariate logistic and autologistic models for primary plots only (P) and
Horvitz-Thompson estimators of neighborhoods (P+S) for three bamboo species,

Bashaniafabri, F argesia robusta, and Y ushania chungii. *p<0. l , **p<0.05,
***p<0.01. ................................................................................................................ 82

Table 4.3. Electivity Index values for slope, culm height, basal diameter and culm
density for Bashaniafabri, Fargesia robusta, and Yushania chungii. First column
for each species represents categories and sample size in parentheses. Second
column indicates Electivity Index values. Electivity Index values are scaled

viii

between —I and +1 (<0 = avoidance, 0 = random, >0 = preference)(Lechowicz 1982,
Reid and Hu I991). p<0.l8*, p<0.10**, p<0.01*** ............................................... 84

Table 5.1. Mean (+ SE) values for reforestation and natural forests with mean
comparison tests (Binom. Prop.= Binomial Proportion Test, T-test= 2 sided t-test,

Wilcox= Wilcoxan Rank Sum Test). ...................................................................... 101
Table 5.2. Univariate Logistic Regression P-values. ..................................................... 105
Table 5.3. Final multivariate logistic regression models. .............................................. 105

Table 6.1. A list of environmental, forest, and bamboo characteristics measured in the
field and used in the various models to produce a map of giant panda occurrence.125

Table 6.2. Error rates associated with the various models. Horizontal error rates denote
omission errors while vertical errors denote errors of commission. ....................... 137

ix

LIST OF FIGURES

Figure 1.1. Location of the People’s Republic of China and Wolong Nature Reserve, in
Sichuan Province, southwest China. Shaded relief of Wolong demonstrates the
elevation gradient within the Reserve. ........................................................................ 4

Figure 1.2. Wolong Nature Reserve, Sichuan Province, People’s Republic of China.
Wolong is 2000 km2 in size and represents one of the largest nature reserves
established by the Chinese government for the management and protection of the
giant panda (Ailuropoda melanoleuca). Outlined areas represent distribution of
pandas during 1974 census (adapted from (Schaller et al. 1985)). White represent
areas found to have bamboo in the understory (Linderman et al. 2004). Regions of
Zhuanjinggou and Yingxionggou are areas of previous widespread forest harvesting.
..................................................................................................................................... 8

Figure 1.3. Conceptual framework of the relationship between humans, forests, and
wildlife habitats (adapted from Liu et al. (1999b)). Dashed box represents the area
of focus for this dissertation. ..................................................................................... 15

Figure 2.1. Mean values (+SE) of topographic and forest characteristics for 30 x 30
meter sample area in categorized age of forest and size of harvest categories (size of
patch for unharvested) ............................................................................................... 30

Figure 2.2. Proportion of primary plot sites with giant panda feces in areas with varying
forest ages and sizes. Italicized numbers above bars represent plot sample size.
Symbols (*, **, +, ++) above bars represent significance groupings in the proportion
of survey plots with panda feces by age class for each size using binomial proportion
test. The same number (and color) of symbols above bars indicates no significant
difference. ................................................................................................................. 33

Figure 2.3. Regression trees of A) all forest and geographic characteristics, B) age and
size of harvested and unharvested patches, C) geographic characteristics, slope.
ecological aspect, and basal area (mZ/ha), and D) forest characteristics. Trees show
classification based on the presence of giant panda feces. Residual mean deviance =
A) 0.175, B) 0.183, C) 0.186, and D) 0.159. Values at tree leaves represent
proportion of plots with feces and sample size (within parentheses). ...................... 37

Figure 3.1. Mean characteristics (+ SE) of three bamboo species in Wolong Nature
Reserve, China. ......................................................................................................... 56

Figure 4.1. Percentage of survey plots in areas with three species of bamboo and areas
with no bamboo that have giant panda feces present in various aged forests ........... 78

Figure 5.1. Mean (+SE) percent coverages for total overstory, midstory, shrub and
bamboo strata in artificially regenerated forest and surrounding natural forest for
areas with and without giant panda feces ................................................................ 103

Figure 5.2A. The area of natural forest impacted by various levels of forest harvest (1.33-
10.6 cubic meters/year) in different aged natural forests. Values represent mean
ages from 5 age classes (0-5, 6-20, 21-40, 41-80 years, and unharvested forest) and
forest volume from 342 field plots which established volume per acre. ................ 106

Figure 5.28. The area of natural forest impacted by various levels of forest harvest (1.33-
10.6 cubic meters/year) in natural forests with different volume stands. This direct
relationship describes harvesting levels and forest densities found in Wolong Nature
Reserve, 2001—2003. ............................................................................................... 108

Figure 5.3. A model of the relationship between tree woody volume (m3) and potential of
finding feces using data collected in the field. ........................................................ 109

Figure 5.4. Habitat degradation determined by the change in panda feces proportion
found in 342 field plots (high decrease = high degradation) from various levels of
forest harvest in natural forests that range from having low to high initial forest
volume ..................................................................................................................... 1 10

Figure 5.5. Pole volume from plantations of Larix potaninii var. macrocarpa and the
number of households that can harvest 100% of their fuelwood needs (10.6 cubic
meters/year) from the total volume in these plantations. The Simulation was
conducted between 10-90 years because these were the years available for the
species growth equation used in the model ............................................................. 1 12

Figure 5.6. Proportional increase in the number of natural forest sites used by giant
pandas when plantations are harvested instead of natural forests. This relationship
was established based on the woody volume of plantation forests over time (see
Figure 5.5) and the proportion of sites with feces in natural forests of varying woody
volumes (see Figure 5.3). The simulation was conducted between 10-90 years
because these were the years available for the species growth equation used in the
model ....................................................................................................................... 1 13

Figure 6.1. The location and elevation of Wolong Nature Reserve, Sichuan, China. 122

Figure 6.2. A characterization of the scaling-up process. Seven variables are sampled in
the field (A) and used to generate a Full Model (B). Numbers within each sampling
plot represent probability of species occurrence values for a theoretical scaling-up
modeling exercise. A Reduced model (C) includes only five of the original
variables because not all variables are mappable. Only three of the variables are
presently available to be used as input into a landscape model (D), thereby creating a
simplified model that may have higher errors than the Full and Reduced models. 128

xi

Figure 6.3. Calculated probabilities of giant panda occurrence grouped and averaged by
survey plots with/without panda feces. ................................................................... 132

Figure 6.4. The proportion of survey plots that did/did not have giant panda feces
classified by probability levels of finding feces. Gray bars represent plots without
feces. Black bars represent plots with feces ........................................................... 134

Figure 6.5. The probability map of panda occurrence showing the distribution of field
plots (both primary and secondary plots) used to create logistic regression models.
Field plots are focused within the central region of Wolong known to have
concentrated panda activity ..................................................................................... 138

Figure 6.6. Calculated probabilities of giant panda occurrence grouped by age of forest
for the Full and Reduced models (Age Class l= 0-5 years, Age Class 2 = 6-20 years,
Age Class 3 = 21-40 years, Age Class 4 = 41-80 years, and Age Class 5 =
unharvested forest). ................................................................................................. I42

xii

CHAPTER 1

BACKGROUND AND RESEARCH SUMMARY

CHAPTER 1

BACKGROUND AND RESEARCH SUMNIARY

Background

Forests are perhaps the most significant renewable resource in the world.
Covering nearly one-third of the world’s land area (approximately 3800 million ha),
forests provide many ecological services (Costanza et al. 1997) in addition to supplying
much of the world with many goods and raw materials including timber and fuelwood
(Convention on Biological Diversity 2001). As levels of atmospheric carbon dioxide rise
and issues of climate change become a part of modern political debate (Stier and Siebert
2002, NCAA 2004), forests are also valuable for the carbon sequestration (forest
vegetation and soils currently contain nearly 40% of all carbon stored in terrestrial
ecosystems (Matthews et al. 2000)).

Despite all the benefits that forests offer, they are frequently poorly managed
and extensively logged, or worse, converted to other land uses altogether. Timber
harvesting for wood products is responsible for approximately one-third of global
deforestation (Pottinger 2003), and roughly half of this harvesting is done illegally.
Illegal harvesting alone may produce nearly ”10 of the total global timber worth US$150
billion annually (Pottinger 2003). Many countries have adopted policies to improve
management, increase conservation or limit illegal harvesting of forests (F A0 2005).
However, deforestation and its effects may continue to remain a concern for decades, if
not centuries, to come.

Ecological impacts of forest harvesting are significant because forests are thought

to be the largest biodiversity banks in the world, housing 60-90% of all terrestrial species

(IUCN: The World Conservation Union 2005). However. many of these species’ habitats
are not protected (Rodrigues et a1. 2004), and even areas that are protected may not
necessarily benefit the resident species (Liu et al. 2001).

Logging impacts on biodiversity are particularly significant in Asia. Asia not
only has one of the lowest forest to area ratios in the world (432 million ha forest to 3.09
billion ha land (FAO 2000), but also one of the highest deforestation rates (2.5 million
ha/yr (FAO 2000)).

Within Asia, the People‘s Republic of China (Figure 1.1) is by far the largest
country (9.6 million kmz) with a wide range of ecosystems (599 terrestrial ecosystems
and the second highest diversity of species in the world (6347 vertebrate species, 14% the
world total (FAO 2000, Li and Wilcove 2005)). China is also one of the areas in the
world at most risk to the environmental problems. including pressures on biodiversity (Li
2004. Liu and Diamond 2005).

China, which is the world’s third largest country in size, has one of the highest
(approximately 22% of the world’s population) and densest (roughly 120 people/km2)
human populations (as opposed to 20 people/km2 in the United States or 60 people/km2 in
Europe) (Peng and Guo 2000). With a high degree of human pressure, China has one of
the largest ecological footprints, second only to the United States (Liu and Diamond
2005).

Because of the pressures on its environment, China has used its forest resources
extensively (Li 2004). Deforestation has taken place over thousands of years and has
increased to high rates during many periods of Chinese history (Kristof 1993. Liu and

Diamond 2005).

     

.25 . ..
'-

‘ 3"“ " % Elevation (MSL)

, ‘ I: 1000 - 2000
‘ [:1 2000- 3000
3’“ 3000 - 4000
- 4000 - 5000

 

5000 - 6250
[I No Data
10 0 10 20 30 Kilometers

 

 

 

Figure 1.1. Location of the People’s Republic of China and Wolong Nature Reserve, in
Sichuan Province, southwest China. Shaded relief of Wolong demonstrates the elevation

gradient within the Reserve.

Currently, China’s rapid economic growth is significantly increasing its demand
for forest products. China is now one of the largest consumers of various types of wood,
including industrial roundwood (2nd), sawnwood (5th), and pulpwood (2nd), in addition to
being the world’s largest producer of plywood (Sun et al. 2004, Sun et al. 2005).
Logging has been found to be the primary cause of habitat destruction in China, affecting
33% of vertebrate species, including 49% of mammal species (Li and Wilcove 2005).
This has resulted in approximately 15% of China’s vertebrate species currently being
considered threatened with extinction (Li and Wilcove 2005).

Harvesting pressures are particularly high in southwest China, which has the
largest amount of forestland (71 million ha) and the highest volume of standing timber (5
billion m3) in China (Shi and Xu 2000). Harvesting pressures help to explain why the
southwestern region lost 39% of its mature forest stock and 25% of its total timber forest
stock between 1989-1993 (Shi and Xu 2000). In southwest China, timber harvesting has
been one of the leading driving forces in biodiversity loss (Xu and Wilkes 2004).

Located in southwestern China, Sichuan Province has the largest tract of
forestland (11.5 million ha) and timberland (10.3 million ha) in the region and the third
largest in the nation (Shi and Xu 2000). In the early 19505, Sichuan had approximately
1.6 billion m3 of forest growing stock (Wang and Chen 1992, State Environmental
Protection Administration of China 1998). However, Wang and Chen (1992) estimated
that by 1980 this forest cover had decreased by 550 million m3, primarily through illegal
logging and conversion of forestland to agriculture. Other recent estimates place

Sichuan’s current timber stock at 1.3 million m3 (Shi and Xu 2000).

To reduce extensive pressures on its forest resources, China has been
implementing and enforcing new forest management regulations. In response to severe
flooding in 1998 and in order to avoid additional disasters due to unchecked forest
harvesting, the central government established two policies, the Natural Forest
Conservation Program and the Sloping Lands Conservation Program (or Grain to Green),
which attempt to limit timber harvesting and fuelwood collection and reforest areas on
steep slopes in return for grain subsidies, respectively (Xu et al. 2004, Liu and Diamond
2005). These two programs may have long-tenn effects on China’s biodiversity and
wildlife habitats, as well as having consequences for other countries in Asia and the
world. Due to the recent regulations on harvesting in China, foreign forests now must
supply much of China’s consumption. Between 1997 and 2003, Chinese forest product
imports rose 75% to US$13 billion (Sun et al. 2004). China is currently importing nearly
95 million m3 of wood, equal to 150% of Indonesia’s total annual timber harvest (Sun et
al. 2004). Therefore, the threats to biodiversity and wildlife habitats from harvesting due
to China’s forest product demands, as well as China’s new forest regulations, are not only
felt in-country, but may impact the entire Asian region.

One of the most charismatic species influenced by timber harvesting that has
concerned the Chinese, as well as others throughout the world, is the endangered giant
panda (Ailuropoda melanoleuca) (Schaller et al. 1985). The influence of timber
harvesting and fuelwood collection on the giant panda is of particular concern to
researchers and scientists, not only because of its endangered status, with total population
numbers estimated around 1750 individuals (Xinhua 2004), but also because the panda is

now limited to only a fraction of its original habitat. To protect biodiversity, the Chinese

government has established approximately 2,000 nature reserves by the end of 2003 (Liu
and Diamond 2005). F orty-five of these reserves were established specifically for the
conservation of giant panda habitats. One of the largest of these is Wolong Nature

Reserve.

Study Area

The Wolong Nature Reserve (102°52’-103°24‘E. 30°45‘-3I°25’N) is
approximately 200,000 ha in size, and is located in Wenchuan County, Sichuan Province.
southwestern China (Figures 1.1 and 1.2). Although Wolong is a nature reserve
established to conserve panda habitat, majority Han, and minority Tibetan, Chang, and
Hui Chinese are still permitted to live within its borders. In 1998, a total of 4320 people
were counted living in 942 households (An et al. 2001). This number is a dramatic
increase from earlier population estimates (70% increase from 1975 to 1998).
Concurrently with this human population increase was a rapid decline in the Wolong
panda population. In 1974, 145 animals were estimated to live in Wolong (Schaller et al.
1985) and these numbers fell to 72 animals in 1986 (China's Ministry of Forestry and
World Wildlife Fund 1989). The most recent panda census used new techniques and
found a larger population (Xinhua 2004), though researchers agree this does not
necessarily indicate an increasing population trend (World Wildlife Fund 2004). High
quality panda habitat was also found to be declining even afier the area was established
as a nature reserve (Liu et al. 2001).

In an attempt to understand the relationship between humans and panda habitat,

Liu et al. (1999b) found that there are many specific human activities causing a rapid

 

Figure 1.2. Wolong Nature Reserve, Sichuan Province, People’s Republic of China.
Wolong is 2000 km2 in size and represents one of the largest nature reserves established
by the Chinese government for the management and protection of the giant panda
(Ailuropoda melanoleuca). Outlined areas represent distribution of pandas during 1974
census (adapted from (Schaller et al. 1985)). White represent areas found to have
bamboo in the understory (Linderman et al. 2004). Regions of Zhuanjinggou and
Yingxionggou are areas of previous widespread forest harvesting.

8

decline in panda habitat. Liu et al. (1999b) note that approximately 41% of the reserve
would be suitable habitat for the giant panda on biological factors alone. However, when
human influences are considered, approximately 17,000 ha (21%) of panda habitat was
lost. leaving a much less suitable, highly fragmented reserve. The majority of these
changes took place along the main highway that bisects the reserve and is due mainly to
the prevalence of historical timber harvesting and present day fiIelwood collection, as

well as agricultural practices and livestock grazing.

The Forests and Bamboos of Wolong

Timber harvesting and fuelwood collection have caused a dramatic decrease in
suitable panda habitat. In Wolong, the collection of wood for fuel has doubled over the
past two decades to more than 9,000 m3/year (Liu et al. 1999b). Similar to farming and
agricultural practices, this activity removes much of the vegetation in the impacted area.
C learings vary in size. but are typically 2 to 6 hectares (Bearer, unpublished data).
Location of clearings also varies. but areas harvested for fuelwood collection are
typically within 2 to 4 km from the nearest road.

Extensive timber harvesting had been performed throughout Yingxionggou and
Zhuanjinggou areas of Wolong (Figure 1.2) throughout much of the 19605 to the mid-
19705 by one primary timber extraction company. Personal observation indicated typical
forestry practices included removal of >50 % of the overstory canopy cover in areas
usually between 0.1 km2 and l kmz. Timber was used in construction and furniture

making, as well as for pulpwood. In 1974-75, the primary timber extraction company

\

was relocated to northern Sichuan Province after Wolong was established as a nature
reserve. The Sichuan Ministry of Forestry officially banned timber harvesting within
Wolong in 1985, however, there is still a high amount of timber harvesting (fluctuating
between l,000-1,400 m3/year) occurring in the Reserve (Liu et al. 1999b).

The forests in Wolong can generally be divided into different categories along an
elevational gradient (Schaller et al. 1985). An evergreen and deciduous broadleaf forest
grows between an elevation of 1600 m and 2000 m, covering approximately 13% of the
Reserve (Table 1.1). Between 2000 and 2600 m, the forest is composed of mixed conifer
and deciduous broadleaf species. Hemlock (T suga spp.), spruce (Picea spp.), and larch
(Larix spp.) are the main conifers, while birch, maple, basswood, and cherry are the most
common deciduous trees. F argesia robusta is the principal bamboo Species with small
populations of Bashania fangiana also being found in this range (Table 1.1). Extending
from 2600 to 3600 m is the subalpine coniferous forest, which accounts for the most
extensive forest type, covering 317 km2 (16%) of the total area of the Reserve (Table
1.1). Fir species are the most common in this forest. along with birch and rhododendron.
In this elevational zone, B. fangiana is the most widespread bamboo (Schaller et al.
1985). Because the giant panda’s main range occurs within the latter two forest types
(between 2000-2600 m and 2600-3600 m), the majority of the investigations of forest
composition related to giant panda habitat have focused on these areas. Schaller et al.
(1985) along with Taylor and Qin (l988b, 1988c) provided information on the structure.
composition, and regeneration patterns of some of these forests. More importantly, they
offered some insight into how the forest may be influenced after harvesting activities

such as fuelwood collection (Taylor and Qin 1988a). These studies presented a good

10

Table 1.1. Elevational gradient and distribution of forest types, bamboo species, and

panda habitat within Wolong Nature Reserve, Sichuan, China.

 

 

 

 

General

Elevation (MSL) Forest Type Bamboo Species Panda Habitat

Categories

5000 fiock 8 Snow No bamboo

4600 Alpine Coniferous No bamboo

4000 Alpine Coniferous No bamboo

3600 Alpine Coniferous Bashania fabri Occassional

Subalpine Coniferous

3000 Subalpine Coniferous Bashania fabri Main Range
Yushania Chungii

2600 Subalpine Coniferous Bashania fabri Seasonal
Yushania chungii & Fargesia robusta

2000 Coniferous and Deciduous Broadleaf Fargesia robusta Seasonal

1600 Evergreen and Deciduous Broadleaf Fargesia robusta Rare

1000 Evergreen Broadleaf Other species

 

 

(from Schaller et al. (1985))

understanding of dynamics of the Wolong forests. However, this is incomplete unless
forest disturbance—bamboo dynamics are described.

The significance of understanding bamboo characteristics and elevational
placement lies in the fact that it is the main food of the giant panda (Schaller et al. 1985).
Because this plant is directly related to the panda habitat, the ecology of the bamboo
within Wolong has been investigated on several occasions. Characteristics such as culm
dynamics (Taylor and Qin 1987), regeneration patterns (Taylor and Qin 1988b),
environmental influences affecting its growth (Reid et al. 1991c), and how it responds to
flowering (Taylor and Qin 1993b) have been described. Although the biology of the

plant has become well understood through these and other studies, information about

11

bamboo distributions within Wolong has, until recently, been general. DeWulfet al.
(1988) generated a general bamboo distribution map using Landsat MSS imagery while
Linderman et al. (2004) used artificial neural networks to generate a more accurate
bamboo map with Landsat TM data.

As humans continue to impact the local forests, and as the bamboo itself
continues to mass-flower and regenerate, bamboo distributions within Wolong Nature
Reserve fluctuate over time. This investigation occurred at a time when bamboo
densities within the reserve were high. so the locations of panda activity noted in this
study may be different than activity during times of low bamboo density. Bashania
fangiana, which dominates the understory of most subalpine forest stands, mass-flowers
every 45-60 years. After these flowering events, the bamboo dies off so that vast
expanses of bamboo throughout the reserve are no longer suitable panda forage. The last
flowering episode of Bashania in Wolong was in 1983. Many studies were done on
bamboo dynamics and panda use during this time (Taylor and Qin 1987, Taylor and Qin
1988a, Taylor and Qin 1988b, Taylor and Qin 1988c, Taylor and Qin 1989, Reid et al.
1991c, Taylor et al. 1991, Taylor and Qin 1993b, 1997). Taylor and Qin (1988a)
estimate that it takes 15 years for bamboo populations to recover to fully-stocked
densities after a flowering episode. By calculating Electivity Indices and comparing
these to previous use, this study provides information concerning panda habitat use
during these fully-stocked periods, and will allow for comparisons about how pandas

may shift habitat use during times of high and low food availability.

12

Giant Panda Habitat Use

Although pandas may eat a variety of plants, and occasionally meat, more than
99% of their diet consists of bamboo, including stems, branches, and leaves (Schaller et
al. 1985). However, panda digestion of bamboo is inefficient (17% dry matter
digestibility) so the animal must invest more than 60% of the day on foraging alone
(Schaller et al. 1985). For this reason, pandas are limited regionally and locally by the
presence of bamboo. In Wolong, the main panda range is within the subalpine coniferous
forest areas where bamboo can occur. Two major species of bamboo are extant within
this region. Umbrella bamboo (Fargesia robusta) is typically found in the lower
elevations, while arrow bamboo (Bashaniafabri) occurs in the upper elevations (Table
1.1) (Schaller et al. 1985). Schaller et al. (1985) found that pandas spent the majority of
the year (>85%) above 2600 m in B. fabri bamboo patches. Use below these elevations
was limited due to habitat disturbance or steepness of terrain, however pandas did use
these areas when new shoots of F. robusta were developing in May and June (Schaller et
al. 1985). A third, less prevalent bamboo species, Yushania chungii, is scattered
throughout the higher elevations along with B. fabri, but panda use of this species has

limited documentation.

Research Summary
This dissertation arises from my interest in determining the role of forest
management and timber harvesting in affecting and conserving wildlife habitats. More
specifically, as fuelwood continues to be essential for the survival of billions of people

worldwide (Gardner-Outlaw and Engelman 1999. Arnold et al. 2003), 1 investigated both

13

larger area harvesting along with fuelwood collection activities and their effect on panda
habitats. It is my hope that the research described in this dissertation can provide another
piece to the large puzzle of enhancing human welfare and alleviating poverty, while
committing to sustainable forest use and conservation of wildlife habitats. By analyzing
field data collected in Wolong Nature Reserve, 1 can demonstrate how harvesting for
timber and fuelwood influences one of the most rare, most peculiar of mammalian
species, the giant panda.

The research described in this dissertation is part ofa larger socioeconomic
ecological research program led by .1 ianguo Liu at the Center for Systems Integration and
Sustainability in the Department of Fisheries and Wildlife at Michigan State University.
The purpose of this program is to understand the relationships between human
populations and government policies, forests, and giant panda habitat (Figure 1.3) (Liu et
al. 1999b). From this framework, I focused on the interaction between the forests of
Wolong and panda habitat.

I examined habitat use patterns (determined by feces presence) in areas associated
with logging and reforestation activities within the forested and non-forested regions of
Wolong. The next five chapters of this dissertation were written as individual
manuscripts that will be submitted for peer-review in appropriate journals. Chapters 2
through 5 concentrate on how forest harvesting influence various aspects of panda
habitat. Chapter 2 is an investigation into the effects of harvesting on the forest
components of giant panda habitat and how this affects panda use. After exploring the

forest community, I focus my attention in the next two chapters (Chapters 3 and 4) on the

 

 

 

  

 

 

 

 

 

 

 

 

 

 

 

 

Attitudes/ I '
Perceptions : Dynamics
I L

Population ctivities i Structur Integrity
Household A L

NeedsNVants Function

i
Policies Quantity

 

 

 

A

Timing

 

 

 

 

—_——-—-—I-—_——-——_——_—-

Figure 1.3. Conceptual framework of the relationship between humans, forests, and
wildlife habitats (adapted from Liu et al. (1999b)). Dashed box represents the area of

focus for this dissertation.

bamboo areas of Wolong. Chapter 3 examines the effects of harvesting on bamboo
populations and bamboo stem characteristics. In Chapter 4, I explore how giant pandas
are using the bamboo populations throughout Wolong, and how harvesting may be
affecting panda use of these bamboo stands. After describing how forest harvesting
activities, that is, timber harvesting and fuelwood collection, have affected giant pandas
and their habitat, I then illustrate the efforts that are being made by the Wolong
Administration and the Chinese central government to protect and regenerate panda
habitats. To do this, I research and model the potential benefits of reforestation projects
in Wolong in Chapter 5. This chapter uses data collected in reforested areas and

surrounding natural forest in Wolong to determine whether reforestation can create or

15

mitigate giant panda habitat. Finally, in Chapter 6, I use the data that I’ve collected in the
field to quantify information loss in landscape models as variables are withdrawn from
the dataset to be applied spatially. Conclusions from all chapters and a summary are

presented in Chapter 7.

I6

CHAPTER 2

EFFECTS OF TIIVIBER HARVESTING AND FUELWOOD COLLECTION ON

GIANT PANDA HABITAT USE

l7

CHAPTER 2
EFFECTS OF TIMBER HARVESTING

AND FUELWOOD COLLECTION
ON GIANT PANDA HABITAT USE

Introduction

Global forestland declined by 0.22 percent each year during the 1990’s (F A0
2003). Loss of forest has been particularly significant in developing countries, which
experienced a total loss of forest area in 1990—1995 equal to 13.7 million ha per year
(F A0 2001 ). In some developing countries that have increasing amounts of forestland,
such as China, which increased the size of its forest between 1980 and 1995, the per
capita amount of forestland continues to decline due to rapid population growth (PAI
1999). Decreases in forest area can have a significant effect on biodiversity (Caldecott et
al. 1996, Jenkins 2003). Wildlife are particularly sensitive to the changes in the
ecosystem and increase in habitat heterogeneity that results from logging.

Recent investigations are beginning to uncover the processes involved in how
wildlife respond to harvesting, thereby allowing better management policies to be
developed and implemented. More specifically, timber harvesting activities have been
found to affect various types of wildlife, including invertebrates (Hill 1999, Ghazoul
2002, Hamer et al. 2003, Hill et al. 2003), reptiles (Prior and Weatherhead 1994),
amphibians (Petranka 1994, Harper and Guynn 1999, Harpole and Haas 1999, Duguay et
al. 2001, Duguay and Wood 2002, Knapp et al. 2003), birds (Ward et al. 1994,
Sallabanks et al. 2000, Duguay et al. 2001. King and DeGraaf 2004), small mammals

(Cole et al. 1998, Malcolm and Ray 2000, Sullivan et al. 2000. Simard and Fryxell 2003)

18

and large mammals such as bear (McLeIlan and Shackleton 1989, Wielgus and Vernier
2003).

Effects of harvesting vary between species and even related species within the
same guild can have different responses (Lindenmayer and Lacy 1995). For example,
mammalian specialists that rely on early successional habitats have traditionally benefited
from widespread logging (Heydon and Bulloh 1996, 1997), but are now under pressure as
habitat matures (Litvaitis and Villafuerte 1996b, Smith and Litvaitis 2000, Thompson and
DeGraaf 2001). On the other hand, mammal species that prefer interior habitats have
been negatively impacted from loss or fragmentation of habitat associated with
harvesting (Zheng et al. 1995, Green and Catterall 1998, Potvin et al. 1999, Wormington
et al. 2002) and from increased exposure to humans after logging roads become
developed in core areas (Robinson et al. 1999, Litvaitis 2001).

Harvesting effects on less specialized mammal species can be difficult to
ascertain because they demonstrate little response to harvested areas (White 1992,
Litvaitis 2001), continue to use remnant patches (Potvin et al. 1999, deBellefeuiIle et al.
2001 ), or show a preference for certain stages of regeneration (Smith and Lindenmayer
1992, Johns et al. 1997, Chamberlain et al. 2003). Effects on large mammals, especially
bear species, are particularly difficult to quantify because of their rareness, large home
range size, and problems in direct sampling (Johns et al. 1997). Recent studies have
suggested grizzly bears (Ursus arctos Ord) are not deterred by harvesting activities
(McLeIlan and Shackleton 1989) and may actually prefer early succession forests

(Mattson 1997, Waller and Mace 1997). The giant panda (Ailuropoda melanoleuca

I9

David), on the other hand, appears to Spend little time in heavily logged areas (Schaller et
al. 1985) and prefers interior forest.

Adding to the complexity is the fact that timber harvests occur in different forms
along a disturbance gradient where various levels of overstory, midstory and understory
cover are removed. The various types of harvests, such as clearcutting (Kirkland 1990.
Sullivan et al. 1999), shelterwood cutting (Anderson and Crompton 2002) and variable
retention harvests (Sullivan et al. 2000. Sullivan and Sullivan 2001) have diverse effects
on wildlife.

One of the most important forms of harvest in the world today is fuelwood
collection. Fuelwood acts as the primary energy source for approximately 3 billion
people worldwide (PAI 1999). Over half of the wood harvested in one year is used for
fuel (PAI 1999), accounting for over 7 million ha of forest area harvested annually.
Developing countries are the primary users of fuelwood, and 75% of forest harvesting in
these countries is for fuelwood use (FAO 2000). Yet, despite the widespread use and the
potential impacts harvesting can have on forested landscape and wildlife habitats, we are
unaware of any study that investigates the effects of fuelwood collection on wildlife

habitat use.

We therefore investigate the effects of forest harvesting (i.e., timber harvesting
and fuelwood collection) on giant panda habitat use in Wolong Nature Reserve, Sichuan,
China. The location was ideal because Asia is one of the largest users of fuelwood,
accounting for 46% of global fuelwood production (WEC 2004). Within Asia, the
People‘s Republic of China is the largest country (9.6 million kmz) with the highest

population (1.3 billion) but has one of the lowest percentages of forest area (18%, (Liu

20

and Diamond 2005)). In addition, China has one of the largest ecological footprints,
second only to the United States (Liu and Diamond 2005) and consumes approximately
440 million tons of fuelwood per year, which accounts for 34% of the total annual
consumption of energy in rural China (Kunshan et al. 1997).

The Wolong Nature Reserve (102°52’-103°24’E, 30°45’-31°25’N) is a nature
reserve specifically established for the conservation of panda habitat. However,
extensive timber harvesting was performed in Wolong throughout much of the 1960’s to
the mid 1970’s. Typical practices included removal of >50% of the overstory canopy
cover in areas usually approximating 10 ha but occasionally exceeding 100 ha. The
Sichuan Ministry of Forestry officially banned timber harvesting within Wolong in 1985,
however, some amount of timber harvesting and fiielwood collection (fluctuating around
7,000-9,400 m3/year) has continued to occur in the Reserve until recently (Liu et al.

199%, Liu et al. 1999b).

Several investigations have given some insight into the way native vegetation has
been influenced by historical timber harvesting and fuelwood collection and how the
habitat matrix of the giant panda would be affected by later forest development (Schaller
et al. 1985, Taylor and Qin 1988b, Taylor and Qin 19880, Reid et al. 1989, Taylor and
Qin 1989, Reid et al. 1991c, Taylor and Qin 1993a, 1997). When large openings in the
canopy are formed. such as those created by clearcut harvesting operations, the density of
arrow bamboo increases to such a degree that tree establishment is severely hindered
(Taylor and Qin 1987, Taylor and Qin 1988a, Reid et al. 19910, Taylor and Qin I993a,
b). Because of this encumbrance, regenerating forest is slow in areas with this type of

harvest regime. Harvesting methods of this kind occurred in the Fanzipeng region during

21

the 19305 (Figure 1.2). This area is now primarily a hardwood stand composed mainly
of Betula spp. with few conifers remaining (Taylor and Qin 1988b, Taylor and Qin
1988c, Taylor and Qin 1989). Large clearcuts also occurred in the Yingxionggou and
Zhuanjinggou regions during the 1970’s (Figure 1.2). These activities have left many of
those areas in a second growth shrub layer consisting mainly of Rubus spp., Sorbus spp..
and Rosa spp. (Reid et al. 19910).

Selective timber logging and fuelwood collection impact the forest differently
than all-tree removal methods. Typically, fuelwood collection activities occur over a
much smaller area, as usually only several trees are removed at one time. Yet, the
cumulative effect of continual fuelwood collection over several years does increase the
size of the area impacted. Because of this collective effect, harvest areas vary in size
from several square meters (single tree removal) to several hectares or larger, depending

on the temporal extent and the number of local people who collected fuelwood.

Despite these insights into how forests regenerate and interact with bamboo, there
has been no investigation to date that monitors the age or size of harvest area and how
associated forest characteristics directly relate to giant panda use. In addition, the timing
of this study is crucial because no studies of panda habitat have been conducted after the
bamboo recovery of a mass-flowering episode (Taylor and Qin 1993b). This
investigation reports on the effect of timber harvesting and fuelwood collecting on the
use of habitats by the giant panda approximately seventeen years after a bamboo mass-
flowering event. Our four primary objectives were to: 1) evaluate forest structure
composition changes in various sizes and ages after harvesting, 2) assess if a size of

harvest effect on panda use was present (i.e., does use in fuelwood collection sites (< 10

22

ha) differ from use in timber harvested areas 10-100 ha), 3) determine if time since
harvest affected panda use, and 4) quantify the environmental factors important to giant

panda use over periods of time after harvesting.

It is our hope that by providing a better understanding of how giant pandas use
previously harvested areas, we can not only ensure the long-term sustainability of panda
habitats. but also provide a baseline for understanding how fuelwood harvest and forest

resource utilization may coexist with wildlife conservation.

Methods
Field Sampling Method:

To determine whether harvesting a forest for timber or fuelwood affects the use of
habitats by the giant panda, we examined use patterns in previously harvested forests and
compared use to uncut forests. Complex site histories made it difficult to characterize
any harvesting as a clear-cut or selectively cut operation, so we used size of harvest to
quantify type of cut, where smaller sized harvests are typically associated with fuelwood
collection. or selective cutting, and clear-cuts were normally larger in size. Data
collection took place during May-August 2001, May-November 2002, and June-August
2003.

Use was determined by the presence of fecal droppings, which was considered a
useful indicator of panda use because although actual Sightings of animals are extremely
rare, feces are deposited frequently (97 droppings/day) (Schaller et al. 1985) and remain
for several months (personal observation). In an earlier study of panda habitat use, Reid

and Hu (1991) used the sample unit of feces groups to determine use mainly associated

23

with feeding behaviors. Our study used general presence/absence of feces to indicate use
because it also incorporated habitat use not associated with feeding, such as traveling
routes. This was significant because panda use in harvested areas may be more closely
associated with daily or seasonal movement, escape or thermal cover, or mate location
and may not be as important as foraging habitat (Schaller et al. 1985). Therefore, until a
better survey technique is developed and the giant panda radio-collaring ban lifted, feces
are the most efficient and effective way to determine use of habitats. Although frequency
of defecation is high, the small number of panda individuals, extent of the forest and
ability of the animal to travel large distances make detection of feces infrequent.

Due to the infrequency of observing giant panda feces in bamboo forests, we used
adaptive cluster sampling to concentrate our sampling effort on areas where feces was
more likely to occur. This allowed us to avoid a situation where most sample plots had
no feces present. Adaptive cluster sampling is valuable because it is designed to sample
rare events (i.e., feces) and allows for focused research in areas where the rare events
occur while maintaining proper statistical hypothesis testing protocol (Thompson 1991b,
Thompson et al. 1992). Whenever feces were found (i.e., presence=l) in a 30 x 30 m
sample plot, all units in its neighborhood (those 30 x 30m plots to the forward, rear, left,
and right) were added to the sample. If in turn any of these subsequently added plots had
feces that were detected by the observer, the plots of its neighborhood were also added to
the sample, so that finally the sampling neighborhood contained 1, 5 (1 primary, 4
secondary), or 8 (1 primary, 7 secondary) sample plots. Adaptive cluster sampling
allowed us to increase our sampling efforts around areas that had a higher potential of

giant panda activity.

24

Thirty x thirty-meter sampling plots were located while the team was in the field
and were identified by local residents knowledgeable of historical, recent past and present
land-use in the area. Age of forest was defined by the last time the forest within the
sample plot had been harvested, which was known by the locals. Size of harvest was
defined as a relatively homogenous area that differed from its surroundings (Forman
1995). The center position of each 30 x 30 m plot was geo-referenced with a Pathfinder®
Pro XRS GPS unit (Trimble Navigation Limited) for subsequent analysis and modeling
of the data. Due to logistical and time constraints which prohibit revisitation and
reexamination of sample plots for passed droppings, equal probability of feces detection
in all habitats was assumed. This assumption was supported by Reid and Hu ( 1991), who
found a low ratio of missed droppings (1.3%) throughout their study sites in Wolong.

The. number of sample plots that were surveyed depended upon the size of the
harvested area that was sampled, so that approximately 5% of the surface area within the
harvest site would be sampled. For example, small harvest areas (5 2 ha) had one
primary 30 x 30 m plot within the harvested area and one plot in the surrounding uncut
forest while large harvested areas (3 9 ha) had four primary plots inside and four in the
surrounding forest. Using this method, we attempted to sample all age classes of forest in
approximate proportion to their extant within the study area. However, several areas.
especially those near households, had a complex site history with large, old harvests
having smaller, newer harvests within, and were therefore not sampled by this method.

Forest characteristics measured within each plot include: basal area (mz/hectare,
IO-factor prism), total overstory canopy cover (all canopy covers measured with spherical

densiometer). conifer overstory cover, deciduous overstory cover, average overstory

25

height, total midstory canopy cover, average midstory height, total shrub cover, average
shrub height, total bamboo cover, total understory cover, and average understory height.
Topographic factors include: ecological aspect (aspect converted into ecologically

significant units (Parker 1982)), elevation (MSL), and slope (degrees).

Statistical A nalysis:

Sites were grouped into four harvest age categories to reflect stages of forest
succession and were based on our ability to identify forest ages in the field as well as
sample size issues: 0-10 years, 1 1-30 years, 31-100 years and old-growth forest. Sites
were also grouped into two patch/harvest size categories: small 510.0 ha and large
harvest 10-100 ha. Sample sizes varied because only sites that could definitively be
categorized were included in this analysis.

We used analysis of variance (ANOVA) along with Tukey’s post-hoe
comparisons to test for the differences of forest and topographic characteristics across
forest age groups. For comparison, we performed all ANOVAs using the primary plots
only (simple random samples) and the neighborhood mean and standard error using the
Horvitz-Thompson estimator calculated from the adaptive cluster sampling method
(Thompson and Seber 1996, Turk and Borkowski 2005). To test for differences in site
characteristics between smaller sites and larger sites, we performed nested ANOVAs with
age-class as the main factor and size-class nested within.

We used two-way contingency tables along with the binomial proportion test (Ott
1993) to test for differences in the proportion of feces in forests of various ages and sizes.

Proportion of feces in harvested areas was compared to that in unharvested sites. We

26

then used the binomial proportion test (Ott 1993) to determine significant differences
between proportions of feces present at sites.

To enhance the interpretation of contingency table results, we used logistic and
autologistic regression by incorporating patch age and harvest size variables as well as
forest and geographic characteristics to determine which were important in predicting
panda feces presence. We constructed two independent logistic regression models for
each age class, both of which used feces as the response variable. One logistic regression
model used data from the primary plots only while the second logistic regression used all
primary and secondary plots (from the adaptive cluster sample). Prior to running each
logistic regression analysis, we conducted univariate logistic regression analysis (PROC
LOGISTIC, SAS Institute) and removed all variables that had P values less than 0.25
associated with the likelihood ratio test scores (G-tests). This somewhat liberal P vaer
was chosen based on the distribution of P values of the variables. We then ran a
correlation analysis (PROC CORR, SAS Institute) on remaining variables to detect
collinearity. Pairs of variables that had r > 0.6 were considered for variable elimination.
We eliminated the variable with the less significant univariate G-test score. If scores
were similar, we eliminated the variable that led to the most parsimonious multivariate
logistic regression model. We tested a variety of logit models by deliberately adding
certain topographic and forest variables and determining which made a significant
contribution to the prediction of feces presence. Akaike information criteria (AIC) were
used to determine which model had the best fit (Quinn and Keough 2002).

Finally, classification trees (Venables and Ripley I999, De'ath and Fabricius

2000) were used to nonparametrically explore the data and provide further insight into

27

how forest and topographic factors, as well as age and size of harvest, may influence
giant panda habitat use. Classification trees were useful because throughout the analysis.
all possible binary splits were considered for all of the predictor variables, yet the split
that was created was the one that produced two groups with the smallest within group
sums-of-squares in the response variable (Quinn and Keough 2002). Four separate trees
were performed on: 1) all independent variables, 2) the age of forest and size of forest
patch, 3) geographic variables including distance to road, distance to house, elevation,
slope, ecological aspect, and basal area, and 4) forest characteristics, including
components of the overstory, midstory, shrub, bamboo, and understory layers. All
classification trees used a minimum node size of 5 with deviance=0.05 and incorporated a
cost complexity pruning using deviance pruning methods to return a minimal node tree as
implemented in S-Plus (Venables and Ripley 1999, De'ath and Fabricius 2000). Tree
sizes were selected after running a series of 10-fold cross-validations using the l-SE rule

(De'ath and F abricius 2000).

M
We sampled a total of9l3 30 x 30 m sample plots (443 primary and 470
secondary) during the 2001-2003 sampling period. We found significant changes in all
forest and topographic characteristics over time except average understory and shrub
height, which did not vary over time (Table 2.1). Results from both the simple random

sample of primary plots and the Horvitz-Thompson mean estimations of neighborhoods

28

Table 2.1. Univariate ANOVA results for forest and geographic characteristics measured

in primary plots only and neighborhoods (primary + secondary plots). Neighborhood

calculations were based on Horvitz-Thompson means of all plots in that neighborhood.

 

Variable

Elevation

Distance to Roadway
Distance to Household
Slope

Ecological Aspect

IBasal Area

Total Overstory Cover
Conifer Overstory Cover
Deciduous Overstory Cover
Avg. Overstory Height
Total Midstory Cover
Conifer Midstory Cover
Deciduous Midstory Cover
Avg. Midstory Height
Total Shrub Cover

Avg. Shrub Height
Bamboo Percent Cover
Total Understory Cover
Avg. Understory Height

 

 

 

Primary Plots Only Neighborhood Means
F df p F df p

54.65 3,437 <0.001 54.65 3,437 <0.001
97.66 3,437 <0.001 97.9 3,437 <0.001
106.97 - 3,437 <0.001 106.96 3,437 <0.001
8.03 3,436 <0.001 9.01 3,436 <0.001

8.5 3,436 <0.001 8.83 3,436 <0.001
84.11 3,435 <0.001 69.76 3,435 <0.001
84.86 3,422 <0.001 85.52 3,422 <0.001
34.4 3,395 <0.001 31.72 3,395 <0.001
20.05 3,394 <0.001 23.47 3,394 <0.001
110.79 3,396 <0.001 111.02 3,396 <0.001
33.98 3,435 <0.001 36.39 3,435 <0.001
4.61 3,396 <0.004 4.41 3,396 <0.005
30.05 3,396 <0.001 34.02 3,396 <0.001
69.47 3,418 <0.001 75.66 3,419 <0.001
3.36 3,419 0.019 3.62 3,420 <0.001
40.87 3,396 0.013 41.96 3,396 0.013
57.75 3,434 <0.001 57.76 3,431 <0.001
26.67 3,417 <0.001 27.84 3,418 <0.001
2.25 3,412 0.02 1.76 3,413 0.15

 

 

29

 

Basal Area (sq m/ha)

100

 

 

 

 

 

 

- 0-10 years: < 10 ha
12:1 0-10 years: 10-100 ha
80 - - 11-30 years: < 10 ha
:1 11-30 years: 10—100 ha
,3 - 31-100 years: < 10 ha
2?, m 31-100 years: 10-100 ha
53 60 - - Old-grth ‘
> E .
O i
E o '
(U vo-a
c ..
g 40 4 '
03 f
a. .
.. .- i .1
20 a a. II if
II I 1| é
,_ I l ,
33 get «6‘ v6‘ 009
a\ O 0 00 \5‘3
gas ago “we, 5&0“ Bamboo 30390

 
 
   
   
    
   
   

2500

_ 2000

~ 1500

_ 1000

_ 500

 

Figure 2.1. Mean values (iSE) of topographic and forest characteristics for 30 x 30

meter sample area in categorized age of forest and size of harvest categories (size of

patch for unharvested).

30

Distance to Road

("1)

were generally the same for all forest and topographic characteristics using ANOVA
(Table 2.1). Old harvests (31-100 years) and old-growth sites occurred significantly
farther away from the roadway, farther into the interior forest, than newly (0-10 years)
and mid-aged (1 1-30 years) harvests (Figure 2.2). Old harvests and old-growth forest
also had significantly higher basal areas than younger forests. Overstory trees were
tallest in old harvests and old-growth forest, and significantly shorter in the most recently
cut areas, especially in large cuts (Figure 2.2). Conifer, deciduous and total overstory
percent cover all increased as the forest aged (Figure 2.2, Appendix 1). Total overstory
canopy cover increased through the age classes and was significantly higher in old
harvests and unharvested areas than middle-aged and new harvests (Table 2.1, Figure 2.2,
Appendix I). The conifer component of the overstory was greatest in the large,
unharvested areas but was significantly lower in areas that were cut 0-30 years ago (Table
2.1, Appendix 2.1), while the deciduous overstory was lower in new harvests and highest
in large, old harvests (Table 2.1, Appendix I). Midstory percent cover generally
increased as the forest aged, with one exception. Both conifer and deciduous midstory
coverages were lowest in new harvests and highest in old harvests (Table 2.1, Figure 2.2).
The lowest mean bamboo percent cover was in large, new harvests while higher bamboo
cover was found in old-growth areas. Smaller. unharvested patches had the highest
amounts of bamboo (Figure 2.2).

Age of harvests in Wolong ranged from several months to approximately 100
years. Older harvests were typically associated with large-scale timber harvesting

operations and newer harvests associated with multiple tree removal for fuelwood and

31

small-scale timber harvesting. The age of the harvested forest had a significant effect on

giant panda use (primary 1,2 =58.7, p=0.01; neighborhood Z32 =1 15.1, p=0.01).
Size of patch (primary 1,3 =0.2. p=0.83; neighborhood 1,2 =0.37, p=0.6) and size of

harvest (primary 1,2 =0.01, p=0.97; neighborhood 1,2 =0.41, p=0.5) indicated no

difference between the proportion of feces found in small and large patches and fuelwood
collection and timber harvest sites. Patch sizes were significantly related to distance from
roadway, with smaller sized patches occurring close to the roadway (0.06 ha, 0-500 m
from roadway), and larger patches farther from the roadway (0.29 ha, 3-3.5 km from
roadway)(Fg,362=6.159, p<0.001). In harvested areas that were 0-10 years old, use was
not detected in the larger harvest sizes (10-100 ha) (Appendix I).

In primary plots, pandas did not use larger harvested areas (IO-100 ha) in new
harvests (Age = 3.0 i 0.6 years)(mean : SE)(Figure 1). Use generally increased as the
forest aged (X20, #99): 24.9, p < 0.001). There was some use in large mid—aged forests
(22.1 i 2.1 years), but it was not significantly higher than new harvests. Use was 5.0 and
4.7 times higher, however, in mature harvest (50.0 i 2.3 years) and old-growth forests
(203.6 : 3.6 years), respectively, than in mid-aged harvests (Figure 2.2).

For primary plots, in smaller harvested areas (< 10 ha), use in old-growth forest
(180.7 : 4.3 years) and mature harvests (44.6 i 1.7 years) was 8.0 and 5.3 times higher
than in new harvested forests (3.7 i 0.4 years), respectively (X20, N=295)= 35.2, p < 0.001),
suggesting even small-scale tree removal, such as single-tree selection timber harvesting
and fuelwood collection, may reduce use by pandas for decades (Figure 2.2).

Overall, in primary plots we found giant panda feces more frequently in older

forests than in newly harvested areas (X29, #393): 56.3, p < 0.001). The reason is that as

32

0.5

 

 

  
  
   

 

 

 

 

 

 

 

 

 

24
to [:3 10 -100 ha
8 [:1 Overall
<0
“- 0.3 -
5 .
.g 1 .
c g
E 0.2 ~ .
Q
CL
* + ----- 7,7,5.
91
45 ,,
0.0 - I “3

 

 

 

 

0-10 11-30 31-100 Old Growth

Age of Forest (Years)

Figure 2.2. Proportion of primary plot sites with giant panda feces in areas with varying
forest ages and sizes. Italicized numbers above bars represent plot sample size. Symbols
(*, **, +, ++) above bars represent significance groupings in the proportion of survey
plots with panda feces by age class for each size using binomial proportion test. The same

number (and color) of symbols above bars indicates no significant difference.

33

the forests recovered, basal area, overstory forest and bamboo covers increased
(F3_390=72.8, p<0.001; F3,375=60.9, p<0.001; and F3,335=43.0, p<0.001, respectively). The
increasing forest cover and bamboo cover created a more suitable environment for the
giant panda (Figure 2.2).

For each forest age class, we performed four varieties of logistic regression using
the primary plots only and all plots in the neighborhood (including primary and
secondary plots) with both logistic and autologistic (includes autocovariate term)
regression techniques (Table 2.2). We believe the autologistic model that includes all
neighborhood plots is the most reliable model because it includes a high number of plots
while accounting for spatial autocorrelation through the use of the autocovariate term.
The logistic model that includes only primary plots is the next most reliable model
because it uses a lower number of primary plots but has low effects of autocorrelation.
The logistic model that uses all neighborhood plots is the least reliable because it may
have a high amount of autocorrelation in the model that is not accounted for.

In recently harvested forests (0-10 years), presence of bamboo was the most
significant indicator of giant panda feces presence, as well as shallow slopes and distance
from households and roadways (Table 2.2). In mid-aged forests (1 1-30 years), bamboo
cover was also a significant indicator of panda feces in the logistic and autologistic-
primary models, but the deciduous component of the overstory and midstory and distance
from roads were also important (Table 2.2). In older forests, high bamboo cover
continued to predict the presence of giant panda feces in addition to southwest aspects.
longer distance from roadways, higher conifer midstory cover, and lower understory

height. In old-growth forests, higher bamboo cover, low slope, low ecological aspect

34

Table 2.2. Significance of forest and geographic characteristics in logistic and

autologistic models developed from primary plots only and Horvitz-Thompson models

developed from all neighborhood plots (primary and secondary plots).

 

 

 

 

 

 

 

 

 

 

 

 

 

LOGISTIC AUTOLOGISTIC
AIC Variable Estimate P-value AIC Variable Estimate P—value
0-10 years
Primary Plots 28.36 Intercept -6.14 <0.01 27.39 Intercept -8.54 <0.01
Bamboo cover 0.05 0.02 Bamboo cover 0.05 0.02
Distance to Road 0.00 0.12
Euclid Dist. 1000 m 622.70 0.11
Neighborhood $7.25 Intercept -9.06 <0.01I60.48 Intercept -6.56 <0.01
Avg. Midstory Height 0.13 0.01 Slope -0.08 0.03
Bamboo cover 0.08 0.01 Bamboo cover 0.05 0.01
Distance to House 0.00 0.02
Euclid Dist. 750 m 130.20 0.01
1 1-30 years
Primary Plots 38.42 Intercept -3.59 <.0001 31.43 Intercept -4.93 0.00
Bamboo cover 0.03 0.02 Bamboo cover 0.04 0.04
Euclid Dist. 750 m 95.90 0.01
Neighborhood I83.88 Intercept -5.04 <.0001156.4rlntercept 44.40 <0.01
Decid. Overstory Cover 0.04 0.02 Decid. Overstory Cover 0.08 0.01
Decid. Midstory Cover 0.05 <0.01 Decid. Midstory Cover 0.07 <0.01
Bamboo cover 0.03 <0.01 Distance to Road 0.01 <0.01
Euclid Dist. 750 m 324.00 <0.01
31-100 years
Primary Plots 56.51 Intercept -3.23 <.0001 29.36 Intercept -2.05 0.03
Distance to Road 0.00 <0.01 Conif. Midstory Cover 0.12 0.05
Avg. Understory Height -0.08 0.031
Euclid Dist. 50 m 183.40 <0.01
Neighborhood 109.71 Intercept -3.28 <0.01 159.87 Intercept -2.86 <.0001
Ecological Aspect -0.16 <0.01 Ecological Aspect -0.06 0.05
Bamboo cover 0.05 <.0001 Bamboo cover 0.04 <.0001
Avg. Overstory Height 0.06 0.02 Euclid Dist. 50 m 39.42 0.01
Old-growth
Primary Plots 93.58 Intercept -6.52 <0.01 51.93 Intercept -1 1.96 <0.01
Avg. Overstory Height 0.23 <0.01 Avg. Overstory Height 0.27 0.01
Conif. Overstory Cover -0.05 <0.01 Slope 0.12 0.03
Decid. Overstory Cover -0.05 0.02 Euclid Dist. 500 m 351.70 <.0001
Distance to House 0.00 0.02
Neighborhood 378.01 Intercept -1.91 0.01 3§28 Intercept -3.57 <0.01
Slope -0.07 <.0001 Slope -0.05 <0.01
Ecological Aspect -0.10 <0.01 Ecological Aspect ~0.08 <0.01
Avg. Overstory Height 0.07 <0.01 Avg. Overstory Height 0.11 <.0001
Decid. Midstory Cover 0.02 0.01 Decid. Overstory Cover -0.01 0.08
Avg. Understory Height 0.01 0.00 Bamboo cover 0.01 0.04
Patch Size -1.79 0.00 Avg. Understory Height 0.02 <.0001
Distance to House 0.00 0.00 Patch Size -2.15 <.0001
Euclid Dist. 1250 m 320.40 <.0001

 

 

35

(South-Southwest), tall overstory trees with a low deciduous component, and a far
distance to households were the best predictors of giant panda feces (Table 2.2).
Euclidean distances for autologistic models in old-growth forest varied from 500-1250
meters.

Classification trees were used to provide further understanding as to which forest
and geographic characteristics are used by the giant panda (Figure 2.3). Classification
trees using the Horvitz-Thompson means achieved the lowest misclassification rates (12-
17%) with models using primary plots only also having low misclassification rates (I 3-
17%). Errors were higher when using all neighborhood plots (primary and secondary
plots) (misclassification rates: 24—32%) and no weighting was factored in for the spatial
autocorrelation inherent in the dataset. Therefore, results from these classification trees
were considered less reliable and are not included in this discussion. The first
classification tree used all independent variables to determine the best overall model that
characterizes the presence of giant panda feces. Both primary and primary + secondary
Horvitz-Thompson model types selected bamboo percent cover as the primary
distinguishing factor, with cover percent ranging from 14-17.5% (Figure 2.3). Other
useful factors included being a kilometer from households, a tall overstory (>26m) and a
high basal area (>0.13 mz/ha). When considering age of forest and size of forest patch
only, primary and Horvitz-Thompson models predicted similar results due to their
corresponding datasets. Nevertheless, a forest that has had 38 years to regenerate from a
harvesting event is more likely to have giant panda feces (Figure 2.3). When considering

only geographic and topographic characteristics of plots, distance from households

36

Figure 2.3. Regression trees of A) all forest and geographic characteristics, B) age and
size of harvested and unharvested patches, C) geographic characteristics, slope,
ecological aspect, and basal area (mz/ha), and D) forest characteristics. Trees show
classification based on the presence of giant panda feces. Residual mean deviance = A)
0.175, B) 0.183. C) 0.186, and D) 0.159. Values at tree leaves represent proportion of

plots with feces and sample size (within parentheses).

37

 

6:25 332:.

as Ed 2 2 v osm

 

2 o: v 85 so So

 

 

$232 28» 2 v s?

 

 

memo» msm v ow<

 

 

aamuzv E _ .o ”cam .25 8:35:32 .882

3:25 mum—V86

:9 So 2 3 v 85

 

a. a: v 8% :9 Bo

 

 

$232 was» a. v 8<

 

 

memo» msm v ow<

 

 

nomnzv E _ .o sex cam cofiocasaa: zoo:

 

 

 

 

EC: :3 33 So
A3; Gd 55:. :6 V «0.50 Emmm
a mom v 2%:
@9330 owfio>< :3: Ed

 

 

o\om.b_ V ._®>OU 0098mm

 

 

 

sanzv N _ .o nose 85 Seaeasoaz .882

 

o._ 38 mmd

E v._m~m v umso: 2 3:320 AS: mmd

 

 

 

 

 

E wHN v EwTJ: b93920 .w>< 3.2 35

 

5.24: v om8:2 3:35 325.;

 

 

exp: v 3.5.9 oonEmm

 

 

3312C is cam .25 aeaoEasoaz .882

 

 

Aime—58m + .03:th
mZ<m=2 DOOIMOmIOEZ ZOmmZOIH-N.E>~—OI

>420 wHOtE >m<_>:~E

38

 

 

 

 

 

 

 

 

23...: 8225
At oo o :9 mm c $8 on o {3.3 v 5.60 oonEnL
9. i. am v 5.80 Eu n mm v BED
Cemego 3:500 535 .55
E mom v 2301 .CEmB>O w>< Com. 5 o

 

 

 

3.2.— v 550 ooh—Sm

 

 

 

$3.16 26 ”8mm hohm— :o_EoEmmm_om__>_ Eco—2

a: 33 8Q med SE Ed at cmd

 

 

 

 

 

an: v sad so 2: v awe:
.583 soaeoo BEES 88.2.
a QR v awe: .0285 es. 3...: :3
as: v 8.8 oscam

 

 

Gomuzv Ed ”83* E25 :othEmmn—ofl—z .2022

 

 

$20.3 Ann—.ovc

3: v .83. aoaeoom a; 3

 

 

ESE w: v 32 .33 32 3o

 

 

 

SSE 3 o v 82 3am i8 So

 

 

 

 

E 0.3: v 3.5: 2 8535

 

 

a 9.123 2 .o Noam Loam cofiocaaaaz coo:

 

Am_ v mod 35 36

E mwmom v cones—m So: NNd

 

 

E 0.3—N v 33oz 2 35.35 3: cod

 

E 2.8m v 33. 2 SEED Ammmv mod

 

 

E wit: V BE: 9 3535

 

 

3:123 m _ .o use 55 ceaéeaaez .282

 

 

Abmwcgom + Ema—EA:
mz<m2 DOOIMOmEOEZ ZOmmEOIh-N._._>mOI

>420 mqum >m<3=~E

39

(>l 100 m) was the most important factor in determining panda feces presence (Figure
2.3). Additionally, a neamess to roadways (<340 m) was negatively related to feces
present and a forest with low basal area (<0. 125 mZ/ha) was also not used. Classification
trees on forest characteristics not only reiterated the importance of bamboo cover (>14%)
and overstory height (>26 m), but also indicated a deciduous component to the midstory
(>l 1.5%) may have higher feces presence, or if overstories are tall (>26 m), the conifer

overstory component should not be too great (<59%) (Figure 2.3).

Discussion

Timber harvesting and fuelwood collection have created a patchy mosaic of
forests in various stages of succession throughout Wolong Nature Reserve. This has
several implications for the use of these habitats by the giant panda and also for the
conservation of this endangered species. Previous studies on panda use of harvested
areas and the surrounding forests show that these activities decrease use because of lower
canopy cover, increased bamboo densities, decreased bamboo seedling densities and
changed bamboo growth form (Schaller et al. 1985, Reid and Hu 1991, Reid et al. 1991c,
Taylor and Qin 1993a, b). However, the majority of those studies were conducted during
or shortly after the arrow bamboo flowering and subsequent die-off in 1983. During the
die-off period, giant panda activity was altered because of resource limitations, especially
those associated with food availability. Our study was performed at a time when bamboo
populations would have recovered from the flowering event (i.e., approximately 15 years

after flowering) and returned to normal, pre-flowering densities (Taylor et al. 1991).

40

Information from this study therefore shows patterns of use during a period when the
bamboo populations are high and therefore may not be a limiting factor to pandas.

Our study investigated the influences of fuelwood collection and timber
harvesting directly by comparing use in harvested areas of various ages to unharvested
forests. Use of forested habitats was lowest in newly harvested areas (O-lO years) and
highest in unharvested areas and old harvested areas that had been cut between 30-100
years ago. The high amount of feces in old harvests was notable because previous studies
have found use to be lower (Schaller et al. I985) and habitat to be less suitable (Schaller
et al. 1985, Reid and Hu 1991, Taylor and Qin 1997) in harvested areas.

There are two primary reasons why the proportion of feces in old harvested areas
is as high as unharvested, old growth forests, which has traditionally been considered the
primary habitat for the giant panda. F irst, early harvesting occurred in highly suitable
habitat and less suitable habitats (i.e., steep slopes) were not harvested because of
difficulty extracting the wood. Second, several decades of forest regeneration has created

a forest cover that provides more suitable habitat characteristics for the giant panda.

Highly suitable habitats occur commonly within the interior of Wolong and are
therefore significant for movement, foraging, and reproduction. Despite their
significance, use decreased after clear-cut harvesting due to the associated change in
forest characteristics, including decreased overstory canopy cover, decreased conifer
component, increased hardwoods component, and increased bamboo densities (Schaller
et al. 1985, Taylor and Qin 1988b, Taylor and Qin 1988c, Taylor and Qin 1989, Taylor
and Qin 1997). We found the overstory and midstory conifer component of forests ll-30

years after harvest to be similar to. and in some cases higher than, unharvested forests

4l

(except in the large harvested areas), indicating establishment of conifers. The deciduous
component of the overstory and midstory, as well as bamboo densities, were also
comparable between old harvests and unharvested forests. Because these forest
characteristics do not differ from unharvested forests we can assume that the giant panda
would begin using these areas sooner and more frequently than if it were within a less
suitable habitat area that was harvested at the same time.

The primary objective of historic harvesting was to cut the large conifers (Abies,
Picea, and Tsuga) that grew within the subalpine coniferous forests of Wolong. Once in
a stand, trees would be selected based on their overall size and the ability to access the
tree and transport it out to the nearest road. Since site quality was highest on the shallow
slopes, and because these areas were easier to access, harvesting was most frequently
done on the interior ridge tops and nearby areas with low slope. However, the giant
panda also prefers sites with low slopes (Reid and Hu 1991) because there is a large
increase in the metabolic requirements to move up steep inclines (Taylor et al. 1972).
For this reason, we would also expect giant panda use to be higher in the old harvest sites
since after these sites become marginally utilizable, they would be exploited by the panda
who is trying to conserve energy. On the other hand, steep areas were often left uncut by
early loggers since it was difficult to access precipitous forests and the ability to move the
timber was difficult and often dangerous. To the giant panda, steep, uncut forests are not
ideal for the same metabolic reasons and tend to be underutilized, helping to explain why
use in unharvested areas was less than that of old harvests.

Another reason why the proportion of feces is high in old harvests is because

early timber harvesting was done within the higher elevation (above 2700 m, Table 2.1)

42

subalpine coniferous forest. Elevations here include the primary vertical range of the
giant panda as well as its primary food, arrow bamboo (Schaller et al. I985). In addition,
higher elevation forests are not directlyjuxtaposed along the Pitiao River valley, where
the majority of local people reside. Therefore access to these regions by humans is more
difficult, allowing areas to remain somewhat undisturbed since the harvesting events
occurred. We should note that humans still use these areas intermittently for wildlife
research, herb collecting, bamboo gathering, and occasionally for snaring musk deer

(currently illegal; personal observation).

In contrast, newer harvested areas typically occur in lower elevations near the
Pitiao valley (below 2600 m, Table 2.1), where human disturbance is more common and
the less preferred umbrella bamboo grows (Schaller et al. 1985). Lower elevation forests
have a complex site history that stems from intense human use over the centuries. This is
especially true in areas where illegal fuelwood collection, poaching and medicinal herb
collection are still active. These activities have created a landscape in the lower
elevations that is heterogeneous and frequently visited and disturbed by humans. This
may help to explain why use is low in these areas.

In the thirty to one hundred years since the old harvest areas were cut, most sites
have been left to regenerate and many have succeeded to the understory reinitiation stage,
where a forest floor stratum of herbs, shrubs, saplings and poles has developed (Oliver
and Larson 1996). Bamboo stands are also affected by the change in environmental
variables as the forest progresses through succession (Taylor and Qin 1989, Reid et al.
l99lc). Therefore, the high amount of use in the old harvest areas may also be explained

by the relationship between overstory canopy density and bamboo cover percent that

43

coincides with the findings of Taylor and Qin (1997), with the frequency of feces
increasing as forest cover and bamboo cover percent increase. As the forest recovered
from the harvesting activity, and as the midstory and later the overstory cover percent
increased, feces was found more often in the harvested area, as demonstrated by the
increased proportion of use in old harvested areas. Old harvested areas ranged in age by
40 years, making a detailed recovery timetable unavailable. However, it can be seen that
as the forest progresses through succession following a harvesting event, suitable habitat

can be restored.

Parsimonious logistic regression models demonstrate that bamboo. percent cover
continues to be the best predictor of giant panda habitat. In newly harvested forests, the
traditional characteristics of panda habitat (i.e., high bamboo cover, low slope, and far
distance from households (interior forest)) were found to be the best predictors of giant
panda feces. As the forest aged (1 1-30 years), the deciduous component of the midstory
and overstory became important indicating the significance of regenerating forests and
how the cooler, shaded forests provided by the deciduous trees may be a useful
component of the forest as pandas move through the lower elevation forests that have
been harvested more recently. Old harvested areas that were used by pandas had bamboo
cover, were farther away from the roadway (more in the interior forest), had a low
understory, some conifer in the midstory, and a SSW aspect (low ecological aspect).
Unharvested and old-growth forest had the most complex characterization of forests used
by giant pandas. Logistic and autologistic regressions indicate smaller forest patches

with more bamboo. taller understories, less conifer and deciduous trees in the overstory,

44

and low slope with SSW aspect (low ecological aspect) to be the old-growth forest with

the greatest likelihood of encountering panda feces (Table 2.2, Figure 2.3).

There were some differences in the variable selection between the logistic
regression that used only primary plots and the autologistic regression that used
neighborhood (primary + secondary plots) means. However, both methods led to a
characterization of a similar type of forest that is selected by pandas. Both methods also
showed how this selection changes as the forest ages afier harvesting events. A newly
harvested forest (0-10 years) requires the presence of bamboo for pandas to use these
areas. A forest of l l-30 years also should have bamboo but may also be used if it has a
deciduous component in the midstory or overstory, indicating a cooler forest that is
farther from the roadway is most preferred. There were discrepancies between the
logistic and autologistic models in the characterization of where panda feces occurs in old
harvested forests (3 I-IOO years). Logistic regression of primary plots indicated a
significance to interior forest (a far distance from roadways), while the autologistic
regression showed that bamboo cover, a southern aspect, conifers in the midstory, and
low understory are most important for giant pandas. Both logistic and autologistic
regressions suggest that forests with tall overstories with a low deciduous component in

the overstory is significantly related to the presence of panda feces.

We applied a series of autocovariate terms associated with various Euclidean
distances in each of the autologistic regressions and used stepwise procedures to select
the best distance associated with panda feces presence. Euclidean distances provided a
unique way to determine the spatial level that pandas may be selecting the forest

characteristics of their environment and may be useful when determining the connectivity

45

of panda habitats. Euclidean distances selected as significant predictors of feces presence
for each autologistic regression model varied by forest age. In younger forests aged 0-30
years, a significant Euclidean distance of 750-1000 meters indicated that pandas use these
forests on a larger spatial level, perhaps roaming widely from one usable forest patch to
another. As the forest matures (3 l -l 00 years), the amount of forest that is usable by
pandas increases. Here, the Euclidean distance of the autologistic regressions of both
primary and neighborhood plots decreases to 50 meters, demonstrating that pandas begin
concentrating their activities on forest patches with the most suitable habitat. These
distances increase in old-growth forests, suggesting that old-growth forests are widely
usable and pandas do not need to concentrate activities in these areas, but may traverse

across these forests.

The findings of this study are significant because they suggest that panda habitats
may become usable in a rather short period of time. If forestland is maintained and not
converted to agriculture or grazing land, forests that were previously harvested may
eventually regenerate and increase the potential for the species long-term survival. This
potential for regeneration of habitats of species that prefer old-growth forests may be

significant when planning for the long-term conservation of other species worldwide.

46

CHAPTER 3

BAMBOO CHARACTERISTICS AFTER FOREST HARVESTING

IN WOLONG NATURE RESERVE

47

CHAPTER 3
BAMBOO CHARACTERISTICS AFTER FOREST HARVESTING
IN WOLONG NATURE RESERVE

Introduction

Logging activities such as timber harvesting and fuelwood collection can have
various effects on understory vegetation (Fredericksen et a1. 1999, Fredericksen et a1.
2000, Scherer et al. 2000, Frey et al. 2003). Removal of overstory and midstory canopy
trees alters understory microclimate (Aust and Lea 1991, Laporte et al. 2003) and soil
chemistry (Kim et al. 1995, Goodale et al. 2000, Goodale and Aber 2001, Piirainen et al.
2002, Carmosini et al. 2003, Laporte et a1. 2003, Howard et al. 2004). These effects may
change the structure of understory vegetation already present or alter the species
composition on the site (Duffy and Meier 1992, Roberts and Gilliam 1995a, b, De
Grandpre and Bergeron 1997, De Grandpre et al. 2000, Roberts 2002, Roberts and Zhu
2002). Changes in understory characteristics by harvesting may then have dramatic
effects on the habitats of resident plant and animal species.

Bamboo (Poaceae) is a ubiquitous understory species found throughout much of
the world (Linderman et al. 2004). Bamboo species are often a significant component of
forests because they compete extremely well with trees (Tanaka 1988, Cao I995,
Takahashi and Kohyama 1999) and have close interactions with other forest components
(Takahashi et al. 2003, Ito and Hino 2004, Noguchi and Yoshida 2004, Taylor et al.

2004). In addition, many wildlife species rely on bamboo as understory habitats (Reid et

48

al. 1991a, Johnson et al. 1993, Fidgett et al. 1996, Choudhury 2001, Pradhan et al. 2001,
Matsubayashi et al. 2003, Sakuragi et al. 2003).

Understory bamboo is a primary componenet of habitat for the giant panda
(A iluropoda melanoleuca), which relies on bamboo as its primary food source (Schaller
et al. 1985). Timber harvesting and fuelwood collection may therefore significantly
affect giant panda habitat by influencing the local bamboo populations (Linderman et al.
2005b). To properly conserve giant panda habitats, ecosystem management in these
areas must consider the interactions between bamboo and forest dynamics (Reid et al.
1989, Reid and Hu 1991, Taylor et al. 1995, Noguchi and Yoshida 2004).

It is critical to understand the relationship between various types of logging and
bamboo populations because many people living near highly suitable panda habitat rely
on fuelwood collection for heating and cooking while the same forests provide vital
habitat for the giant panda (Liu et al. 1999b, An et al. 2001, Liu et al. 2001, Ruiz-Perez et
al. 2001, An et al. 2002, An et al. 2003, Linderman et al. 2005a). Timber harvesting
affects not only the spatial distribution of bamboo (Linderman et al. 2005b), but also
various growth and population characteristics of bamboo species. Previous studies have
shown that bamboo stem densities increase (Reid et al. 199 l b), seedling densities
decrease (Taylor and Qin 1993b) and germination, establishment, and overall
regeneration from seed were lower in clearcuts (Taylor and Qin 1988a). When large
openings in the canopy are formed, such as those created by clearcut harvesting
operations, the density of arrow bamboo increases to such a degree that tree
establishment is severely hindered (Schaller et al. 1985, Taylor and Qin 1987, Taylor and

Qin 1988a, Reid et al. 1991b, Taylor and Qin 1993a, b). Because of this encumbrance.

49

regenerating forest is slow in clearcut areas. Selective logging, on the other hand, may
maintain enough deciduous overstory cover to continue to provide suitable habitat for the
giant panda (Reid et al. 1991b, Taylor and Qin 1997).

Recently, several regions around the world (including the Amazon, Africa,
Siberia) have taken steps to limit the effects of forest harvesting via the use of logging
bans. In Asia, many countries (e.g. the Philippines, Thailand, Vietnam, and China) have
implemented bans on timber harvesting and fuelwood collection in an effort to curb past
overexploitation of forest resources (Durst et al. 2001). Harvesting bans have the
potential of conserving wildlife habitats by allowing the forest to retain standing
overstory and midstory densities. However, species that rely on understories in these
areas may continue to be affected for some time into the future as previously logged
forests regenerate and competition for light, moisture, and space resources evolves
through successional stages.

The Chinese government has implemented new policies in an effort to eliminate
harvesting of natural forests within its borders, and particularly in nature reserves (Durst
et al. 2001, Zhao and Shao 2002, Liu and Diamond 2005). The primary policy, the
Natural Forest Conservation Program (N FCP), includes a logging ban that halts all timber
harvesting and fiielwood collection in natural forests throughout China (Zhao and Shao
2002, Liu and Diamond 2005). This policy has the potential of influencing the
conservation of giant panda habitats because it will decrease the amount of illegal
harvesting thereby increasing stand volume and overstory and midstory canopy covers.
This, in turn, may allow bamboo to return to preharvesting densities and growth form and

forests to progress to an old-growth state, both of which are preferred by the panda.

50

To determine how fuelwood collection and timber harvesting influence the
characteristics of understory bamboo populations, we studied bamboo culm dynamics in
Wolong Nature Reserve, Sichuan, China. The objectives of our study were to determine:
1) how different bamboo species are influenced by different sizes of harvesting
(fuelwood collection typically in smaller and timber harvesting often in larger areas), 2)
how long these effects last, and 3) which forest and geographic characteristics are
associated with productive bamboo characteristics. Information gathered on how various
harvesting activities influence bamboo populations and culm characteristics will help to
determine more appropriate harvesting regulations as well as to conserve giant panda

habitats.

Methods

Data collection took place during May-August 2001, May-November 2002, and
June-August 2003. A sample plot consisted of a 30 x 30 m area. The center position of
each 30 m2 plot was gee-referenced with a Pathfinder® Pro XRS GPS unit (Trimble
Navigation Limited). Geographic and forest characteristics as well as bamboo percent
cover were measured over the entire 30-m2 area. Characteristics measured included:
slope, aspect, forest basal area (IO-factor prism), overstory height and percent cover.
midstory height and percent cover, shrub height and percent cover, bamboo percent
cover, and understory height and percent cover. We then established four bamboo
subplots measuring 1 m2 in each 30 m2 plot. The l m2 bamboo subplots were located 10

m from the center point in the four cardinal directions. We counted density of bamboo

51

stems within each of the four 1m2 subplots, and measured culm basal diameter and culm
height from 5 random stems in each subplot.

We monitored characteristics of three bamboo species in various aged forests
throughout the Wolong Nature Reserve. Three native, naturally occurring bamboo
species included arrow bamboo (Bashaniafabri), umbrella bamboo (F argesia robusta),
and yushania bamboo (Yushania chungii). We studied the first two species because of
their documented importance to giant panda foraging (Schaller et al. 1985, Taylor and
Qin 1987, Reid and Hu 1991, Reid et al. 1991b, Taylor and Qin 1993b). The third
species was identified to also be potentially important for giant panda habitat, especially
in higher elevation habitats, and was therefore added to the sampling design. In addition
to sampling areas with wild bamboo populations, we sampled areas with no bamboo for
comparison.

Documentation of timber harvesting and fuelwood collection history of specific
locations in Wolong is limited, therefore local guides knowledgeable of site histories
helped us to establish forest age in the field. We examined bamboo characteristics in
previously harvested forests of various ages and in uncut forests. Sites were grouped into
four harvest age categories based on their general forest stand ecology, our ability to
accurately identify forest ages as well as sample size issues: 0-10 years, 1 1-30 years, 31-
100 years, and unharvested forest.

In order to determine if a relationship existed between patch size and bamboo
characteristics, we examined both forest patch size (included unharvested forest patches)
and harvest patch size (did not include unharvested areas). Size of patch was difficult to

accurately measure in the field. We therefore estimated sizes of both forest and harvest

52

patches using a length and width measurement of distinct forest boundaries and
calculated size based on the resulting rectangle. Although we realize this type of patch
size measurement is a rough approximation, more descriptive forest and harvest patch
size measurements were infeasible given time constraints and the topographic relief of
the Wolong terrain. Both patch size and harvest size were grouped in small (< 10 ha) and
large (10-100 ha) size categories.

Complex site histories made it difficult to characterize any harvesting as a clear-
cut or partial cut, so no distinction was made between the two. Instead, we used size of
harvest to quantify type of cut. Smaller sized harvests are typically associated with

fuelwood collection, or selective cutting, and clear-cuts were often larger in size.

Statistical Analysis.“

We compared bamboo cover, density, culm height and width between species
using one-way ANOVA. In addition, we tested whether age of forest and forest/harvest
patch size had any relationship to bamboo characteristics using two-way ANOVAs for
each species. To determine how these characteristics changed over time after harvesting
events, we compared bamboo cover, stem density, culm height and width by age for each
species using ANOVA. The effects of various size harvests were tested by comparing
the same characteristics by size of harvest and size of patch for each species using
ANOVA.

We used regression techniques to test the relationship of changes in forest
characteristics and geographic characteristics to bamboo (B. fabri, F. robusta, Y. chungii)

culm characteristics. Initially, forest and geographic variables that were correlated (r2 >

53

0.60) were removed from analysis. We then used remaining variables in combinations of
single and multiple linear regressions to determine the variables that demonstrated the
most significant relationship to bamboo variables.

Significant forest and geographic variables from linear regressions were then used
in single and multivariate combinations to establish the best localized multivariate loess
model. Loess smoothing models were chosen because of their unrestrictive ability to
investigate the relationships between bamboo characteristics and forest and geographic
information. Variables chosen for the final loess models represent the forest and
geographic variables that best explain variation in bamboo characteristics and may be

used to determine which variables are most important in conserving giant panda habitats.

B<Lult§

We measured a total of 367 plots (30 m2 each) and 1468 bamboo subplots (l m2
each) throughout the sampling period. We found age of forest where each bamboo
species occurred to be significantly different ((F4,362=60.75, p<0.001). B. fabri occurred
in the oldest stands and frequently in unharvested, old-growth forests (Table 3.1). F.
robusta and Y. chungii were also in older forests. while areas with no bamboo were in
younger forests (Table 3.1 ).

Species of bamboo varied between forest patch sizes (F4‘354=5.04, p<0.001). B.

fabri and areas with no bamboo occurred in larger forest patches while F. robusta and Y.
chungii were found in smaller patches (Table 3.1). The size of harvest was also

significantly different among bamboo species. The largest harvest sites had either B.

54

Table 3.1. Mean age, patch and harvest size characteristics of three bamboo species and

areas with no bamboo in Wolong Nature Reserve. China.

 

 

 

 

Species N Age Patch Size Harvest Size

(years) (ha) (ha)
Bashania fabri Mean 83 147.5 21.0 29.4
SE 8.7 3.7 6.8

Fargesia robusta Mean 112 50.7 5.8 6.2
SE 6.2 1.0 1.2

Yushania chungii Mean 20 79.1 6.7 8.1
SE 21.1 3.8 5.0
No bamboo Mean 116 26.5 25.7 26.9
SE 3.9 5.4 5.7

 

55

 

 

segoeds ooqweg to
JQAOQ iuecued pue Atgsueg mats ueew

.250 Ghana 2382 mac—03 E 380% 02:53 3.588 Amm Hy 8558220 :32 ._.m 2:»;

83.0.569sz meowam oonEmm

5.8.x. E. 8%?ch €552; Egg»:

 
   

 

 

 

 

 

 

- N M
e
u
4 m
m. w.
mu H
- w m.
a w m
o w.
-. m p
m m
.8238 mega: g m
3.839 $39284 _liL - or W.
Ea magnum I m
1 NF

 

 

 

56

fizbri or no bamboo while F. robusta and Y. chungii were found in smaller harvested
areas (Table 3.1).

Mean bamboo percent cover across the 30 m2 parcels differed between species
(Figs—1161.7, p<0.001). All three bamboo species (B. fabri, F. robusta, and Y. chungii)
had significantly higher percent cover than areas with no bamboo indicating greater than
sparse densities. B. fabri had the highest percent cover, significantly more than F.
robusta (Table 3.1). Bamboo stem densities in 1 m2 subplots also differed between
species (F4‘3(,3=193.6, p<0.001). The three bamboo species had significantly higher
densities than areas with no bamboo (Table 3.1). Culm diameter and height were
different between species (F4,363=948.0, p<0.001 and F4,363=738.9, p<0.001, respectively),
with F. robusta being tallest and B. fabri being shortest (Table 3.1). Culm heights were
significantly different for all species except B. fabri. (Table 3.1). For culm width, F.
robusta was significantly more robust than both Y. chungii and B. fabri (Table 3.1). All
stem widths were significantly different for each of the species.

Few B. fabri subplots were available in newly harvested areas and none were
available in 1 1-30 year old forests due to the low frequency of this species in younger
forests, resulting in low sample sizes. For B. fabri, bamboo cover, stem density and culm
height were not statistically different among age classes (Table 3.2). However, bamboo
coverage and stem density were highest in old growth forests and lowest in newly
harvested areas. Culm width was the only B. fabri characteristic that changed since
harvesting, with wider culms occurring in newer forest and narrower culms in old growth
forests (F 176:3.32, p=0.04). Size of harvested patch did not have any effect on B. fabri

characteristics over time since harvest.

57

Table 3.2. Mean bamboo characteristics (iSE) over time since last timber harvest or

fuelwood collection activity.

 

N Bamboo Cover

Stem Density Culm Width Culm Height

 

 

 

 

 

(%) (#Im2) (mm) (m)
Bashania fabri
0-10 years Mean 8 52.38 42.91 4.20 0.92
SE 14.91 12.23 0.35 0.10
11-30 years Mean 0 n/a n/a n/a n/a
SE n/a n/a n/a n/a
31-100 years Mean 21 70.24 51.36 4.12 0.99
SE 5. 75 4. 74 0.20 0. 08
Old Growth Mean 54 81.28 65.24 3.94 0.87
SE 3.56 4.03 0. 08 0.04
Fargesia robusta
0-10 years Mean 41 47.17 14.55 12.84 3.49
SE 4.97 2.15 0.39 0.13
11-30 years Mean 29 56.41 14.03 11.60 3.57
SE 5.48 1.17 0.41 0.16
31-100 years Mean 22 48.18 19.77 10.38 3.18
SE 5.53 1.95 0.50 0.16
Old Growth Mean 20 61.83 15.47 11.76 3.66
SE 7.29 1. 88 0.45 0. 15
Yushania chungii
0-10 years Mean 5 71.00 32.95 6.97 2.07
SE 12. 88 7. 64 0.49 0. 17
11-30 years Mean 4 49.50 26.00 7.85 2.32
SE 14.61 7.21 0.53 0.32
31-100 years Mean 6 55.33 19.96 6.55 1.88
SE 13.26 5.34 0.34 0.09
Old Growth Mean 5 88.40 44.20 6.64 2.01
SE 6. 54 11.89 0.46 0. 08

 

58

 

F argesia robusta stands occur at lower elevations and are more prone to human
disturbance. This resulted in a higher number of Fargesia plots in newly harvested forest
and a lower number in old growth forest. F argesia robusta showed differences in stem
density (F3,108:4-039 p<0.01) and culm width (F3,.03=3.91, p=0.01 l) with forest age
(Table 3.2). Mean F. robusta cover percent was highest in old growth forest. Densities
were lowest in areas 0-30 years after harvesting, but increased considerably in the period
31-100 years after harvesting, and then generally declined as the forest reached an old
growth stage (Table 3.2). Culms were widest in newly harvested forests and generally
fluctuated in width afterwards. Size of forest patch was related to the percent cover of F.
robusta (F1,103=2.72, p<0.10), and may be an important characteristic when trying to
determine suitable giant panda habitat at low elevations.

Yushania chungii did not occur as frequently as the other two native bamboo
species and therefore had lower sample sizes. Yushania was found in relatively equal
proportions of all age classes, indicating it may be an opportunist species. It showed no
differences in its growth characteristics in different aged forests or in various sized
patches (Table 3.2), although mean cover percent and stem densities were high both in
newly harvested and old growth forests (Table 3.2).

Multivariate loess smoothing models of forest and geographic characteristics
ranged from predicting bamboo culm characteristics moderately well (F. robusta culm
height: r2=0.48) to very well (Y. chungii cover %: r2=0.97). Slope and elevation occurred
most frequently, helping to predict B. fabri percent cover, stem density, culm diameter
(elevation only), and culm height; F. robusta stem density (elevation only), culm

diameter

59

Table 3.3. Multivariate loess model results showing most significant forest and
geographic variables that relate bamboo stem characteristics. Variables are listed in
alphabetic order: Age = forest age since last harvest, Basal Area = forest basal area (10
factor prism), BBC over = bamboo cover (%), Elevation = MSL elevation, Slope = slope
(degrees), Size =forest patch size 01a), T otOverCover = total overstory cover (%),
TotalShrubC over = total shrub cover (%), and TotUnderC over = total understory cover

(on).

60

 

 

 

880880.92 .. 880mm .. 82 .880 5.0 $8 .88: 5.30.
8802.80.88» .. 8.88.0 .. 8:. .880 80 mg 888.0 5.30
88.0 . 880mm no.2 8.0 88.80 8.0
0050.09.20... .. co_~m>0_m :4. 3 Bd .050 200.00 .000 0.2059»
8.608828 . 88.0 .. 8.0 .. 8.888. Ed 88 .88.. 5.30
8808828 .. 88.0 . 8.0 .. 8.88.0 mom .3 888.0 2.30
88085.8 .. 8.0 .. 8.88m. .. 8.6000 8.8 ad 8880 8.0
0N5 .. 00>00000c206h .. 0050025404008. .. 002. _000m 24.3 Ed 060 0:00.00 000200. 03090.4
88.0 .. 8.88m. .. 8:... 88m. .. 8< 58 88 .88: 5.30
8.0 .. 8.8.8.0 .. 8.6000 .. 82 880 8.0 N88 8885 5.30
88.0 .. 8.88.0 .. 8.6000 .. 82 .880 2. a 8.0 3880 8.0
050.0320... .. 00>00.0>Ouo._. .. 000.0 .. :o_00>0_m_ 3.9 and .0>00 E00000 .50. 0.20500

 

 

 

 

0.20605 E00555

Lotw 209.000
885.80

.—

I‘

r

0050:0000..ch

oonEmm

00.0000

088%

 

 

61

and height; and Y. chungii cover % (elevation only), stem density (slope only), and culm
diameter (elevation only)(Table 3.3). We found bamboo percent cover, forest basal area,
understory cover, shrub cover, and overstory cover to be the most important forest
characteristics in determining bamboo culm characteristics (in order of decreasing
frequency)(Table 3.3). We also found size of forest patch to be significant in determining
B. fabri culm diameter and F. robusta percent cover, stem density, culm diameter and

height (Table 3.3).

Discussion

We explored the relationship between bamboo characteristics and timber
harvesting and fuelwood collection over time. Culm characteristics were not different for
different forest patch and harvest sizes. This indicates that whether harvesting occurs at
either the small (< 10 ha) or large (IO-100 ha) spatial scale (i.e., fuelwood collection or
timber harvesting, respectively), bamboo growth characteristics may not be adversely
affected in a particular way. Changes did occur in bamboo culm characteristics over time
as various bamboo species responded to the changing understory environment after
harvesting.

Our data supported previous findings of widespread, dense stands of Bashania
fabri and less dense stands of thick-culmed Fargesia robusta (Schaller et al. 1985, Taylor
and Qin 1987, Reid and Hu 1991, Reid et al. 1991b, Taylor and Qin 1993a, b) throughout
the panda range in Wolong Nature Reserve. We found Bashania stands in older forest.
occurring farther from the Wolong roadway (2342.4 1 40.6 m) and at higher elevations

(2478-3214 MSL) than younger Fargesia stands, which were lower in elevation and

62

closer to the road (1771-2726 MSL; 1324.7 i 52.4 m, respectively). Therefore, although
both species may be impacted by fiiture human activity, including possible harvesting
activities, the proximity of Fargesia to the roadway and therefore human settlements
make it more likely to experience impacts.

Interpretation of the results for B. fabri remain conservative because we were
unable to sample B. fabri stands growing in forests harvested “-30 years ago due to their
limited availability in the field. Because of this, we could neither validate nor refiite the
findings of others (Taylor and Qin 1988a, Reid et al. 1991b, Taylor and Qin 1993b) who
determined that clearcuts favored development of short, dense B. fabri stands in turn
reducing panda habitat quality. Despite this lack of data, we did observe an increase in
overall abundance as well as a decrease in culm width over time since harvest. B. fabri
was rarely found in forests younger than 30 years old, but in these forests it generally had
lower percent cover and stem density, as well as thicker stems. Our data support
previous research which identified elevation, slope, forest basal area and bamboo percent
cover to be among the factors that contribute to a healthy and productive B. fabri stand.

In F. robusta stands, percent cover and stem density also generally increased over
time, with the highest percent cover occurring in unharvested, old growth forests.
Fluctuations in percent cover, stem density, and culm width and height indicated an
interaction as the forest reinitiates and grows to a mature stand. Increased percent cover
in l l-30 year old forests demonstrates F. robusta attempting to colonize forest floor
while not substantially increasing its culm width or height. When the forest stand
matures to 31-100 years, percent cover decreases while stem density increases. This is

most likely because the F. robusta stand has established itself spatially and culms at the

63

perimeter died back or were outcompeted and shaded out by surrounding deciduous trees.
thereby allowing an increasing density primarily from concentrated root sprouting.

Our study documents the characteristics of another bamboo species, Yushania
chungii. Few studies are available in English or Chinese that describe the characteristics
of this species, even though it occurs throughout the panda’s range and offers the
potential as an alternative food source. Y. chungii was not as common as B. fabri or F.
robusta, and was typically interspersed among stands of B. fabri. It occurred in smaller
patches and was found in both very young forests (which were uninhabitable by other
bamboo species) as well as old-growth stands, demonstrating that it may act as an
opportunistic pioneer bamboo species in a diversity of habitats. Little is known about the
life history of this bamboo species, though others have found that culm densities decrease
as crown canopy cover of fir species increased (Gratzer et al. 1999). Our data suggest
that growth forms of Y. chungii are similar in different aged forests, whether in a newly
harvested forest or one with old-growth characteristics. This has significance for panda
habitat because Y. chungii therefore has the opportunity to play a key role when other
species of bamboo mass-flower and die off.

As part of the effort to protect bamboo populations and giant panda habitats, the
local government of Wolong has begun a program of bamboo planting. The primary
purpose of the planting is to provide food for the growing number of giant pandas in the
Hetaoping Giant Panda Breeding Center, located in the Reserve. Planted bamboos can
minimize the impacts of harvesting large sections of naturally occurring bamboo that
occur both inside and outside the Reserve. Also, planted bamboo stems offer an

alternative to the bamboo collecting by local farmers who use the bamboo for numerous

64

household purposes (brooms, bean poles, roof thatching, etc.). Although the amount of
bamboo harvested by locals is undocumented, there is the potential that its impacts on
giant panda habitats may be significant, and mitigation of human effects through planting
bamboo may be important. We urge further study of these planted bamboo stands to
determine how they may be affecting panda habitats.

Understanding the relationship between logging and bamboo dynamics is key to
understanding the sustainability of giant panda habitats. Large patches of highly suitable
giant panda habitats have been affected by harvesting activities for centuries (Liu et al.
2001). Yet new policies may improve conditions. The expansion of China’s reserve
system may increase the amount of areas protected and the Natural Forest Conservation
Program can increase the quality of protection of habitats. However, the creation of
nature reserves and logging bans do not necessarily mean that habitats will be protected
(Durst et al. 2001, Liu et al. 20011 , Zhao and Shao 2002, China-Daily 2004). The
continued risk of harvesting impacts suggests understanding effects of logging remains
critical. Our study indicates that logging activities influence understory bamboo
characteristics, and while some bamboo species experience perturbations in growth
characteristics alter a harvesting event, other bamboo species may not. This has
significant implications for all understory wildlife habitats in Asia and other parts of the

world that are now restricting or expanding harvesting activities.

65

CHAPTER 4

USE OF BAMBOO PATCHES BY THE GIANT PANDA AFTER FOREST

HARVESTING

66

CHAPTER 4
USE OF BAMBOO PATCHES BY THE GIANT PANDA
AFTER FOREST HARVESTING

introduction

Timber harvesting and fuelwood collection can have various effects on understory
vegetation (Duffy and Meier 1992, Roberts and Gilliam 1995a, b, De Grandpre and
Bergeron 1997, F redericksen et al. 1999, De Grandpre et al. 2000, Fredericksen et al.
2000, Scherer et al. 2000, Roberts 2002, Roberts and Zhu 2002, Frey et al. 2003) and the
wildlife species that rely on it (Carey and Johnson 1995, Rawat I997, Fredericksen et al.
2000). Species that prefer early successional habitats with high understory coverage
often benefit from harvesting (Litvaitis and Villafuerte 1996a, Heydon and Reynolds
2000, Smith and Litvaitis 2000, Thompson and DeGraaf 2001), while species that prefer
mature or interior forests characteristically do not gain any advantage from harvesting
(Zheng et al. 1995, Green and Catterall 1998, Potvin et al. 1999, Wormington et al.
2002)

Bamboo, a prolific understory species found throughout much of the world, is
affected by harvesting activities in various ways. Previous research indicated clearcutting
affects bamboo by increasing stem density (Reid et al. 1991b), decreasing seedling
densities (Taylor and Qin 1993b) and lowering germination, establishment, and overall
regeneration from seed (Taylor and Qin 1988a). In large clearcuts, the density of arrow
bamboo increases to such a degree that tree establishment is severely hindered (Schaller

et al. 1985, Taylor and Qin 1987. Taylor and Qin 1988a, Reid et al. 1991b, Taylor and

67

Qin 1993b). Regenerating forest is therefore slow to reestablish in clearcut areas. If
harvesting occurs in smaller areas, such as in selective logging or fuelwood collection,
enough deciduous overstory cover may be retained which can provide suitable conditions
for bamboo species that prefer late successional forest (Reid et al. 1991b, Taylor and Qin
1997. Saenz and Guariguata 2001).

Effects of harvesting on bamboo are significant because bamboo is the primary
habitat and/or food source for a number of wildlife species, including the endangered
giant panda (Ailuropoda melanoleuca) (Schaller et al. 1985, Reid et al. 1989).

Harvesting may also affect the spatial distribution of bamboo, which can have long term
effects on the distribution of panda habitats and connectivity between patches of habitat
(Linderman et al. 2005b). A recent census, which began in 1999 and reported results in
June 2004, estimated giant panda numbers have increased from 1,100 animals in 1988 to
over 1,590 animals currently (China State Forestry Administration 2004, World Wildlife
Fund 2004). However, these trends are questionable due to differences in census
methods (World Wildlife Fund 2004). Regardless of trends in panda populations, the
habitats upon which the giant panda relies continued to decrease in quantity and quality,
both in protected and unprotected areas (Liu et al. 2001).

Panda habitats are continuously progressing through competition and successional
stages after disturbance events such as harvesting. Logging effects on bamboo have
frequently been used as a proxy to describe potential impacts on panda habitat. However.
the effects of timber harvesting and fuelwood collection on panda habitats have not been
directly studied, and there is a lack of a detailed assessment of the effects of harvesting

on giant pandas. In addition. there is currently no description of panda use of bamboo

68

patches after the recovery of arrow bamboo from the last flowering event in 1983.
Therefore, the goals of this study were to: 1) determine how giant pandas are using
present day bamboo patches in both harvested and unharvested forest, 2) characterize
changes in panda habitat selection after timber harvesting and fuelwood collection, and 3)
compare selection of habitat features immediately afier bamboo flowering as documented

by Reid and Hu (1991) to selection of features today.

Methods
Field Methods:

We collected data during May-August 2001, May-November 2002, and June-
August 2003. A sample plot consisted of a 30 x 30 m area. Geographic and forest
characteristics as well as bamboo cover were measured over the entire 30 x 30 m plot
area. including elevation, slope. aspect, forest basal area (IO-factor prism), overstory
height and percent cover, midstory height and percent cover, shrub height and percent
cover, bamboo percent cover, and understory height and percent cover.

Use by the giant panda was determined by the presence of fecal droppings, which
was considered a useful indicator of panda use because although actual sightings of
animals are rare, feces are deposited frequently (97 droppings/day, 4 droppings/hour)
(Schaller et al. 1985) and remain for several months (personal observation). Despite the
frequency of defecation, the extremely low density of pandas, the extent of the forest and
ability of the animal to travel large distances make detection of feces infrequent. All

detected feces, regardless of time since defecation, were surveyed within a plot.

69

The center position of each 30 x 30 in plot was geo-referenced with a Pathfinder®
Pro XRS GPS unit (Trimble Navigation Limited) to allow for subsequent analysis and
modeling of the data. Due to logistical and time constraints which prohibit revisitation
and reexamination of sample plots for passed droppings, we assumed an equal probability
of feces detection in all habitats. Reid and Hu (1991) support this assumption as they
found a low ratio of missed droppings (1.3%) in Wolong Nature Reserve.

We surveyed giant panda use in three bamboo species in various aged forests in
Wolong. The three naturally occurring bamboo species included arrow bamboo
(Bashania/abri), umbrella bamboo (F argesia robusta), and yushania bamboo (Yushania
chungii). In addition to sampling areas with bamboo populations, we studied areas with
no bamboo for comparison. We examined bamboo characteristics in previously
harvested forests of various ages and in uncut forests. Local guides helped us to estimate
harvesting time in specific sites because there is little documentation of harvesting in
Wolong. Sites were grouped into four harvest age categories based on general forest
ecology. our ability to determine forest age in the field, and sample size issues: newly
harvested forest = 0-10 years, middle aged forest = 1 1-30 years, old harvest = 31-100
years, and old-growth forest.

We examined both patch size (included unharvested forest patches) and harvest
size (did not include unharvested areas) to determine if a relationship existed between
patch size and bamboo characteristics. Given topographic relief and time/access
constraints, forest and harvest patch sizes were estimated using a length and width
measurement of distinct forest boundaries and size of the patch was calculated based on

the resulting rectangle. Sites were then grouped into two size categories. Both patch and

70

harvest sizes were categorized as follows: Size Class 1 = < 10 ha, Size Class 2 = 10-100

ha.

Adaptive Cluster Analysis:

We used adaptive cluster sampling to assist in detecting differences in panda use
(feces presence) between the various harvested and unharvested areas. This method is
valuable because it is designed to sample rare events (e.g., panda feces) for species that
exhibit a degree of territoriality and allows for focused research in areas where the rare
events occur while maintaining proper statistical hypothesis testing protocol (Thompson
199 1 b, Thompson et al. 1992). Whenever we found feces (i.e., presence=1) in a 30 x 30
in sample plot, we added all units in its neighborhood (those 30 x 30m plots to the North,
South, East and West) to the sample. If in turn any of these subsequently added plots had
feces, the plots of its neighborhood were also added to the sample, so that finally the
sampling neighborhood contained 1, 5 (1 primary + 4 secondary), or 8 (1 primary + 7
secondary) sample plots. The sizes of these neighborhoods were ideal because of time
constraints during field sampling. In addition, bias in the final estimate of panda feces
presence/absence is introduced into the sample when these stopping rules are
implemented in the adaptive cluster (Su and Quinn 2003), but sampling neighborhoods of
5 and 8 plots are ideal when using the Horvitz—Thompson estimator (Su and Quinn 2003),
which is not very sensitive to the stopping rule and bias is reduced with low numbers of
sample plots (Su and Quinn 2003).

Adaptive cluster sampling allowed us to increase our sampling efforts around

areas that had a higher potential of giant panda activity. In addition to adding to the

71

number of sample units, cluster sampling increased the proportion ofsites with feces
presence and therefore provided a more insightful analysis of forest characteristics
influencing panda use. In other words, by focusing our efforts on areas with a higher
probability of feces presence, we were able to understand more about the use of forested
habitats by the rare giant panda. The spatial autocorrelation bias associated with
neighboring plots was factored into all analyses via use of the Horvitz-Thompson
estimator, Euclidean weights and autocovariate logistic regression.

The original Horvitz-Thompson estimator for determining a value for a

population is

YHz—I’Zyi

Ni=17ri

where 71,. is the probability that the ith unit is included in the sample and y,~ is the
measurement for the ith unit in a sample of n specified units (Thompson and Seber 1996,
Cochran 2002, Turk and Borkowski 2005). In this study, we used a modification similar
to that described by Lo et a1. (1997). The mean proportion of sites with giant panda

feces, )7, , in the ith stratum (i.e., bamboo type), was estimated by

.1k_,-a

 

 

 

yr Aijzla

where A,- is the total area for the ith stratum, a". is the area of the network (number

ofplots in neighborhood with feces [including the primary plot] x 30 m2), and k is the
total number of neighborhoods sampled in the ith stratum (Lo et al. 1997). The

probability of finding feces in each neighborhood, )7” in thejth observed plot,

72

’"1

71:2”!

u=l mj

 

is the sum of plots with feces present in the neighborhood, 7,." , divided by the
total number of plots in the neighborhood, m j. Here. we defined a]. , the intersection

probability (the probability that two neighborhoods, k and h, will intersect by the initial

sample), similar to that used by Dryver and Thompson (2005), where

IIN‘N*1+I”‘”"‘-IN‘”*””11
a.(n.>0&n >0)=1— " n l n

I ‘ " N)
["2

Because Horvitz-Thompson estimator values were mean proportions of panda

 

 

use, we calculated differences among groups using prop.test in S-plus.

Statistical Analysis:

Initially, we tested whether giant panda feces were found in greater numbers in
different bamboo stands using 12 analysis. We then used 12 analysis on each bamboo

species to test whether harvested forests had fewer feces than unharvested, old growth

forests. To test whether the size of the forest (or harvest) patch influenced the proportion

of feces found, we again used 12 analysis.

73

A utalogistic Regression:

We performed both stepwise logistic and stepwise autologistic regressions to
determine which forest and geographic variables most suitably predicted the presence of
giant panda feces in each of the three bamboo species stands. Initially, we performed
univariate logistic regressions on all independent variables and retained those with P-
values for likelihood ratios (G-test) less than 0.25 (SAS Institute 1985). We then
performed correlation analysis on remaining variables to detect collinearity (SAS
Institute 1985). Variable pairs found to be covarying (r > 0.60) were examined, and we
eliminated the variable with the least significant univariate G-test score. We then
considered all two-way interactions among remaining variables and retained those
interaction terms with P-values for likelihood ratios (G-test) less than 0.25 (Klute et al.
2002).

We used the spdep extension for the R statistical package to determine the degree
of spatial autocorrelation using join-count statistics for the binary response variable
(feces presence/absence). We generated one thousand Monte Carlo permutations at
distances: 0-50, 50-100, 100-250, 250-500, 500-750, 750-1000, 1750-2000 to
represent possible levels at which the giant panda may be responding to its environment.

We adapted the autologistic regression procedure described by Klute et al. (2002).
The logit of probability of panda feces presence (p;) was used with the addition of an

autocovariate term to measure the spatial relationship:
1081“,“): fl() + fl1X1i+ + flani + Igmauto COVi

where:

74

p,- is the probability of feces presence at the ith stop, ,30 is the model intercept,

,B,,..., ,8" are parameter estimates, X I ,....,X n, are the values of the predictor variables at the

ith stop.

In
2141ny
1

auto cov,~ = J:

k.
2 wii
i=1

1
wt] : —
hli

represents the weight given to stop j as the inverse of the Euclidean distance (h,_-,-)

between stops i and j and y]. is 0 when feces is absent, 1 when feces is present (Klute et al.

2002).

We then used PROC LOGISTIC (SAS Institute 1985) to test all autologistic
regression models for each of the three bamboo species. For each logistic regression,
stepwise regression techniques were used to select the initial variables. Results from both
the logistic and autologistic methods were described to show differences from adding in
the autocovariation term. We performed each method on both the primary plots only and

the neighborhood (primary + secondary plots) to demonstrate differences in results.

E lectivity Index:
The last flowering of B. fabri occurred in 1983. Shortly alter the subsequent

bamboo dieoff, Reid and Hu (1991) studied the forest characteristics giant pandas were

75

selecting and presented the data in terms of Vanderploeg and Scavia‘s Electivity Index
(Lechowicz 1982, Reid and Hu 1991). In an effort to better understand what
characteristics pandas may be selecting in bamboo areas now after the recovery from the
previous flowering event, we also calculated the Electivity Index (see below) (Reid and
Hu 1991). Comparisons between giant panda preference shortly after B. fabri flowering
(Reid and Hu 1991) and during the period of our field study (2001-2003) are discussed
through the Electivity Index results.

We calculated the Electivity Index to offer a comparison of association with
habitat features now that the bamboo populations have recovered from the previous die
off. Selection of habitat characteristics was calculated using Vanderploeg and Scavia‘s

Electivity Index (E,):

(W,- —(-‘—))
E, = ———’—l’——
(W.- +(-))
n
where
W _ (n/P.)

lCWZZri/P

P,- is the proportion of sample units in the ith class, and r,- is the number of sample
units with feces present in the ith class. The index is scaled between —1 and +1 (<0 =

avoidance, O = random, >0 = preference)(Lechowicz 1982, Reid and Hu I991).

76

B.e_5.u_lt§

We sampled a total of 913 plots in 443 neighborhoods (443 primary and 470
secondary sites). Of the total sample plots, B. fabri and F. robusta were encountered
more frequently (38.6 % and 29.9%, respectively) than areas with no bamboo (16.2%)
and Y. chungii (9.1%). Species of bamboo had a significant effect on the presence of
giant panda feces ( 1505‘, = 136.05, p<0.001). Areas with Y. chungii bamboo had the
highest proportion with feces (52.63%), significantly higher than F. robusta stands
(19.01%) and areas with no bamboo (0.71%) (Figure 4.1). Bashaniafabri (32.17%) also
had a significantly higher proportion with feces than areas with no bamboo (Figure 4.1).

For different bamboo species (B. fabri, F. robusta, and Y. chungii), different

forest ages had significantly different proportions of feces. Time since harvest did not

influence the presence of feces in B. fabri ( 130,, = 87.14, p<0.001)(Figure 4.1 ). F.

robusta ( 15053 = 38.89, p<0.001)( Figure 4.1), and Yushania spp. stands ( 1:0,, = 174.33,

p<0.001)( Figure 4.1)(Appendix 2A-C). We did not test areas without bamboo because

of their extremely low amounts of giant panda activity.

77

 

 

 

 

 

 

 

 

 

 

 

 

 

100 e
- 0-10 years
__ 1:: 11-30 years

5 80 J 31-100 years
3%: m 1:1 Old-growth
.9 § - Total
0
— L1.
0. 60 -

m
l.—
o 'U

c
$58
88 4° ‘
t5 (9
0.

20 -
0 . .. T _
. fab“ (1513 ung“ boo
Bashania Fargesia fob V(“wattle 0“ N0 bar“

Bamboo species

Figure 4.1. Percentage of survey plots in areas with three species of bamboo and areas

with no bamboo that have giant panda feces present in various aged forests.

78

Bashaniafabri plots had little giant panda use in new and mid-aged harvests, but

use increased significantly as the forest aged ( 1505, = 87.14, p<0.001)( Figure 4.1). Both

old harvests and unharvested forests with B. fabri had significantly higher proportions of
feces than newly harvested and middle aged areas (Figure 4.1)(Appendix 2A). Newly
harvested forests with F. robusta had significantly less panda feces present than mid-
aged, old harvest, and unharvested forests (Figure 4.1)(Appendix 2B).

Newly harvested areas with Y. chungii bamboo had relatively higher proportions
of giant panda feces but were probably not as important to pandas than stands in
unharvested forests (Figure 4.1) (Appendix 2C). Few recently harvested areas and old
harvest sites contained Y. chungii bamboo, so sample sizes of these areas were low,
limiting our ability to interpret results for this age class. Proportion chi-square results

indicate significant differences in feces presence among Y. chungii forest age classes

( 150,3 = 174.33. p<0.001), however, results should be interpreted cautiously as the

bamboo does not occur as frequently as other species and sample sizes were therefore
low.

There was no difference in the probabilities of encountering feces between small
(< 10 ha) and large (IO-100 ha) forest patches for B. fabri or F. robusta. No Y. chungii
stems were found in large forest patches (IO-100 ha) and were therefore not tested.
Similarly, within small and large harvests, there was no relationship with the proportion
of giant panda feces found between B. fabri and F. robusta.

All forest and geographic characteristics were initially tested using univariate

logistic regression to determine their inclusion in the final logistic regression models.

79

Univariate models using both the primary plot only dataset and the neighborhood
(primary and secondary plots) dataset showed similar results (Table 4.1). Both logistic
and autologistic stepwise models suitably characterized (p < 0.05) the presence of giant
panda feces using univariate and multivariate combinations of forest and geographic
characteristics for all bamboo species except for the autologistic primary plot model of
Yushania chungii (Table 4.2). The autologistic models were considered more reliable
because they accounted for spatial autocorrelation of the dataset, which was an issue due
to our use of the adaptive sampling methodology. Because the autologistic model
accounts for autocorrelation, use of all neighborhood plots, including primary and
secondary plots, provides the most useful characterization of variables influencing
presence of giant panda feces. Because the logistic model that uses all neighborhood
plots does not account for spatial autocorrelation, we believe its results are less reliable.
For B. fabri, the logistic model selected the same characteristics (slope, ecological
aspect) to describe the presence of panda feces as the autologistic using all neighborhood
plots (Table 4.2). Overstory height, which was correlated with total and conifer overstory
cover, was also found to be a significant predictor. In stands of Fargesia robusta,
overstory height and deciduous midstory cover best predicted giant panda feces presence,
along with conifer overstory cover (Table 4.2). Elevation and understory height were the

most useful characteristics of panda habitat in Yushania chungii stands (Table 4.2).

80

Table 4.1. Univariate logistic regression models of forest and geographic characteristics

on primary plots only and Horvitz-Thompson estimators of all neighborhood plots

(primary and secondary). The variable ecological aspect is based on the

recharacterization of aspect into 21 ecologically significant units according to Parker

(1982).

Elevation

Distance to Roadway
Distance to House
Forest Age

Patch Size

Slope

Ecological Aspect

Basal Area

Bamboo Cover

Total Overstory Cover
Conifer Overstory Cover
Deciduous Overstory Cover
Average Overstory Height
Total Midstory Cover
Conifer Midstory Cover
Deciduous Midstory Cover
Average Midstory Height
Total Shrub Cover
Average Shrub Height
Total Understory Cover
Average Understory Height

81

P-value
primary plots only primary and secondary

<0.001
<0.001
<0.001
<0.001
0.84
0.39
0.95

<0.001
<0.001
<0.001
0.01
0.04
<0.001
<0.001
0.180
<0.001
<0.001
0.090
<0.001
<0.001
0.00

P-value

<0.001
<0.001
<0.001
<0.001
0.54
0.04
0.51

<0.001
<0.001
<0.001

0.14
<0.001
<0.001
<0.001

0.54
<0.001
<0.001

0.00
<0.001
<0.001

0.30

Table 4.2. Multivariate logistic and autologistic models for primary plots only (P) and

Horvitz-Thompson estimators of neighborhoods (P+S) for three bamboo species,

Bashaniafabri, Fargesia robusta, and Y ushania chungii. *p<0.1, **p<0.05, ***p<0.01.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Bamboo Species Regression Samples AIC Intercept Variable Estimate
Logistic P 99.53 -3.713“ Overstory Height 0122'“
P+S 321.37 1.16" Understory Height 0012‘“
Ecological Aspect -0.059"
Slope 0046'“
Bashania fabri Autologistic P 61.03 -2.377*“ Euclidean Distance = 250 155.1'“
P+S 282.65 -O.792 Euclidean Distance = 1000 137.0‘"
Overstory Height 0049*"
Ecological Aspect -0.103"“
Slope -0.057“*
Logistic P 98.3 -2.003*“ Overstory Height 0.073m
Understory Height -0.028“
P+S 268.43 -3.749“* Overstory Height 0082"“
Deciduous Midstory Cover 0026""
Bamboo Cover 0015'“
Fargesia robusta
Autologistic P 64.84 -3.717“* Euclidean Distance = 500 214.4'“
Conifer Overstory Cover 0052“"
P+S 233.467 4942‘“ Euclidean Distance = 1000 125.6‘“
Overstory Height 0129*"
Deciduous Midstory Cover 0.02"
Logistic F 18.566 8.689" TBamboo Cover 0.081 "
P+S 72.899 -18.448**' Elevation 0.007'“
Understory Height -0.041"‘
Autologistic P 10.071 -171 Euclidean Distance = 500 1673.2
Yushania chungii Elevation 0.072
Bamboo Cover 0.579
Distance to Road -0.047
P+S 72.89 -18.448*“ Euclidean Distance = 500 -58.407"“
Elevation 0.008'“
Understory Height -0.041*""

 

82

 

According to Electivity indices, giant pandas exhibited significant preference for

and avoidance of certain habitat features (Table 4.3). Both B. fabri and F. robusta

bamboo stands that occurred on steep slopes were avoided ( 12 =8.34, p<0.01 and

13 =1 .93, p<0.18, respectively). Similarly, in Y. chungii patches, preference was shown
for flat areas (<15 degrees)( 12 =2.12. p<0.18). Giant pandas also preferred very tall F.
robusta stems (> 5 m)( 2’2 =1.77, p<0.18) and avoided short (2 — 2.9 m)( 12 =2.36,
p<0.18) and wide (> 13 mm)( 2’2 =1 .80, p<0.18) stems. Wider Y. chungii stems (> 6 mm)

also appeared to be avoided (6-7.5 mm; 12 =2.73, p<0.10), with pandas seeming to

prefer narrower Y ushania stems instead. Electivity values indicate bamboo areas that had
low stem density were avoided for the three bamboo species while areas with higher

densities were preferred. Pandas were shown to be avoiding very sparse stands of B.

fabri and F. robusta ( 12 =3.21, p<0.10 and 12=3.68, p<0.10, respectively) (Table 4.3).

Discussion
Bamboo is not a stagnant resource, but a dynamic organism that exhibits dramatic
fluctuations in growth form and densities during sexual reproduction events and afier
logging activities. In order to best conserve the giant panda, a better understanding of
bamboo characteristics, and how pandas perceive bamboo over time, is therefore
necessary. We describe use of bamboo stands for three wild bamboo species over time
after timber harvesting and fuelwood collection as well as areas without bamboo. This

study was performed at a time when B. fabri, the most extensive bamboo species

83

Table 4.3. Electivity Index values for slope, culm height, basal diameter and culm
density for Bashaniafabri, F argesia robusta, and Y usham'a chungii. First column for
each species represents categories and sample size in parentheses. Second column
indicates Electivity Index values. Electivity Index values are scaled between —I and +1
(<0 = avoidance, 0 = random, >0 = preference)(Lechowicz 1982, Reid and Hu I991).

p<0.18*, p<0.10**, p<0.01***

 

 

 

 

 

 

 

 

 

B. fabri F. robusta Y. chungii
Slope <15 (111) 0.153 <15 (37) -0.089 <15 (22) 0.123 "
(degrees) 15-24 (100) 0.140 15—24 (72) 0.096 15-24 (35) -0.035
25-34 (64) 0.109 25-34 (103) 0.081 25—34 (19) -0.032
>35 (39) -O.873 **" >35 (59) -0.127 * >35 (7) -0.082
Culm Height 007 (29) -0.264 0-1.9 (5) -0.027 0-1.6 (4) 0.200
(meters) 0.8-1.0 (22) 0.169 202.9 (29) -0.210 * 1.7-2.2 (12) 0.200
1.1-1.2 (22) 0.169 3.0-3.9 (55) 0.016 2.3-2.6 (3) 0.333
1.3-1.4 (7) 0.233 404.9 (18) -0.118 2.7-3.2 (1) N/A
1.4 + (2) N/A 5.0 + (6) 0.224 " 3.2 + (0) N/A
Basal Diameter 0-3.9 (39) -0.049 0-9.9 (24) -0.040 0-5.9 (5) 0.282
(mm) 4.0-4.4 (23) 0.094 10.0-12.9 (58) 0.144 6.0-7.5 (8) -0.382 **
4.4 + (20) 0061 130+ (30) -0.150 * 7.5 + (7) -0.133
Culm Density 0-29.9 (12) -0.268 ** 0-6.9 (19) -0.387 ** 0-9.9 (2) N/A
(stems/m2) 30.0-59.9 (31) 0146 7.0-14.9 (42) -0.250 10.0-19.9 (5) -0.346
60.0-89.9 (27) 0.053 15.0-22.9 (34) -0.006 20.0-29.9 (3) -0.105
900-1199 (9) 0.013 23.0-29.9 (13) 0.127 30.0-39.9 (3) 0.237
120.0 + (3) 0.212 30.0 + (5) 0.253 40.0 + (7) 0.351

 

 

84

throughout Wolong, returned to normal pre-flowering densities after the 1983 sexual
reproduction event (based on the findings of (Taylor et al. 1991)) who estimated
approximately 15 years after 1983 for recovery).

Our research corresponds with other results that suggest giant pandas prefer
higher elevation habitats (> :2400 MSL). In our study, feces were found more frequently
in higher elevation Y. chungii and B. fabri stands, and less frequently in lower elevation
F. robusta. These results were expected, as others have noted that pandas spend the
majority of their time in areas of higher elevations and come down to lower elevations
only sporadically in early spring to eat new shoots and in the late summer to feed on
leaves and mature shoots (Schaller et al. 1985). The impacts of timber harvesting and
fuelwood collection are relevant to all native bamboo species because. logging occurred at
all elevations (1771 - 2984 MSL), though forests in lower elevations are more often
logged due to the proximity to households and the roadway that bisects the Wolong
Nature Reserve.

Giant pandas did not use recently logged (0-30 years) forests that contained B.
fabri and F. robusta. However, forests did not need to be in an old growth stage to be
used. Pandas used areas that had been harvested 31-100 years prior as frequently as old
growth forest, indicating the ability of these forests to regenerate into suitable habitat.
More specifically, our data indicate that forests approximately 60 years in age were
frequently used. This amount of time may therefore be necessary for forests to
regenerate to a stage where bamboo grows at a level where giant pandas can forage.

Forests that contained Y. chungii showed high amounts of use in all age classes

except 1 1-30 years, including 100% use in newly harvested areas (0-10 years). Fewer

85

sample plots were obtained in Y. chungii stands and may be a possible explanation for the
high percentage of use. Y. chungii is less prevalent than B. fabri and F. robusta and
therefore had a lower frequency of occurrence. However, the amount of use in these
areas indicates it may be a valuable food source, and its occurrence at higher elevations
makes it a prime alternative food source during B. fabri die offs.

Within B. fabri stands, the ability to predict the presence of feces was best
described with slope and aspect. Pandas preferred areas with low slope and forest with a
southwestern aspect. Other habitat features, such as conifers in a tall overstory, were also
helpful at predicting giant panda feces presence in B. fabri stands. Forests with F.
robusta were used by pandas when the bamboo stands were associated with a forest that
had a tall conifer overstory and a dense deciduous midstory. We believe these forest
variables may be more indicative of habitats required for the cool, moist conditions
preferred by the F. robusta bamboo, but the combination of these factors is ideal F.
robusta panda habitat. Pandas preferred stands of Y. chungii that had low ecological
aspects (SW facing slopes), higher elevations and low understories.

Electivity values for all three bamboo species indicate pandas do prefer some
areas to others. Reid and Hu (1991) also calculated electivity values for B. fabri shortly
afier the 1983 flowering episode. Comparing our electivity values to theirs we find
several similarities and differences in panda habitat preferences shortly after B. fabri
flowering and almost two decades after flowering. Currently, giant pandas are generally
less selective (i.e., they act more like generalists) in their choice of habitats than they
were shortly after the flowering event. For example, after the flowering event (Summer

1986), pandas avoided areas with sparse bamboo growth (culm density <59 stems/m2)

86

 

and were forced to select areas with higher bamboo density (Reid and Hu 1991). We
found that in 2001-2003, pandas avoided areas that had the lowest bamboo stem density,
but otherwise used all other areas with no preference. Still, steep areas and areas with
low culm density are avoided during both time periods (Reid and Hu 1991).

This study indicates that it may take at least 30 years for B. fabri and F. robusta to
regenerate to levels where panda use was similar to what it was before harvesting. We do
not recommend logging as a prescription within Wolong, but our data do suggest that
habitat can recover from such events after 30-100 years (mean age B. fabri = 53.41 years.
mean age F. robusta = 50.31 years, Appendix 2A & 28)

A bamboo species. Yushania chungii, is used by pandas even immediately after a
harvesting event. This species may therefore be an overlooked resource for giant pandas,
especially during B. fabri dieoffs, but its smaller areal extant limits its overall importance.
Areas with no bamboo were often in grazing areas, old fields or shrublands. Grazing by
cow, goats and horses continue to require large amounts of land be kept in grassland.
Although we did find one solitary panda feces in the middle of a cattle-grazing pasture.
our data suggest pandas do not use these areas even though they are in close proximity to
highly suitable habitats. We therefore suggest the Wolong Reserve Administration begin
to address the loss of habitat to grazing lands and the potential for these areas to return to
forestland and panda habitat.

In addition to the three species of naturally occurring bamboo, a new program has
been developed by the local Wolong government to plant bamboo. Planting bamboo
began in 2001 as part of the Grain to Green policy in which households are subsidized

with grain to plant bamboo. Planted bamboo was located primarily among crops of

87

cabbage, but also among potato and cornrows. In the areas of lower elevation (Gengda
Township), planted bamboo stems were growing well, while at higher elevation (Wolong
Township) the planted stems were not as successful. We found no evidence of use of
planted bamboo by wild giant pandas. However, we did observe a giant panda using corn
crops in Crow Valley (LaoYaShan; Gengda Township), indicating pandas may be willing
to descend to lower elevations for a human-grown food source. Tempting wild pandas
down from their higher elevation habitats should not be encouraged and may only lead to
further problems associated with pandas interacting with humans.

Giant pandas eat approximately 150 shoots/day (Schaller et al. 1985). With the
current number of captive pandas at the Hetaoping Breeding Center (N=40),
approximately 6,000 stems must be harvested daily to sustain the captive population.
Forty percent of these stems given to captive pandas are F. robusta culms harvested by
local people within the Reserve while the remaining 60% is bought from outside the
Reserve (Siqiang Zhou, pers. comm.). Our data indicate an average of 15.48 stems of F.
robusta per square meter. Therefore, if we assume that 40% of captive food comes from
Wolong, over 150 m2 /day of native F. robusta bamboo habitat must be harvested to
support the panda population in Hetaoping. This means that a large amount of bamboo is
cut inside the Wolong Reserve, while an even larger amount is also trucked in several
times per week. Transporting bamboo from surrounding counties increases the amount
of labor costs and financial pressures, as well as increasing noise and air pollution within
Wolong. Bamboo planting within Wolong may help reduce these costs while at the same

time offering grain subsidies to local people and local bamboo to the pandas.

88

The results of this study suggest that harvesting for timber and fuelwood
influences understory bamboo growth characteristics and this, in turn, affects the use of
habitats by giant panda. Immediately after the harvesting event, different bamboo species
respond in different ways, and some species may provide better habitat than others. Over
time, bamboo characteristics change as does the pandas selection of different species.
Findings from this research can be used toward understanding the relationship between

other species which rely on understory vegetation that is influenced by forest harvesting.

89

CHAPTER 5

IS REFORESTATION EQUIVALENT TO RESTORATION OF PANDA

HABITATS: A CASE STUDY IN WOLONG NATURE RESERVE

90

CHAPTER 5
IS REF ORESTATION EQUIVALENT TO RESTORATION OF PANDA
HABITATS: A CASE STUDY IN WOLONG NATURE RESERVE

Introduction

Plantation reforestation is becoming increasingly common, with over 180 million
ha already planted worldwide (F A0 2005). Many countries are using artificially
regenerated forests and plantations of certain trees as a means of reestablishing a forest
canopy (Kanowski et al. 2003), helping to retain topsoil or prevent desertification (Liu et
al. 2002, Stott and Mount 2004, Liu and Diamond 2005), and as a way to provide woody
biomass for fuelwood, paper, or crop production (He and Barr 2004, FAO 2005). Asian
countries, in particular, have been actively attempting to conserve their forests through
plantation reforestation. Asian reforestation projects have accounted for a large portion
of global efforts, with over I 15 million ha of plantations currently throughout Asia (FAO
2005). Within this region, several countries including Indonesia, Lao PDR, Philippines,
Thailand, and Vietnam have all launched programs for reforestation and together have
planned to reforest nearly 10 million ha over the next several years (Gilmour 1999. Sayer
et al. 2004).

The Chinese government has recently established one of the largest reforestation
programs in the world. With one of the highest (approximately 22% of the world’s
population) and densest (roughly 120 people/kmz) human populations (as opposed to 20
people/km2 in the United States or 60 people/km2 in Europe) (Peng and Guo 2000, Liu

and Diamond 2005). China has experienced extensive deforestation (Smil 1984, Liu et al.

91

1999b). To counter deforestation and its effects, the Chinese central government recently
adopted two policies, namely the Natural Forest Conservation Program (N FCP) and the
Returning Steep Agriculture Slopes to Forest Program (Grain to Green or GTG), which
will limit timber harvesting and reforest areas that were previously cleared. respectively
(Xu et al. 2004, Liu and Diamond 2005).

China has one of the largest planted forest areas in the world (Zhao and Zhou
2005). Current estimates place the total amount of plantation forest in China to be over
45 million ha (FAQ 2005). The Grain to Green program will increase the amount of
artificially regenerated forest even fithher, with estimates of over 7 million ha already
planted (Xu et al. 2004) and 14.67 million ha of croplands being converted to forest by
2010 (Xu et a1. 2004). Given the current funding for the GTG program (over US$40
billion), it will likely continue to convert a large number of croplands to artificially
regenerated forest (Xu et al. 2004).

Yet despite the high level of financial resources being committed to reforestation
projects in China and other countries around the world, there are some debates on the
potential benefits plantation reforestation will provide biodiversity. More specifically,
the possible resources that plantation forests provide to wildlife remain unclear (Stanturf
et al. 2001, Hartley 2002).

There is some evidence that reforested areas may be beneficial to wildlife because
they provide a certain habitat component that may not be available if an area is left
unforested and may do so in a shorter time period than if allowed to naturally regenerate.
For example, wildlife may profit from plantations of mast species because they provide

readily accessible forage (Twedt and Wilson 2002, Grossman et al. 2003). Wildlife have

92

also been found to use reforested areas (Latham et a1. 1996, Matlock et a1. 2002, Hemami
et a1. 2004), with some reforestation species (Twedt et al. 2002), certain forest ages
(Nagaike et a1. 2003), or particular seasons (Palmer and Truscott 2003) offering more
suitable habitat.

Species diversity may also increase after reforestation. Sykes et al. (1989)
recorded mammalian diversity and found a marked increase following reforestation of a
rangeland. Prior to reforestation activities, nine mammalian species were using the area.
Twelve years afterwards, this number increased to 18 species, 12 of which were breeding
regularly (Sykes et al. 1989). In addition to species diversity, reforestation areas may
influence population abundance. For example, Sykes et a1. (1989) found a general
increase in the number of roe deer harvested throughout their study area after
reforestation. Also, the number of breeding pairs increased from 271 before afforestation
to 724 afterwards (Sykes et al. 1989).

Although some species may benefit from reforestation, converting an area to
artificially regenerated forest may be detrimental to other species. Some wildlife species
may not be able to use areas that are reforested (Twedt and Portwood 1997, Ericsson et
al. 2000). Species diversity may also be negatively influenced. For example, Armstrong
et al. (1998) determined that reforested lands had a marked negative impact on the
biodiversity of some areas in South Africa. They found that several plant and animal
species had become locally extinct or threatened within newly reforested lands, and that
reforestation projects had been largely responsible for the endangered status of some
species (Armstrong et al. 1998). In another study, Allan et al. (1997) compared habitats

and found grasslands had higher bird species diversity and had a higher number of

93

endemic and threatened species than reforested areas, even if the plantation area was very
small.

Plantations also affect small and large mammal abundance. Taylor and Perrin
(1996) caught a high number of small mammals in new, young pine plantations
(approximately 0-14/trapnight). However, small mammal numbers declined as the timber
trees matured, so that there were very few remaining by the time the stand matured.
These numbers, even in the new plantation, were still lower than the small mammal
numbers in surrounding natural habitat, where numbers ranged from 18 to 25
mammals/trapnight (Maddock I988). Hulbert et al. (I996a) also noticed a decline in the
presence of two species of lagomorphs using plantations as the forest matured. However,
because of the species’ adaptability, the authors concluded that in the early years
following planting, reforestation is unlikely to influence mountain hare and rabbit
populations as long as an abundant ground cover remains available (Hulbert et al. 1996b).
Changes in large mammal abundance have also been documented. Extirpation of an
African antelope, the oribi, after reforestation was recorded by Viljoen (1982). The oribi
were not seen in forest plantations and were found to be avoiding the areas due to
preferences for grasslands (Viljoen I982).

The extent of reforestation occurring throughout Asia, especially in China, has the
potential to greatly affect the resident wildlife occurring in these regions. This is
especially true for the mammal populations, as China has one of the richest mammal
populations in the world (State Environmental Protection Administration of China 1998).
However, we are unaware of any study that has identified the relationship between

Chinese reforestation and the neighboring wildlife habitats. Therefore, we chose to study

94

habitats of the giant panda (Ailuropoda melanoleuca) in Wolong Nature Reserve,
Sichuan Province. The giant panda occurs throughout the steep, mountainous regions of
Sichuan Province, and its habitats may be influenced directly by GTG reforestation
projects (Liu and Diamond 2005).

To understand the relationship between reforestation areas and giant panda
habitat, we investigated whether current artificially regenerated forests in Wolong
provided habitat to pandas and how artificially regenerated forests may mitigate the
impacts of fuelwood harvesting on panda habitats. Our first objective was to determine
how pandas currently use reforestation areas in order to establish whether areas that have
been reforested can create panda habitat. To accomplish this, we compare panda use of
various aged artificially regenerated forest to natural surrounding forests. Our second
objective was to evaluate the potential of reforestation areas to mitigate impacts on panda
habitat by providing alternative fuelwood sources. For this, we developed a model of
reforestation mitigation potential based on field data and information from the literature

to address the basic issues related to panda habitat.

Methods

Potential ofReforestation Areas to Create Habitat:
Field Survey Methods:

We surveyed giant panda use in various aged reforestation areas as well as natural
forests throughout much of the Wolong Nature Reserve. Use by the giant panda was

determined by the presence of fecal droppings, which was considered a usefiJI indicator

95

of panda use because although actual sightings of animals are rare, feces are deposited
frequently (97 droppings/day, 4 droppings/hour) (Schaller et al. 1985) and remain for
several months (personal observation).

Data collection took place from May-August 2001, May-November 2002, and
June-August 2003. We sampled a total of 69 plots in randomly selected reforestation
areas and 69 plots in neighboring natural forests. A sample plot was 30 x 30 m in size.
The center of each plot was spatially referenced using a Pathfinder® Pro XRS GPS unit
(Trimble Navigation Limited). Geographic and forest characteristics as well as bamboo
cover were measured over the entire 30 x 30 m plot. We located sample plots while in
the field. Sites were grouped into four harvest age categories based on their forest stand
ecology, our ability to age the forest in the field, and sample size issues: 0-10 years, 1 1-
30 years, 31-100 years, and unharvested or old—growth forest.

Statistical Analysis:

 

To determine the potential of artificially regenerated forest to provide suitable
panda habitat, we first tested for differences in forest and geographic characteristics
between reforestation areas and neighboring natural forests. To test for differences
between natural and artificially regenerated forests, we used the paired t-test (when
variances tested equal) or the more robust Wilcoxon signed-rank test (when variances
tested unequal). Variances were tested using an F-test for variance equality (Insightful
Corp. 2003) and were considered equal if P > 0.2.

To gain a general understanding of where giant panda feces occurred in
artificially regenerated and natural forests, we tested for differences in mean canopy

covers (overstory, midstory, shrub, and bamboo strata) in areas with and without feces

96

using ANOVA and Tukey’s posthoc comparisons. We then estimated which forest and
geographic characteristics best predicted the presence of giant panda feces using logistic
regression. We fit two final logistic regression models, one for reforestation areas and
one for the surrounding forest. For each model, we initially fit univariate logistic
regressions to all variables and retained each variable only if the likelihood ratio test (G-
test)(PROC LOGISTIC; (SAS Institute 1985) had P < 0.2. Variables that were
significant in univariate tests were then tested for collinearity using PROC CORR (SAS
Institute 1985). For pairs of variables with r > 0.60, we eliminated the variable with the
less significant univariate G-test score. A multivariate, stepwise logistic regression was
fit using the remaining variables. Significant predictors in the logistic regression helped

to describe habitat characteristics preferred by pandas.

Conservation Potential of Reforestation Areas:

To establish the potential of reforestation areas to mitigate impacts on natural
giant panda habitats, we first determined the possible extant of fuelwood harvesting
impacts on natural giant panda habitats (i.e., natural forest). We simulated impacts by
subtracting firelwood volumes collected by households from natural forest stands. In
other words, we estimated forest stand biomass using data collected in the field and then
subtracted amounts of biomass used for fuelwood to determine a decrease in forest
biomass. Initial stand volumes of natural forest were based on DBH and tree height
measurements and tree densities gathered from 342 field plots throughout Wolong Nature

Reserve. Each field plot measured 30 x 30 m and numbers of trees in 20 cm categories

97

were counted (IO-30, 31-50...71-90, 91+ cm DBH). We calculated volume of tree
biomass using (Wenger 1984):

V=BAXZBIi

where, BA = mean basal area in each 20 cm category and TH = average overstory
tree height. Average household fuelwood use was estimated to be 10.6 m3/year based on
Liu et al. (199%), who calculated an annual fuelwood consumption of 10.000 m3/year
during a time when the total number of households equaled 942 (Liu et al. 1999b). Since
there is a current ban on fuelwood collection (from the NF CP), we also simulated
potential effects of reduced fuelwood collection at 50% (5.3 m3/year), 25% (2.65

m3/year), and 12.5% (1.33 m3/year) levels.

We then modeled potential impacts of fuelwood collection on panda use by
calculating a change in proportion of feces presence due to a change in forest biomass
from fuelwood harvesting. To do this, we created a model to establish the relationship
between harvesting and change in panda use. The model was a regression that used
measurements of stand volume in 342 forest plots in Wolong and the mean proportion
with feces. Using this regression, we then demonstrated the decrease in likelihood of

finding feces in a given natural forest stand after it had been harvested.

Finally, we estimated the volumetric growth of reforestation trees to determine
their potential for mitigation of fuelwood collection in natural forests. To do this, we
began a simulation based on generic 5-year-old saplings that had dimensions equivalent

to 0.025 m dbh (l-inch dbh) and 0.6 m height. We then adapted a volumetric growth

98

equation for Larix potaninii var. macrocarpa using data from Li (1990). The growth
equation was (r2=0.997):

— - 2 -9
y=6e4x-5e6x +9e x3

(5.1)
where x = age of tree in years and y = volume growth in m3. Because of the difficulty
associated with modeling the initial years of seedling growth, and because plantation
trees in Wolong were initially grown in nurseries, the growth model was most reliable
between year 10 and year 90. Therefore, we only report results between these tree ages.
We chose to base our model on Larix potaninii var. macrocarpa because it is planted
throughout Wolong and was found to have less volume growth than other species making
it a more conservative estimate in modeling mitigation potential of fuelwood. We used
data from Mingchong Liu in Wolong Nature Reserve (2000-2003) to estimate the amount
of forest biomass planted in Wolong for the Grain to Green Program. Tree volumes were
determined using an initial stand density of 5000 trees/ha (5 trees/ 10m2), an average
density calculated in the field, and initial age at year 5. With this estimated initial
density, we calculated initial volume to be 0.5 m3/ha (5000 trees/ha x 0.0001 m3/tree) of
woody biomass 5 years after planting. In addition, we developed a mortality function

that was:

y = 8886’ + 0.0018x2 - 0.1609x + 7.692 (5.2)

99

which killed off a higher rate of younger trees than older. The function was developed by
determining the polynomial regression between the following mortality rates: 7% at year
5. 5% at year 20, and 3% at year 70, and 2% at year 80, year 90, and year 100. The
predicted volume grown in plantations was then used for mitigating fuelwood collection
on natural forests. In other words, we determined the increase in panda use of natural
forests when plantation forests were harvested for fuelwood instead by equating the
amount harvested in plantations to an equal amount of forest volume allowed to remain
standing in natural forests. The difference in the forest volume that remained in natural
forests was then related to a change (an increase) in panda use based on the regression
between forest volume in 342 field plots and the mean proportion with panda feces
described earlier (p. 97). Using the results of this simulation, we then demonstrated the
mitigation potential of reforested areas. We did this by showing the increase in potential
of finding panda feces in a natural forest when forest harvesting occurs in plantations

instead of natural forest.

Results

Potential of Reforestation Areas to Create Habitat:

The ages of the 69 plots in reforestation areas we surveyed ranged from recently planted
(<1 year) to old plantation (50 years). Mean reforestation age was 30.5 i 3.9 years
(_+_SE). Characteristics of reforestation areas were compared to an equal number (N=69)
of neighboring natural forest plots. Distances between the centers of reforest plots and

the natural forest plots ranged from 43.5 to 314.7 meters. Natural forest plots

100

Table 5.1. Mean (: SE) values for reforestation and natural forests with mean
comparison tests (Binom. Prop.= Binomial Proportion Test, T-test= 2 sided t-test,

Wilcox= Wilcoxan Rank Sum Test).

 

 

Reforested Plantation Natural Forest
Mean SE Mean SE Score P Value Test
Percentage with panda feces 1.45 13.04 -2.40 0.017 Binom Prop. 1
Age (years) 30.47 3.87 54.52 7.01 -1.18 0.237 Wilcox
Slope (deg) 23.18 2.89 25.60 1.30 -2.06 0.040 Wilcox
lBasal Area (m2/ha) 9.24 0.99 4.16 0.60 4.12 <0.001 Wilcox
Total Overstory Cover (%) 42.99 2.90 31.08 2.71 2.99 0.003 T-test
Conifer Overstory Cover (%) 41.09 2.90 14.41 2.83 6.56 <0.001 T-test
Deciduous Overstory Cover (%) 1.90 0.86 16.84 2.09 -7.38 <0.001 Wilcox
Average Overstory Height (m) 15.64 0.71 21.21 0.74 -5.46 <0.001 T-test
Total Midstory Cover (%) 9.88 1.65 24.60 2.27 -5.30 <0.001 Wilcox
Conifer Midstory Cover (%) 6.00 1.42 2.72 0.77 1.64 0.100 Wilcox
Deciduous Midstory Cover (%) 3.88 0.89 19.69 2.06 -6.44 <0.001 Wilcox
Average Midstory Height (m) 4.88 0.67 11.65 0.76 -6.65 <0.001 T-test
Average Shrub Height (m) 2.66 0.19 4.35 0.25 -5.33 <0.001 T-test
Total Shrub Cover (%) 19.39 2.20 26.67 2.29 -2.29 0.023 T-test
Bamboo Cover (%) 7.52 2.66 49.82 3.83 -8.57 <0.001 Wilcox
Total Understory Cover (%) 64.35 3.50 25.58 3.42 7.91 <0.001 T-test
Average Understory Height (cm) 48.56 3.31 25.32 2.86 5.30 <0.001 T-test

 

 

 

 

 

 

101

ranged in age from recently harvested (approx. 1 year) to unharvested (estimated mean
age = 150 years), with mean natural forest age 54.5 i 7.0 years (:SE)(Table 5.1).

Based on the presence of feces, giant pandas showed preference for natural
forest habitats to reforestation areas. We found panda feces in only one reforestation
plot, so interpretation of results must be conservative. Natural forest had a significantly
higher amount of feces than artificially regenerated forest (13.0% and 1.4%,
respectively)(Z = -2.3975, p = 0.017)(Table 5.1).

Panda use varied by age of forest (F4,t33=10.49, p=0.002) with higher amounts
of use in older harvested and old-growth forest. In natural forest, feces occurred between
I | to 100 years of age as well as in old-growth forest. No feces were found in areas
harvested less than 10 years ago. We found high proportions of use between 31-100
years (39.1%) and old-growth forest (27.8%). The proportion of feces in natural forest
harvested 31-100 years ago had significantly higher amount of use than areas harvested
earlier than 31 years (Fl,(,3=3.53, p=0.01). The one site in artificially regenerated forest
that had feces had been planted approximately 40 years ago and was one of the oldest,
most remote plantation sites. Giant pandas were not found to use areas that were planted
more recently than 40 years.

Forest structure varied between reforested areas and natural forest sites (Table
5.1, Figure 5.1). Generally, plots in reforested areas with no feces present had higher
overstory cover (F3,130=4.47, p=0.005), lower midstory cover (F3_131=8.73, p<0.001), and
significantly lower bamboo cover (F3,;33=31.37, p<0.001) than surrounding natural forest
(Figure 5.1). On average, artificially regenerated forests had higher overstory coniferous

canopy cover, and a denser and taller understory (Table 5.1). Natural forests had a taller

102

Percent Cover

80

 

 

   
 

- Artificial Forest- Feces

1: Artificial Forest- No Feces
8m Surrounding Forest- Feces
1:1 Surrounding Forest- No Feces

    

601

 

 

 

40«

20‘

 

 

 

 

 

 

 

 

     

Overstory Midstory Shrub Bamboo

Figure 5.1. Mean (iSE) percent coverages for total overstory, midstory, shrub and
bamboo strata in artificially regenerated forest and surrounding natural forest for areas

with and without giant panda feces.

103

overstory with a greater deciduous component and a taller, denser, more deciduous
midstory (Table 5.1). Perhaps most significantly, the surrounding natural forest had
much higher bamboo coverage than the reforested areas (Table 5.1). The differences in
forest structure may help to explain the use of habitats by the giant panda.

Logistic regression models indicated several forest characteristics that best
predicted the presence of panda feces. For reforested areas, average midstory height,
bamboo cover, and average understory height had the most significant univariate logistic
p-values (Table 5.2). When the full logistic model was fit to the reforestation data,
bamboo cover was the only variable that predicted panda presence (Table 5.3). Again,
we urge cautious interpretation of these results because only one plot had feces present in
reforestation areas.

In natural forest, univariate logistic models indicated that conifer midstory
cover, conifer overstory cover, forest age, understory height, and forest basal area were
associated with panda feces presence (Table 5.2). Unexpectedly, bamboo was not a
significant factor in the prediction of panda feces in natural forest surrounding
reforestation areas. The full logistic model for natural forest indicated that the coniferous

component of the midstory best predicted the presence of panda feces (Table 5.3).

Conservation Potential of Reforestation Areas:

The amount of forest area impacted by forest harvesting differed when various
levels of harvesting occurred in different aged forests. Harvesting 10.6
m3/year/household impacted a greater spatial extant of young natural forests more than

older forests. For example, a single household collecting its yearly supply of fuelwood in

104

Table 5.2. Univariate Logistic Regression P-values.

 

 

 

 

Reforestation Natural F375?—
Forest Age (years) 0.7888 0.0351
Patch Size (sq. km) 0.0598 0.1214
Slope (degrees) 0.9940 0.7250
Basal Area (m2/ha) 0.5689 0.0497
Average Overstory Height (m) 0.1302 0.0650
Total Overstory Cover (%) 0.1465 0.1202
Conifer Overstory Cover (%) 0.1343 0.0179
Deciduous Overstory Cover (%) 0.6686 0.1052
Average Midstory Height (m) 0.0013 0.1862
Total Midstory Cover (%) 0.5069 0.4727
Conifer Midstory Cover (%) 0.3520 0.0014
Deciduous Midstory Cover (%) 0.4228 0.7211
Average Shrub Height (m) 0.3659 0.1632
Total Shrub Cover (%) 0.3123 0.1389
Bamboo Cover (%) 0.0424 0.3581
Average Understory Height (cm) 0.0593 0.0479
Total Understory Cover (%) 0.1346 0.5824

 

 

Table 5.3. Final multivariate logistic regression models.

 

 

Estimate P-value AIC l

 

Intercept Variable
Reforestation Areas -6.2458
Natural Forest -2.8182

 

Bamboo Cover (%)
Conifer Midstory Cover (%)

0.056

0.1614

0.051

10.3

2
0.003 36.81J

 

105

Area of Natural Forest Impacted

 

0.6

 

 
 
   
  

 

 

 

 

 

 

o 100% harvest(10.6 m3/yr)
o 5 . o 50% harvest (5.3 m3/yr)
' v 25% harvest (2.65 m3/yr)
A 12.5% harvest (1.33 m3/yr)
0 4 Least-square Regressions
g 0.3 ~
0.2 —
0.1 e
M
0.0 I I I I I I I I
0 20 40 60 80 100 120 140 160

Age of Forest
(years since harvest)

Figure 5.2A. The area of natural forest impacted by various levels of forest harvest (1.33-
10.6 cubic meters/year) in different aged natural forests. Values represent mean ages
from 5 age classes (0-5, 6-20, 21-40, 41-80 years, and unharvested forest) and forest

volume from 342 field plots which established volume per acre.

106

young forest (0-10 years) impacted nearly 0.5 ha, whereas the same amount of fuelwood
collection in old-growth forest (150 years) impacted a smaller area (0.1 ha)(Figure 5.2A).
In terms of forest biomass, a forest with higher volume of woody material had
less area impacted by fuelwood collection (Figure 5.28). If forests had approximately
100 m3/ha of forest volume, over 0.1 ha was cut (at 100% harvest). Alternatively, in high
volume forests (i.e., old-growth forest) where we measured over 500 m3/ha of forest
volume, less than 0.04 ha would be harvested by one household (Figure 5.28). The large
amount of area impacted in newer, low volume forest is due to lower amounts of woody
biomass in young forest, thereby causing more areas to be harvested. As expected,
decreased levels of harvesting impacted proportionally smaller amounts of natural forest.
For example, a decrease in annual fuelwood harvest by 25% (2.65 m3/year/household)
lessened the impact on natural forest by 25% so that only 0.13 ha of young forest were
impacted (Figure 5.2). We modeled the relationship between tree volume (m3) and
potential of finding feces using data collected in the field (Figure 5.3). A 2“d order

polynomial regression fit the dataset best (r2=0.64) with the relationship being:

y = 0.0049x -1 .0e'5 x2 + 0.1626 (5.3)

where, y = proportion of sites with giant panda feces present and x = volume of woody
(i.e., tree) biomass (m3). Harvesting for fuelwood at the full, half, and quarter levels all
reduced the potential of feces presence, but at different amounts. For example,
harvesting 10.6 m3/year in areas with low initial forest volume (10 m3/ha) caused the

greatest decrease (5.2%) in the potential of finding feces (Figure 5.4). Harvesting in

107

Natural Forest Impacted

 

 

 

 

 

 

 

 

 

0.12 -
0,10 ~ + 100% harvest (10.6 malyr)
-O— 50% harvest (5.3 m3/yr)
+ 25% harvest (2.65 m3/yr)
0.08 - —A— 12.5% harvest (1.33 m3/yr)
8
v 0.06 -
0.04 -
0.02 _
0.00 I l T I 1 I ' I '

6-10 11-15 16-20 21-25 26-30 31-35 36-40 41-45 46+

Total Forest Volume
in 30 m2 Plots
(m3/30m2)
Figure 5.28. The area of natural forest impacted by various levels of forest harvest (1.33-
10.6 cubic meters/year) in natural forests with different volume stands. This direct

relationship describes harvesting levels and forest densities found in Wolong Nature

Reserve, 2001-2003.

108

1.0

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

Proportion with Feces

 

v = -1E-05x2 + 0.0049x + 0.1626

R2 = 0.6403 .

 

 

 

 

60

80 1 00 1 20 140
Wood Volume (m3/hectare)

I60

*1?

180

200

Figure 5.3. A model of the relationship between tree woody volume (m3) and potential of

finding feces using data collected in the field.

109

 

0.06

 

   
   

 

 

 

 

 

 

g —o— 100% harvest (10.6 m3/ha)
8 0'05 O 50% harvest (5.3 m3/ha)
8 —v— 25% harvest (2.65 m3/ha)
ti
0, 0.04 -
0
0
0
u.
g 0.03 -
8 o
0. O o
c 0"'--o...
0 ----o._.
a) O O
‘5 O
8 0.018 +‘YT‘7‘—v~_'_+ O
Q ”"~-'_
‘V-—-'_
+~'__'
0.00 r r r fl r r r
0 20 40 60 80 100 120 140 160

Initial Forest Volume
(ms/ha)

Figure 5.4. Habitat degradation determined by the change in panda feces proportion
found in 342 field plots (high decrease = high degradation) from various levels of forest

harvest in natural forests that range from having low to high initial forest volume.

110

areas with higher initial forest volumes also caused a decrease in panda use, but the effect
was less. For example, harvesting 10.6 m3/year in areas that initially had 100 m3/ha
caused a decrease in the potential of finding feces of 3.3% (Figure 5.4). Lowering the
fuelwood harvest by half and a quarter reduced the effects by proportionally equivalent

amounts (Figure 5.4).

Conservation Potential of Larix potaninii var. macrocarpa:

Plantation volume increased over the first 52 years of stand development to 342.8
m3/ha. As the plantation aged past 52 years, total volume decreased as individual growth
declined and natural mortality continued (Figure 5.5). If 100% of the volume produced
were used for fuelwood, it would provide enough annual fuelwood for 10 to 30
households, depending on the age of plantation (Figure 5.5).

To determine the impacts mitigated by harvesting plantation forest, we modeled
100%, 50%, 25% and 12.5% harvesting of plantation forest biomass (Figure 5.6).
Harvesting from plantation forests was correlated to an equal amount of natural forest
that was not harvested. We modeled this by adding the same amount harvested in
plantation to the amount of natural forest volume in natural habitats. The higher the
harvest of plantation forest the greater the increase in proportion of sites with panda feces
in natural forests because more areas of natural habitat were preserved (Figure 5.6). In
other words, a 100% harvest of plantation forests preserved more natural forest than 25%
plantation harvest. Generally, older plantations offered more woody biomass for

fuelwood and therefore preserved more natural panda habitat. For example, if 100%

Ill

 

   

 

 

 

 

 

400 35
350 30
300
'5‘ 25
5
“5’ 250
E — 20
2 200 i ~ — -——;V—_—. _'—' _-————‘_ '—’ ‘ '—.—"— —— —
>0 I Total Volume (Live + Dead)1 ' 15
13 15° “ " 7 A Larix potaninii (Live Stems) 1 ”‘
' ~59 o-— Number of Households l 10
100 - - -_ _ _
50 = = - - 5
[LA
0 - . . 0
1o 20 30 40 50 60 70 80 90
Age of Plantation (Years)

Figure 5.5. Pole volume from plantations of Larix potaninii var. macrocarpa and the
number of households that can harvest 100% of their fuelwood needs (10.6 cubic
meters/year) from the total volume in these plantations. The simulation was conducted
between 10-90 years because these were the years available for the species growth

equation used in the model.

112

 

 

 

 

I s‘fiioii/TeIa—ntét—icinififitfi
0.2 5’ j - - , ,- - , - -. _.= 50% Plantation Harvest l
, A 25% Plantation Harvest l
- x12.5% Plantation Harvest

      

Increase in Proportion of Sites
wrth Panda Feces

 

 

0 ,
1 0 20 30 40 50 60 7O 80 90
Age of Plantation
(Years)

 

 

Figure 5.6. Proportional increase in the number of natural forest sites used by giant
pandas when plantations are harvested instead of natural forests. This relationship was
established based on the woody volume of plantation forests over time (see Figure 5.5)
and the proportion of sites with feces in natural forests of varying woody volumes (see
Figure 5.3). The simulation was conducted between 10-90 years because these were the

years available for the species growth equation used in the model.

113

ofa 50-year-old plantation is harvested, the amount of biomass preserved in natural forest
may increase the potential of panda use in those natural habitats by over 80% (Figure

5.6).

Discussion

We found giant panda feces in 1.4% of plots (1 of 69) in reforestation areas, and
therefore concluded that pandas rarely use this type of artificially regenerated forest.
However, use of this plot is notable because it provides evidence that in some cases, the
giant panda is able to use reforested areas if certain conditions can be met. We believe
use of this particular reforestation plot by a panda is explained by the plot’s location and
surrounding forest characteristics. The plot was located far from human disturbance
(over 2 km from roadway) and was characterized with a healthy arrow bamboo
(Bashaniafabri) stand, as demonstrated by the high proportion of bamboo in the plot
(75%) and evidenced by its significance in the logistic regression of reforestation plots.
However, the vast majority of reforestation plots had no bamboo cover or significantly
lower bamboo cover and were therefore not used. Given the importance of bamboo in
determining the likelihood of use by pandas in reforestation areas, methods to propogate
bamboo under the more acidic conditions of Larix spp. plantations should be considered.

The majority of the tree species planted in Wolong Nature Reserve are fast
growing species (i.e., Larix spp.) which are non-native to the forests and climate of
Wolong. Other species that are native to Wolong, more suitable for Wolong’s climate. or
grow easily in Wolong’s soil may provide a better fuelwood source and may have the

additional advantage of allowing bamboo and other understory, shrub, or midstory

114

vegetation to grow. If different planted species allow more bamboo or other vegetation
to grow in its community, the likelihood of these planted forests creating panda habitat
may be greater. However, the tree species that are currently used to reforest areas of
Wolong do not appear to provide conditions suitable for other native vegetation to grow
and therefore do not create suitable panda habitat. Based on the results of this study, care
should be taken to not interpret the planting of trees or the reforestation of an area to be
equivalent to the creation of panda habitats.

We also evaluate the potential of reforestation areas in Wolong Nature Reserve to
mitigate fuelwood collection effects on giant panda habitats by modeling the relationship
between natural forest volume, Larix spp. growth, and the potential of finding panda
feces. F uelwood collection in Wolong previously occurred in the natural forests
surrounding the households (An et al. 2002, Linderman et al. 2005a), however, since the
implementation of two new conservation policies in China (the NFCP and the GTG),
harvesting has dwindled and amount of reforested areas have increased (Liu et al. 2001,
Liu and Diamond 2005). Although present day harvesting does not currently have a large
impact on China‘s natural forests, the two policies may expire in the next several years. at
which point increased harvesting may again occur throughout China. Given this time
period, trees originally planted as part of the GTG program may grow to a size that may
provide a significant source of fuelwood.

We found that an average household in Wolong Nature Reserve harvesting
natural forest at the levels described by Liu et al. (1999b) impacts approximately 0.4 ha
each year whether harvesting takes place in a young or secondary forest (0-40 years).

Harvesting in older natural forest impacts less of an area. Similarly, harvesting in forests

115

with low volume affects larger areas while cutting in high volume forest impacts a
smaller area. Currently, while the harvesting ban is in place, harvesting of whole trees
continues to be rare and effects of harvesting on panda habitats are therefore minimal.

Our model indicates fuelwood collection in natural forests reduces the likelihood
of finding giant panda feces. Forests with greater woody volume are less sensitive to
changes in probability of panda feces, and therefore less sensitive to harvesting, than
were low volume forests. Harvesting in forests with lower volume has a greater
influence on the potential occurrence of giant panda feces because it impacts a larger area
and because forests with lower wood volume have lower proportions of feces. This
relationship only accounts for proportional change and not total change. Therefore, it
does not indicate that harvesting in old growth natural forests has less of an effect on
pandas than harvesting in low volume forest because the total probability of panda feces
occurring in high volume forest is much greater.

Planted forests can mitigate impacts of fuelwood collection on natural forests,
although this varies with species, age, and amount harvested. During the early stages of
forest growth, roughly 4 households could collectively gather their fuelwood over 1
hectare due to the low amount of biomass. However, if forests are allowed to grow past
this initiation stage, biomass grows steadily until the trees are approximately 70 years
old, at which point it could provide enough wood for approximately 10 households for
each hectare using 10.6 m3/ha of fuelwood/year/household (Figure 5.5). If rates of
household fuelwood consumption are lower, the number of households that these
plantations provide fuelwood for could increase to 20 (at 5.3 m3/yr/household), 40 (at 2.7

m3/yr/household), or more.

116

A cumulative total of 266.7 ha were planted in 2000, 333.3 ha in 2001, 414.6 ha
in 2002, and 448.5 ha in 2003 with primarily Larix spp. for the Grain to Green program
(Wolong Nature Reserve 2003). Assuming a conservative estimate of 1000 trees/ha (1
tree/10 m2)(20% of the average density measured in the field), woody volume produced
by this policy may well exceed 80 m3/ha after 80 years of growth. Permitting harvesting
of the planted trees allows natural forests to regenerate and to experience less human
disturbance. Harvesting of planted trees then provides better giant panda habitat in
natural forests. If rotations are developed so that harvesting only occurs in reforested
areas every 80 years, there is the potential that panda use of natural forests may increase
by greater than 80% (Figure 5.6).

The placement of planted trees and reforested areas should be taken into
consideration when planting artificially regenerated forests. Our findings show that
present reforestation areas do not provide suitable giant panda habitat but do offer a
potential alternative fuelwood source. In order to prevent these areas from impacting
natural wildlife habitats, they should be placed close to households. This would increase
the likelihood of their use, decrease human effort to collect the wood, and decrease levels
of pollution and environmental disturbance. In societies where fuelwood is still a
necessary source of energy, alternatives should exist that reduce the impacts on natural
forests. Artificially regenerated forests may potentially provide that alternative. For this
reason, artificial forests may be an significant yet overlooked component of wildlife

conservation in these areas.

117

CHAPTER 6

SCALING UP FROM SURVEY SAMPLES TO THE LANDSCAPE LEVEL:

GIANT PANDAS ACROSS HETEROGENEOUS LANDSCAPES

118

CHAPTER 6

SCALING UP FROM SURVEY SAMPLES TO THE LANDSCAPE LEVEL:
GIANT PANDAS ACROSS HETEROGENEOUS LANDSCAPES

Introduction

To understand the relationships of organisms to their environment, researchers
typically collect ecological information from a set of spatially distributed sample plots.
They may then attempt to establish key relationships (models) from the plot data based
on the premise that the plots are representative of the broader landscape, and extrapolate
these relationships to landscape level predictions. Throughout this paper, we refer to this
process of using field data to generate landscape level results as “scaling-up.” The
landscape models that result from scaling-up can then be used in various ways, including:
predicting species occurrence (F leishman et al. 2001, Fertig and Reiners 2002), mapping
habitat suitability (Danks and Klein 2002, Gross et al. 2002, Niemuth 2003, Store and
.Iokimaki 2003), analyzing spatial patterns (Franklin et a1. 2002, Ray et al. 2002), and
investigating human impacts (Hawks et al. 2000, Cramer and Portier 2001, Rao and Pant
2001).

At the landscape level, we often must use models because it is not practical to
study the entirety of a large area in detail. Landscape models have proven to be valuable
tools (Turner et al. 1995, Liu and Taylor 2002) because models based on plot data could
extrapolate what a survey would have found from the plot level to the entire landscape.
However, all models have errors. First, errors may be produced during the statistical

analysis used to generate the models. For example, the selection of independent variables

119

may be important in determining the extent of errors that are produced (i.e., the principle
of parsimony) (Leger and Altman 1993, Zar 1996). Second, errors may result from
necessary assumptions and imperfect knowledge. For example, suppose we know that
species ‘X’ lives only in coniferous forest. At any given point in a landscape, we can
reasonably predict its suitability for species X merely by the proportion of conifers
present. Now assume we want to map the habitat ofX across an entire state. A
landscape model is necessary because it is not practical to visit every location. However,
a problem arises because our landscape level forest map does not distinguish conifer
forest from other forest types. This map only includes a “forest” variable. In this case,
the landscape model will perform poorly, not because of a bad model or poor
understanding of the species, but because the data limitation leads to an
oversimplification of reality. Nearly all landscape models must do this to some degree.
However, few models accurately quantify the amount of information lost during this
stage. Rather, models are typically selected based on criteria such as Akaike’s
lnfonnation Criteria (AIC). For example, models may be created or resampled at several
different spatial scales (e.g. grain size) and the relationship between the scale and a
variable are of primary interest. Here, the focus is often on selecting the model with the
spatial scale that fits the data best. In many instances, the loss of information from the
actual scaling up procedure is not explicitly defined and may be neglected altogether. In
addition, even if a variable measured in the field is ecologically significant, it cannot
always be mapped at the landscape level or incorporated into the final analyses. This
forces the modeler to reduce the number of variables to only those that can be mapped,

which may result in the final landscape level model having even less predictive value.

120

To demonstrate the major issues of scaling-up field data to the landscape level, we
did a case study of the giant panda (Ailuropoda melanoleuca) in Wolong Nature Reserve,
Sichuan, China. Wolong was suitable because spatial data collected in the field were
available and could be used towards the development of an effective full model. In
addition, this study enabled us to generate a landscape level occurrence probability map
for the giant pandas in Wolong.

Using the spatial data collected in the field, we developed logistic regression
models to differentiate between areas with and without the presence of giant panda feces.
In this paper, we show that at the survey plot level, the model accurately differentiated
between areas with a high probability and low probability of giant panda use. However,
as we scaled the model up to the landscape level, our ability to differentiate between
areas with/without feces progressively declined in performance. Some of the declines
were expected since we oversimplifyed the world. Other declines were not expected, and
this highlights the future research needs for understanding panda occurrence at the

landscape level.

Methods
Study Area
The study was conducted in the Wolong Nature Reserve (102°52’-103°24’E,
30°45’-31°25’N), Sichuan Province, China (Figure 6.1). Wolong Nature Reserve,
hereafter Wolong, is one of the largest reserves (approx. 2000 kmz) established
specifically for the protection of giant panda habitat (Liu et al. 2001). At the foothills of

the Himalayan Mountains, Wolong has a high degree of topographic relief, rising from

121

1200 m above sea level in the southeast to over 6250 m in the northwest (Schaller et al.
1985, Liu et al. 1999b)(Figure 6.1). This coarse, crumpled topography leads to a wide
diversity of forest-types throughout Wolong. Deciduous forests and umbrella bamboo
(F argesia robusta) predominate below 2000 m (Schaller et al. 1985). Above 2000 m,
coniferous species are often found as well as large patches of arrow bamboo (Bashania
fabri). Subalpine coniferous forests and vast stretches of arrow bamboo predominate
above 2600 m (Schaller et al. 1985). The ages of the forests also vary because local
people living within the reserve have historically harvested trees for timber and
fuelwood. Thus, the study area is characterized by a highly heterogeneous environment
due to factors which include topography, forest and bamboo characteristics, and age of

forest due to harvesting.

   

Elevation

1000 - 2000
[:l 2000 - 3000
its...” 3000 - 4000
- 4000 - 5000
- 5000 - 6250

ll Wolong Nature Reserve

0 10 20 30 40 Kilometers
5::

 

Figure 6.1. The location and elevation of Wolong Nature Reserve, Sichuan, China.

122

Field Sampling Methods for Data Collection

To determine how giant pandas use heterogeneous landscapes, we examined use
patterns in forests of various ages throughout much of Wolong. Data collection took
place during May-August 2001, May-November 2002, and June-September 2003. Use
was determined by the presence of fecal droppings. Droppings are a useful indicator of
panda use because pandas deposit feces frequently, although actual sightings of animals
are rare (97 droppings/day, 4 droppings/hour) (Schaller et al. 1985) and feces remain for
several months (personal observation, SLB & HJY). Presently, a radio-collaring ban on
giant pandas restricts the use of any radio-tracking methods. Until the ban is lifted, feces
remain the most efficient way to determine use of habitats. Throughout this paper we
refer to the presence/absence and probability of finding panda feces as being equal to the
presence/absence and probability of finding the giant panda. Although we realize that
this is not the reality, the presence/absence and probability of finding feces is the best
available method to relate to actual use by pandas.

In the design of the field survey, we sampled a total of 913 30 m2 plots. Primary
plots were located randomly in the field (in areas where physical access was possible),
were placed approximately 200 meters in horizontal distance from the nearest plot, and
were identified by local guides knowledgeable of harvesting history in the area. The
center position of each 30m2 plot was geo—referenced with a Pathfinder® Pro XRS GPS
unit (Trimble Navigation Limited) and OmniSTAR DGPS real-time correction
(OmniSTAR USA, Inc.). Data were then associated with Landsat imagery and other
spatial data for analyses. Due to logistical and time constraints which prohibit

revisitation and reexamination of sample plots for passed droppings, equal probability of

123

feces detection in all habitats was assumed. Reid and Hu (1991) support this assumption.
They found a low ratio of missed droppings (1.3%) throughout their study of panda
habitats in Wolong.

Adaptive cluster sampling was used to assist in determining which areas pandas
might have visited. This method is designed to sample rare events (i.e., feces) and allows
for focused research in areas where rare events occur while maintaining proper statistical
hypothesis testing protocol (Thompson 1991a, Thompson and Seber 1996). Whenever
feces were found (i.e., presence=l) in a 30 x 30 m sample plot, all units in its
neighborhood (those 30 x 30m plots to the North, South, East and West) were added to
the sample. If in turn any of these subsequently added plots had feces, the plots of its
neighborhood were also added to the sample, so that finally the sampling neighborhood
contained 1 (primary only), 5 (1 primary, 4 secondary), or 8 (1 primary, 7 secondary)
sample plots. Of the 913 plots sampled, 443 were primary plots and 470 were secondary
plots.

Forest characteristics measured include a variety of geographical, forest and
bamboo variables (Table 6.1). We changed three variables to pseudo-variables for more
appropriate analyses. Aspect was converted to ecological aspect, a value which
categorizes 360 degrees into 21 ecologically meaningful units (Parker 1982). Elevation
(MSL) was divided into 200 m interval categories and within each category we calculated
the proportion of plots with feces. The proportions were then used to give relative
weights (values 1-10) to each elevation category. Age of forest was also divided into five

class categories (Table 6.1).

124

Table 6.1. A list of environmental, forest, and bamboo characteristics measured in the

field and used in the various models to produce a map of giant panda occurrence.

 

Variables Model Ty

 

. Field Data Full Reduced Simple Landscape
Elevation X X
Age of Forest
Slope
Ecological Aspect
Total Overstory Cover
Conifer Overstory Cover
Deciduous Overstory Cover
Average Overstory Height
Total Midstory Cover
Conifer Midstory Cover
Deciduous Midstory Cover
Average Midstory Height
Total Shrub Cover
Average Shrub Height
Bamboo Cover
Total Understory Cover
Average Understory Height

><><><><
><><><><

><><><

X
X
X

><><

 

 

 

 

 

 

><><><><><><><><><><><><><><><><><

 

 

125

Landscape Data

The landscape data consisted of a forest map and elevation data. The forest map
was from a supervised classification of a Landsat 5 image (13 June 2001). Estimated
accuracy was 86%. The elevation data were a combination of the Shuttle Radar
Topography Dataset (SRTM, available at http://www.jpl.nasagov/srtml) and digitized
topographic maps. We used elevation derived from paper topographic maps to fill in data
voids in the SRTM data. All landscape level analyses were conducted using ERDAS

Imagine 8.6.

Modeling Procedure
As models are scaled-up, or as field data are integrated into a landscape model,

errors might occur in each step of the scaling up process (Figure 6.2). At the initial stage,
there are issues related to field data collection (Figure 6.2A). These may include
problems with sampling strategy (e.g. were proper sampling methods used?),
autocorrelation and sampling errors that may lead to the data being a distortion of reality.

To fully understand the relationship among variables collected in the field, a Full
model can be created (Figure 6.28) that contains all variables from the field data except
for those found to be redundant or non-significant. The model simplifies reality but also
allows general relationships to be more easily defined. This simplification causes the
difference between the model results and reality to become even greater than the
difference between field data and reality.

The most complex models (e.g. Full models) are typically undesirable for scaling-

up to the landscape level. This is because data for many variables are ofien not available

126

across an entire landscape. For example, understory height may be an important
characteristic for a mammal species, but it is useless at the landscape level unless
understory characteristics are known across the landscape. To gain applicability and
parsimony, variables that are correlated with other datasets (i.e., independent variables),
not found to be significant, or not mapped at a landscape level are removed from the
model, creating a model with fewer variables (Figure 6.2C). This Reduced model is
theoretically the best model we could make for a landscape analysis. Our expectation is
that such a model may still perform well, but not as well as the Full model. If all the
remaining variables have spatial datasets available, then the datasets can be implemented
into a landscape model and spatial analyses performed. Typically, the number of
variables needs further reduction because not all spatial datasets are available for the
remaining variables. This creates what we call a Simple model (Figure 6.2D). It is
simpler than the Reduced model because it often has only a few variables, making it
easier to apply but at the cost of accuracy. Variables used are typically the ones that are
available across a landscape, and are not necessarily the ones that are most important for
a species of study (Figure 6.2D). The Landscape model is a spatial implementation of the
Simple model. Both models are mathematically the same, but the Landscape model uses
landscape scale maps, rather than the plot level data used in earlier models.

The resulting loss of modeling power that occurs as variables are removed may
have the undesirable effect of model output becoming meaningless. Yet the loss of

modeling power that occurs as plot sample data are transitioned to higher spatial levels,

127

Figure 6.2. A characterization of the scaling-up process. Seven variables are sampled in
the field (A) and used to generate a Full Model (8). Numbers within each sampling plot
represent probability of species occurrence values for a theoretical scaling-up modeling
exercise. A Reduced model (C) includes only five of the original variables because not
all variables are mappable. Only three of the variables are presently available to be used
as input into a landscape model (D), thereby creating a simplified model that may have

higher errors than the Full and Reduced models.

128

 

  
 

.60 .42 .21 14
”{Wg‘g/ngfl/ D) 8821365814.
/(‘I3/Or27/020/0 21/ MODELS
/0/.43/6.37/0/.33/0/.24/

 

 

VARIABLE 3
VARIABLE 5

 

 

/0~2V C) REDUCED

 

 

 

VARIABLE 2
VARIABLE 3
VARIABLE 5
VARIABLE 6
VARIABLE 7

 

 

/ .
VARIABLE 6 V\ .53 .22 _.

/'

\

 

 

 

 

 

 

 

/0 21 A B) FULL
/ ' / MODEL
/0.37 7/(5 25 //019/
VARIABLE 1 e——/
VARIABLE 2 /
VARIABLE 3
VARIABLE 4
VARIABLE 5
VARIABLE 6
VARIABLE 7 \ _ _
“63%“ . A) REALITY

129

such as the landscape level, is not always fully appreciated. Therefore, we use this
method of scaling-up field data on giant panda occurrence. We first create a Full model
that includes all independent forest and geographic variables measured in the field. Then.
we reduce the number of variables to only those which are significant to develop a
reduced model. Eliminating more variables, we reduce the model further until we
eventually create simple and landscape models. The results of each of these models
allow us to characterize a scaling up procedure. From this, we are able to quantify the

loss in interpretive ability as datasets are scaled up from field data to the landscape level.

Statistical Analysis

Using two-thirds of the field data, we developed logistic regression models. One-
third of the dataset was reserved for later model validation. To remove non-significant
variables, we ran separate univariate logistic regression models for each variable to the
presence/absence of panda feces and removed variables with P-values greater than 0.25
in G-test scores. We then used Pearson correlation coefficients to detect redundant
variables. Correlated variable pairs with a r-value greater than 0.50 were considered for
elimination. For each pair, the variable with the more significant univariate G-test was
retained for further analysis (Klute et al. 2002). Following the removal of insignificant
and correlated variables, we fit a multivariate logistic regression to the remaining
variables. All models were run in SAS v.8.02.

Our logistic regression models used the equation presented in Quinn and Keough

(2002)

I30

fl0+fl1X1+fl2X2 +"'l6po

 

fl-(x) : 1 ‘1‘ eflo+flrX1+flzX2+'°'fl17Xp (6'!)

where 11(x) is the probability that panda feces is present for a given covariate
value (x), and Bp are coefficients for each variable, xp, to be estimated. To determine
probability values of panda occurrence, we acquired coefficients from logistic
regressions, created equations using coefficients (Equation 1), and then calculated
probability values for all survey plots. We considered a probability 3 0.40 to predict
panda feces and therefore panda presence. Although this threshold was arbitrary, we
chose it because one-third of the total field plots (primary + secondary) contained feces in
the original dataset and the threshold of 0.4 allowed us to reduce omission error without

increasing overall errors.

Begum

Logistic Regression Models

Of the 913 plots sampled, 320 (35%) had panda feces present. Seven variables
were either non-significant or covaried with more influential variables and were not used
in the Full model, which contained 10 variables (Table 6.1). The Reduced model
included 5 variables, while the Simple and Landscape models both included 4 variables
(Table 6.1). One variable, total overstory canopy cover, was initially removed because it
covaried with other significant variables. Later we used it in the Simple and Landscape
models because landscape level data were available for total overstory canopy cover but

not available for other covarying significant variables. Two variables, bamboo cover and

131

forest age, were removed from the Simple and Landscape models despite their
significance because spatial data layers across the landscape were unavailable.

All 4 models predicted plots that contained feces as having a higher panda
probability than survey plots without feces (Figure 6.3). A larger difference between
feces and no—feces points indicates a better ability to discriminate between occurrence
and non-occurrence areas or areas with and without giant pandas. This suggests that all
models are working appropriately, but they do vary in performance. The Full model
shows the largest separation, with the average probability of points with feces having a
30% higher probability than the no-feces plots. However, this separation declines as
models become simpler. A comparison of the average probabilities in the Simple and
Landscape models indicates a <10% separation (Figure 6.3). The decreasing difference is
mainly because of a decrease in the probability associated with feces points rather than an

increase in the probability associated with no-feces points (Figure 6.3).

0.7 , =~ f , -i-
0.6 1
0.5 .

0.4 '

 

0.3 -

Probability

0.2 - _ __ __ , - __ j

0,1 . . ,. __ _ __, ____ V--O_:-nofeEes’

Full Reduced Simple GIS

Model version

Figure 6.3. Calculated probabilities of giant panda occurrence grouped and averaged by

survey plots with/without panda feces.

132

To further determine what was happening within each model, we examined the
probability distribution of feces and no-feces points for each model (Figure 6.4). For all
models, the proportion of no-feces points was consistently greater in lower probability
areas while feces points were in higher probability areas. This fiirther supports that the
models are correctly identifying occurrence/non-occurrence areas (Figure 6.4). In the
Full model, the majority of no-feces points occur in areas with very low probability (0.1
probability) and the majority of feces points occur in areas with a probability of 0.4 or
greater (Figure 6.4A). Many no-feces points also occur in areas that have higher
probability of having feces, suggesting that these points may still be viable habitat even
though they do not have feces (Figure 6.4A). The Reduced model shows similar trends
to the Full model, with no-feces points occurring in lower probability areas (greatest
number at 0.1 probability) and feces points in higher probability areas. However, the
number of feces points in higher probability areas declines and there is a more uniform
distribution of feces as well as no-feces points across the range of probabilities (Figure
6.48).

The Simple and Landscape models generally show points with no-feces to have
low occurrence probability and points with feces to have high panda probability.
However, the distinction between areas with and without feces is less apparent.
Histograms of output from Simple and Landscape models appear to have shifted to
within 0.1 probability relative to each other, whereas the distinction was greater in the
Full and Reduced models. In addition, the Simple and Landscape models do not show

the high number of no-feces points in the

133

Figure 6.4. The proportion of survey plots that did/did not have giant panda feces

classified by probability levels of finding feces. Gray bars represent plots without feces.

Black bars represent plots with feces.

Proportion of points

Proportion of points

A. Full Model

(1.40 - i=7 -— ---, , -___ 9---- Am- , --,

 

 

0.10 '

 

 

 

Probability of feces

8. Reduced Model

0.40 a. .7 .— .-. -- , -_

 

0.30 ,. -7 a; .77; “km .._ - - -

0.20 -

 

 

 

 

Probability of feces

134

C. Simple Model

 

 

 

 

 

 

 

 

0.40 - ,- f - -__--_.____._. _ i
.23
.5 03,01--- ---,__.. —— -— “flw
0
a.
“-1
0
c: 0.20 _,
0
a:
H
o
8‘ 0.10 L
In
0..

0.00 . .#

0‘ ’0'? 0°” ,6“ ’0'? 91° lo“ of ,0“
Q 03 61’ of of of of 0" 0% 0
Probability of feces
D. Landscape Model

0.40 . _ . _-_______--
U) |
*0
r:
0—
0
a.
w
0
r:
0
:1:
H
0
a.
0
H
0..

 

 

 

 

ON ’0'} ’0'? 0‘" ’0‘? 91° ’6‘ ’ch 0°’
9 0‘ 61’ 0°? 0* of of 0’1 of 0
Probability of feces

135

very low probability areas (<0.l). Instead, they show an increased proportion of no-feces
points in areas of increased panda occurrence probability (Figures 6.4C and 6.4D).

To better understand why the data were converging as models became more
simplified, we developed error matrices to describe model performance (Table 6.2). The
Full model performs best (25% overall error rate) with a decline in performance to the
Landscape model (39% overall error rate) (Table 6.2). The primary source of increasing
error is in omission errors (plots predicted to have no feces but feces actually found).
Omission errors range from 8% in the Full model to over 30% in the Simple and
Landscape models. Commission errors (points predicted to have feces but none found, a
scenario which is more likely to occur in nature) range from 43% in the Full model to

over 56% in the landscape model (Table 6.2).

Probability Map of Panda Occurrence

The giant panda occurrence probability map from the Landscape model (Figure
6.5) displays trends similar to those in previous panda habitat maps and surveys (DeWulf
et al. 1988, Liu et al. 2001). Almost half of the reserve (~48%) has very low probability
(< 0.1) and 82% is below 0.4 probability, our threshold for predicting presence (Figure
6.5). Previous studies highlighted the potential problem of fragmentation of remaining
habitat, but our map more clearly defines the seriousness of the problem. Several large
areas of high occurrence probability occur with numerous smaller patches of medium to

high occurrence probability interspersed.

136

Table 6.2. Error rates associated with the various models. Horizontal error rates denote

omission errors while vertical errors denote errors of commission.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Full Model prediction
mOdel feces no feces Error rate
Field fecesl 77 7 0.08
data no fecesl 57 115 0.33
Error rate 0.43 0.06 0.25
Reduced Model prediction
model feces no feces Error rate
Field feces 70 20 0.22
data no feces1 66 122 0.35
Error rate 0.49 0.14 0.3]
Simple Model prediction
model feces no feces Error rate
Field fecesl 56 33 0.37
data no fecesl 63 126 0.33
Error rate 0.53 0.21 0.35
Landscape Model prediction
mOdel feces no feces Error rate
Field feces) 64 33 0.34
data no feces1 82 116 0.41
Error rate 0.56 0.22 0.39

 

 

 

 

 

 

137

 

 

 

 

|“ Wolong Nature Reserve
Field Plots

Panda Probability
< 0.2
" 0.2 - 0.4

i
i
x” -
.1 j
\ -/
10 Kilometers
->0.4

 

 

Figure 6.5. The probability map of panda occurrence showing the distribution of field

plots (both primary and secondary plots) used to create logistic regression models. Field

plots are focused within the central region of Wolong known to have concentrated panda

activity.
138

Discussion

Extrapolating field data to the landscape level is becoming increasingly common
in ecology. This raises important concerns over the issue of scale (Levin 1992, Morales
and Ellner 2002, Urban et al. 2002, Emlen et al. 2003). Information and power are lost at
many stages of the scaling-up process, fi'om the field sampling stage through the various
modeling stages. This information loss needs to be thoroughly understood to determine if
the output is usable for further analyses or for management practices. In our study, we
surveyed many aspects of giant panda habitat and included sampling areas known to have
resident panda populations in the Wolong Nature Reserve. From this study, we have
demonstrated the potential risks in scaling-up by using field surveys of the giant panda to
create a landscape level probability of occurrence map.

Our analyses show that there is a considerable decrease in the ability to predict
presence/absence of giant pandas as more variables are removed from the model. The
Full model, which includes all significant, non-redundant variables, works best. It
successfully distinguishes between feces/no-feces survey plots. While the Full model
performs well at the individual plot level, it requires too many data to extend to the
landscape level.

The Reduced model, which is less data intensive and possibly extendable to the
landscape level in the future, performs nearly as well as the Full model. The removal of
five variables between the Full (N=10) and Reduced (N=5) models appears to have had
little impact. However, decreasing the total number of variables by one in the Simple
model (N=4) had a much larger impact. The ability of the model to distinguish between

feces/no-feces points is far less than the Full or Reduced model. Additionally, the

139

proportion of points at the two extremes of panda probability (<0.1= low probability.
>0.9=high probability) converges toward the center as variables are removed.

Bamboo, recognized as the most important component of giant panda habitat and
found to be the most significant variable in the Full and Reduced logistic models, was
removed from the Simple and Landscape models. Total overstory canopy cover, on the
other hand, was removed from earlier analyses because it covaried with other significant
variables. However, because covarying variables were removed from later analyses and
we had spatial data available for this variable, it was included in the Simple and
Landscape models. The removal of significant variables for which we do not have
landscape level data, and their replacement with covarying, significant variables for
which we do have landscape level data, not only decreased the ability of the models to
differentiate between areas with/without feces, but also decreased our ability to interpret
results.

Sources of increasing error in the models come primarily from the omission error
rate, which steadily increased from the Full (8%) to the Simple (37%) and Landscape
(34%) models. Commission error rates, although not as significant because they can be
explained ecologically, also increased from the Full (43%) to the Simple (53%) and
Landscape (56%) models. Results suggest that if a higher number of significant variables
can be included in the Landscape model (i.e., spatial datasets are available for more
variables) then the predictive ability of the model will be better. However, the decreased
distinguishing ability of the Simple and Landscape models suggest that the output from
models that typically contain only a few significant variables needs to be carefully

examined, and sources of error must be addressed.

I40

The loss of the bamboo and age of forest variables in our Simple and Landscape
analyses due to the lack of relevant data not only decreased the models performance, but
also made it more difficult to interpret results. A model that does not include bamboo as
a variable should raise important questions concerning our results because bamboo is the
most important characteristic for the giant panda. Bamboo data for the entire Wolong
Nature Reserve were generated by Linderman et al. (2004), who classified Landsat
imagery using neural networks and found approximately 80% accuracy in bamboo
projections. However, because our study sampled different areas in separate parts of
Wolong Nature Reserve, we found decreased accuracy and therefore could not include
the data in the Simple or Landscape models.

As mentioned previously, it is not only the number of variables, but also the
significance and interpretability of the variables that is important. We demonstrated this
by showing that the removal of five variables had little effect as we moved from the Full
to the Reduced model. Our ability to differentiate between feces/non-feces plots only
decreased by <4%. However, only one less variable was used for the Simple and
Landscape models, but our ability to distinguish feces/non-feces plots decreased by
approximately 14%. The reason for this is because bamboo and age of forest variables
were removed.

Age of forest was the one variable measured in the field that allowed us to
determine human impacts on giant panda habitats. By knowing when and where forests
had been harvested, we were able to determine at what age a forest began to be used by
the panda, and how much panda activity increased with forest age. The probability of

finding feces steadily increases as the age of forest increases (Figure 6.6). This suggests

141

that harvesting forests for timber or fuelwood has a negative impact on giant panda
habitat and as that forest ages, habitat for the giant panda recovers. Once the forest has
40+ years to regenerate, the probability of finding feces in the area is greater than 0.4,
indicating that it is potential habitat (Figure 6.6). Because we are presently unable to
map age of forest across the landscape, it was removed from the Simple and Landscape
analysis. This made our ability to translate human impacts across the Reserve scale
impractical. While removal of forest age and bamboo variables decreased the
performance ofour models, it does clearly identify those variables as priorities if we are

to create better giant panda occurrence models in the future.

0.50 ,---..- »- - - . -- -L-1

 

0.40 - —

0.30 _ __-

   

 

   

.. -— Fullrnodel
.— -" Reduced modell

0.20 —- ~———

 

Probability of panda occurrence

 

0.00

 

AgeClass

Figure 6.6. Calculated probabilities of giant panda occurrence grouped by age of forest
for the Full and Reduced models (Age Class l= 0-5 years, Age Class 2 = 6-20 years, Age

Class 3 = 21-40 years, Age Class 4 = 41-80 years, and Age Class 5 = unharvested forest).

I42

The output of our final Landscape model is a map of the probability of panda
occurrence (Figure 6.5). Comparison of this map to previous giant panda habitat maps
and diagrams (Schaller et a1. 1985, DeWulf et al. 1988, Liu et a1. 2001) suggests that our
model provides an accurate assessment of panda habitat. In addition, the output of the
Landscape model offers a high level of spatial resolution (30x30 m), and allows more
detailed spatial analyses to be performed throughout Wolong (Figure 6.5). Results
suggest that low probabilities throughout the center of Wolong are due, in part, to their
proximity to the road system and human dwellings (Figure 6.5). This impact from
humans is also evidenced by the decrease in probability of use in areas with little forest
cover (e.g. 0-5 years). This information can be used by future researchers and reserve
managers working toward the protection of giant panda habitat.

Our analyses demonstrated that there are many issues concerning scaling up of
field survey~ data to the landscape level. First, information is lost as variables collected in
the field are removed in order to achieve a more parsimonious and applicable model.
Second, the results from a reduced model need carefiIl examination to determine if it
continues to output information usefiil for its purpose. Third, significant variables may
be replaced by other covarying, significant variables as input data into a landscape model
due to the limitation of landscape level spatial data. This is more likely to occur in areas
with few spatial datasets available. Although this may result in reduced model
performance. it also allows researchers to identify data that are necessary for more

accurate analyses. thereby guiding future research.

I43

CHAPTER 7

SUMMARY AND CONCLUSIONS

144

CHAPTER 7

SUMMARY AND CONCLUSIONS

Forest harvesting has dramatically reduced the total amount of forestland around
the world, especially in China. Yet despite the risk this poses to biodiversity and wildlife.
impacts from forest harvesting remain largely unexplored. To understand how forest
harvesting affects use by wildlife, I examined giant panda (Ailuropoda melanoleuca) use
in Wolong Nature Reserve, Sichuan Province, southwest China.

I monitored giant panda use by observing the presence/absence of panda feces.
Adaptive cluster sampling was used in order to concentrate my research efforts in areas
with a greater likelihood of panda activity. A total of 91 3 plots (443 primary and 470
secondary; total 443 neighborhoods) were sampled in 2001-2003. Due to the binary
nature of my response variable, I used logistic and autologistic regression to determine
whether harvest age, size, forest and geographical characteristics, and various bamboo
characteristics were significant in identifying presence of feces. 1 coupled this with other
analyses, including regression tree analysis, to validate and enhance my interpretation of
results. Analyses were performed on all primary plots as well as the Horvitz-Thompson
estimator of neighborhood plots to compare results of adaptive cluster analysis.

The analytical methods developed in this dissertation can be applied to other
studies that are implementing field investigations of rare and endangered species. I
provide a useful characterization of the differences and similarities found in using simple
random sampling of field plots versus using an adaptive cluster sample with a

neighborhood of 1. 5 or 8 total plots for several statistical methods. Overall, the adaptive

145

cluster sampling was more resource consuming because of the additional samples
necessary to survey the entire neighborhood. Limiting the neighborhood to 5 or 8 plots
helped to keep down costs. However, the additional costs of adding the adaptive cluster
samples must be weighed against the costs of getting the original primary random
sampling plot. In Wolong, where accessibility to the primary random sampling plot can
be difficult and the initial costs can be high, adding the adaptive cluster samples did not
substantially increase the overall costs. In this regard, the adaptive cluster sampling was
useful. Although I did not analyze the differences in variance from each of the models,
results of logistic regression models were systematically more complex when using
neighborhoods as opposed to the logistic regression of primary plots only.

I found significantly less giant panda activity in areas recently harvested. Newly
harvested areas (0-10 years) had the lowest percentages of feces presence (9.9%), while
unharvested forest had the highest (54.1%). Old harvested areas (31-100 years) (48.1%)
had similar amounts of feces presence to unharvested areas. Recent harvests that were
small (0-10 years, <10 ha) had some use (15.0%) while large, recent harvests (0-10 years,
10-100 ha) were not used (0.0%). Large, old harvest areas (10-100 ha, 31-100 years) had
the highest proportion of feces (56.1%), even greater than those in unharvested areas.
Age was an important factor in harvests that exceeded 10.0 ha, suggesting that the larger
harvests require more time to recover.

Logistic and autologistic regression along with classification trees on forest
characteristics suggests an importance in bamboo cover, overstory and midstory

composition, slope and distance to human activity. These results support previous

146

research findings while also providing a useful recharacterization of panda habitat for
several time periods after it has been harvested.

1 also conducted several analyses to determine how forest harvesting influenced
bamboo populations. 1 studied three bamboo species, including Bashaniafabri, Fargesia
robusta, and Yushania chungii. A total of 367 plots (30 m2 each) and 1468 bamboo
subplots (1 m2 each) were sampled and I found that harvesting affected various species in
different ways. For all bamboo species, harvesting at the small scale (i.e., fiielwood
collection) or large scale (i.e., timber harvesting) did not have an effect on bamboo
characteristics. However, the age of forest (i.e., years after harvesting) where each
bamboo species occurred was significantly different. B. fabri occurred in the oldest
stands and frequently in unharvested, old-growth forest while F. robusta and Y. chungii
were also in older forest and areas with no bamboo were in younger forest.

B. fabri occurred in older forest further from the Wolong roadway (2342.4 : 40.6
m) and at higher elevations (2478-3214 MSL) than younger Fargesia stands, which were
found in lower elevation forests and closer to the road (1771-2726 MSL; 1324.7 j: 52.4
m. respectively). B. fabri was rarely found in forests younger than 30 years old, but in
these forests it generally had lower percent cover and stem density, as well as thicker
stems. In F. robusta stands, percent cover and stem density generally increased over
time. F. robusta responded to harvesting with an initial colonization and subsequent
increase in stem density as competition with trees for light and space resources increased.
This study is also the first to document how another bamboo species, Yushania chungii,
which occurs naturally in the forests of Wolong, is affected by harvesting. Y. chungii was

not as common as B. fabri or F. robusta, and was typically interspersed among stands of

147

B. fabri. However, it may be a significant species for pandas because it was found in
both very young forests as well as old-growth stands, and due to its high use by pandas, it
could play a key role in panda survival when other species of bamboo mass-flower and
die off.

In addition to altering the growth characteristics of various bamboo species, forest
harvesting also can influence the habitats of the species that rely on that bamboo. In the
following chapter, I investigated how pandas are using bamboo patches and how the
relationship between pandas and bamboo changed after a harvesting event, how long it
may take for changes in habitat to recover from harvesting, and what policy changes
would be best for conservation of the panda’s bamboo populations. 1 found feces more
frequently in higher elevation Y. chungii and B. fabri stands, and less frequently in lower
elevation F. robusta. Giant pandas did not use recently logged (0-30 years) forests that
contained B. fabri and F. robusta. However, forests did not need to be in an old growth
stage to be used. Pandas were using areas that had been harvested 31-100 years prior as
much as old growth forest, indicating the ability of these forests to regenerate into
suitable habitat. More specifically, our data indicate that forests approximately 60 years
in age were frequently used. This amount of time may therefore be necessary for forests
to regenerate to a stage where bamboo grows at a level where giant pandas can forage.

While forest harvesting may affect giant panda habitats for several decades after
the initial logging event, some effort is being made to reforest certain regions of Wolong
by means of plantation reforestation. Therefore, 1 characterized the potential of these
reforested areas to provide or mitigate human impacts on giant panda habitat. Comparing

use in reforested areas to use in nearby natural forest, I found that pandas rarely used

148

artificially regenerated forest and did so only when the plantation forest had a high
percent cover of bamboo and occurred at a far distance from roadways and human
disturbance. However, although reforestation areas may not necessarily provide suitable
habitat, they do offer a woody biomass source that can be used as an alternative
fuelwood. By using reforested areas as their main fuelwood source, local residents of
Wolong Nature Reserve may limit their harvesting activities in the nearby natural forests
and impacts on giant pandas can then be reduced.

Finally, I used my field data to quantify the amount of information that is lost
during the scaling up process. An increasing number of ecological models have used plot
data collected in the field to predict species occurrence across landscapes. Despite the
popularity of this method, there are problems associated with the process of scaling-up
that have been largely ignored in the literature. Issues with the scaling-up method stem
from the removal of key variables as researchers move from many variables collected in
the field to fewer variables that are used as input into a landscape model. To demonstrate
the loss of information and the potential pitfalls of scaling-up field data to the landscape
level, I did a case study on the occurrence of pandas in Wolong. I collected field data on
the presence/absence of giant panda feces and ran logistic regression models to determine
how results responded to the removal of particular variables.

I found that a model including all significant field variables (N=10) successfully
distinguished between areas with/without panda feces. However, as the model was
reduced to fewer variables (N=4), the ability to distinguish areas with feces present and
absent was reduced. More important than the number of variables is the loss of a few key

ecologically important variables that may significantly help to predict species occurrence.

149

The analysis indicates that care should be taken at each stage of scaling-up and that it is
vital for researchers to understand the nature of this process when extrapolating their plot
data to the landscape level.

Results of this dissertation suggest that forest harvesting activities, such as timber
harvesting and fuelwood collection, can have significant impacts on many aspects of the
forest, especially bamboo populations, for several decades after harvesting. During this
period, panda use may be reduced. However, as the forest regenerates, panda habitat will
most likely recover, in some cases after a period of at least 30 years. In addition, the
reforestation activities currently being pursued by the central, provincial, and local
governments may eventually provide an easily accessible alternative fuelwood source
that can further reduce the impacts on natural forests and allow many areas currently
unsuitable for giant pandas to reinitiate into a forest stand that can be used by pandas.
This suggests that if forestland is maintained via the Natural Forest Conservation
Program (China’s “logging ban”), habitats of the giant panda that were previously
destroyed and made less suitable through forest harvesting may eventually regenerate and

increase the potential for the species’ long-term survival.

150

APPENDICES

151

 

 

 

 

 

 

 

 

 

00.0. 00 .0 00 .0 00 .0 0:. 00.0 00.0 00 .v 00.0 00.0 mm
2.8 8. 3. 8.8 2.8 8.8 8.8 8.8 8.8 8.8 2.8 88.2 .63 .88: 880 8882
00.0 00.0 00.00 00.» 0.0 00... 00.0 0:. 00 .0 00.0 mm
8.8 8.2. 8.8 8.8 8.00 8.8 8.8 2.3 8.2. 5.8 88.2 00 8.60 88880 8.8
00.0 000 00.00 00 .0 00... 00.0 00 .0 0:. 00.0 00.0 mm
8.8 2.8 :8 8.8 8.2 8.8 8.8 00.8 08.8 2.8 88.2 00 8.60 8880.
00.0 00.0 00.0 0.00 00.0 00 .0 00.0 00 .0 00.0 00.0 mm
8.8 8.8 8.8 2.8 2.2 00.8 20.8 8.8 2.8 00.2 88.). 00 880 880 8.8
V0 .0 00.0 00 .0 00 .0 0Y0 v0 .0 00 .0 00 .0 00 .0 00 .0 mm
00.0 8.4 8.0 2.0 8.2 008 8... 5.0 2.0 8.0 88.2 05 88: 880 8883..
00.0 00.0 00.0 0.0 00.0 00 .0 v0.0 00 .0 00 .0 00.0 mm
8.2 8.2 8.8 8.0 8.8 03. 8.: 0m. 2 00.0 8.8 882 .8 088: 8.8.5. 888}.
00.0 00 .0 00 .0 0v.0 00 .0 00.0 00.0 00 .0 00 .0 V0 .0 mm
8.2 8.8 008 8.2 8.2 8.4 8.2 8.2 80.2 008 88.2 3.. 880 88.6.5. ..8880
00.0 00.0 00.». 00.0 00.0 00.0 00.0 00.0 00.0 0 .0 mm
8.4 2.8 8. E 0.8 60.8 8.0 8.0 Fan 2;. 8.2 86.2 3.. 880 886.5. 80.80
0:. 3.0 00.0 3.0 00.0 00.0 00.0 000 0.0 00.0 mm
3.8 2.8 :2 8.8 8.0 8.0 8.8 8.8 8.2 8.0 88.2 3.. 880 88.6.5. .98
00.0 00.0 00.0 00.0 3..» 2.0 00.0 00.0 00.0 00.0 mm
8. 8. 8.8 3.0 2.2 .3. 2.0 2.8 08.8 8. 2 8.0 88.2 E. 088: .8886 8883..
00.0 00.0 00.0 00.0 2.0 00.0 00.0 00.0 00.0 00.0 mm
20.2 8.2 00.2 8.2 00.0 0.3 04.2 8.3 8.0 8.2 88.2 .00 8.60 8885 888.80
00.0 00.». 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 mm
8.8 8.8 8.2 8.2 8.0 2.8 8.8 8.8. 8.2 8.0 88.2 .00 880 8885 8:80
00.0 00 .0 00 .0 00.0 00.0 00.0 00.0 00.0 00.0 2.0 mm.
20.8 5.2. 00.2 8.8 8.0 8.0 5.2. 8.8 8.8 8.0 88.2 3.. 8.60 888.6 .28
00 .0 00 .0 00 .0 00 .0 00.0 .0 .0 00 .0 00. .0 00 .0 00.0 mm
8.0 8.8 8.8 2.0 o: 82 3.2 3% 8.0 3; 885. sea. .88
00.0 00 .0 00.0 00.0 00 .0 0‘0 00 .0 00 .0 00.0 3.0 mm
08.0 8.2 8.0 8.8 2.2 8.2 8.2 00.8 00.0 0... 2 863. .682 8.8.80,
2.0 00.0 00.00 00.0 00.0 00.0 00.0 00.0 00.0 00.0 mm
00.2 8.2 8.8 8.8 2.8 8.2 8.8 04.2 5.8 8.2 88.2 88883 88.0
00 .00 00 .0». 00 .00 0 v.00 0 0 .0» 00 .0» 00 .00 00.00 00 .00 00 .00 mm
2.88 8.28 5.028 8.88 8.028 8.028 8.88 8.28 8.88 8.88 88.2 :22. 8.88.0
00 .0 00 .0 00 .0 00 .0 v0 .0 00 .0 00 .0 00 .0 00 .0 00 .0 mm
8.8 a: 8. 8 8.8 2.8 .88 8.0 2.0 8.0 8.0 882 a... 8.0 8.8.
00.0 00 .0 00 .0 V0 .0 00 .0 00.0 00 .0 0‘0 00 .0 00.0 mm
8.8 8.3. :8 8.8 88 8.8 8.82 2.8.. 8.8 0.8 88.2 8880 8<
8.0 2.0 o 8.0 o 8.0 8.0 8.0 00.0 8.0 88". 5.; 8.88.0
8 3. m 40 2. 8 02 00 8 02 .80 8.0 08800 z
2 02-2 9. 2A 9. 02-2 6,. 2A 2. 02-2 9. 2A £890.20 28.. 02-5 .880 8-: 880 2-0
0.00.» 00790 0.00.. 00-: 9.00» 00.0 .300.

 

 

.220 00 3020202000 20020000 000 80000 00 Ammflv 80:3 000:. 000 80000 53» 00000080 .220 20:08.00 80:52 ._ £00000<

152

20.» 8.00 02.0 02.2 02.0 8.22 8.02 8.8 00.0 8.8 :8. 30.812.68.85 80832

8 .0 20.2 8 .0 2. 8 00 .8 00.20 00 .0 8.8 20.0 2.2 0.2.. 8.60 2.68.800 .98
00.0 00.2. 00 .0 00.2. 00 .0 00.2. 02 .0 2.0... 2 .0 00.2. 222.. .88... 8.2.0 80832
00.2 00. 8 8.0 8.8 00 .2 8.2 00.0 2.8 8.2 00.8 0.2.. 8.60 8.2.0 .98
00 .0 20.2 02.2 00.2 00 .0 8... 00.0 00.2 00 .0 8.02 0... .88... 2.28.5. 8883..
2.02 22.2 80.2. 00.02 00 .0 00.0 2 .0 2.0 8.2 02.2 .02.. 8.60 2.98.... 880.80
20.2 00.2 2 .0 00.0 00 .0 00.0 8.2 8.2 8 .0 22.0 0.2.. 8.60 2.98.5. 8.80
22 .2 00.00 00 .8 00.00 22 .00 00.00 20.0 02.02 02 .2 00.00 202.. 8.60 2.98.5. .98
00.2 :8 2.2 .0 00.00 8.0 00.2 00 .0 2.00 20.2 00.8 0... .88.. 2.6885 8082
02 .2 28.2 8.0 2.2 02 .0 00.0 8.2 8.2 82 .2 20.0 0.2.. 8.60 2888.6 868.80
02.0 00.0.. 8.0 22.00 8 .2 00.00 02 .0 8.22 2 2 .0 00.20 .02.. 8.60 2888.6 8080
00 .0 20.8 8 .0 22.8 20.0 00.00 20.0 22.2 8 .0 8.8 0.2.. 8.60 2.6885 .98
20.0 00.00 8.0 00. 22 00.0 00.02 02.0 00. 2.. 2 .0 2.0.8 0.2.. 8.60 8880
2.02 00.02 8.2 22. 22 20.0 20.0 02 .0 8.0 00 .0 8.02 29.0.... 82 .080
00 .0 02.02 02 .2 00.0 22 .0 8.02 20.2 2 2.02 00 .0 20.02 .082 8.88. .880. .8880
8.2 8.2 00.2 8.2 82 20.2.0 00 .0 02.8 22 .2 00.2 88.08. 88.0
20.0 00.02 00 .0 00.8 00.2 00.2 00.0 8.2 20.0 20.22 8.982.. 8.0 88.0
02 .0 00.02 00 .0 20.00 00.2. 00.2 00 .0 20.0 00 .2 20.002 8.82. 8< .88..
8.00 00.0000 8.20 22.222 00.002 8.222 00.00 00.082 28.20 00.88 8828.. 88: o. 8.88.0
8.00 02.008 8.202 00.82 00.022 8.802 02.00 02.202 22.22 20.88 888.... 20.52.80 2 888.0
00.20 00.008 20.8 00.028 20.00 00.0200 02 .20 00. 208 20.2 02.800 3.0.2. 8.8.8.0
2.0.00 8.8 0 0 2 2.00 8... 88. 2...; .0880
20 00 2 2 22 008880 + 2.8.8. 2

mm 2.522 mm 0002 mm cmmE mm .522 mm 2.505—
539005 0.00.» 002-00 0.00.2 00-22 0.00.» 02-0 .068 0002 0.2.2.000

.C200\.0..2202200m. 80:202. 30:0

200 00.20.202002000 200020000 0:0 000.0,. 00 fimH. 02:0.» :08: 0:0 00000 22.5 :o_:o0o.0 .0220 2080000 2002252 .<m r._0:000<

153

EM.
mad
mad
3N
and
mmd
Ed
mmN
mm.“
MW».
«66
2.»
mad
No.“
S.»
mug.
med
3d

©9an Nodmmw umdwm demt
mmdmh 5999
86% mméwmw 00.9» omNNVN mvgm mwdmmm mmdm mthNN 09:.

mm .vt. modem?

mm

mme
EKN
aim
9%?
5.3
NmKN
mm...
mrdm
omdw
mo.mN
Fmdv
wmdm
wmdm
vm.»
mm. :
voNN
mod
mime

9.3
mm
cams.

£506.20

.36
mvd
.310
mvfi
E d
mmfi
om.“
mad
E .s
E .m
mm.m
Nwé
vmd
mm.»
om.“
mug.
mNN
39m

mm

medm
ome
om.»V
31mm
om.N—.
ooém
mmé
mnhm
Nde
3.9.
mvdr
omNN
omév
Fwd
mmd
oodw
39m
Fmdm

mm

t
:85.

m6?» ooTFm

mmfi
cod
“Md
9 .m.
mv.»
NW6.
“ms
mud
Veg.
mmd
mad
mud
26
nmd
Ed
nos
2..“
ms.“

moémw vodmmv
mmbm

mm

3.8 t .m mmdc
3.: mmé 5mm
mod 2 .o wad
mmdw woe. omdm

8.? odd $6
3.8 cm.“ a:
5.? vmd mm;
mmdm mm.» 36
mud; ox.“ wmd
8.9 2 .5 mad
9: mm.“ Nwé
mo. K. N“ N 8.5
Sdm Efi 3.3.
$6 $6 mm;
mud mad 3.?
ooém mm; mndm
mod mod 00?
mmdu 2‘6 no.0
Exams 3&3.
358 Sin vmsmm
no.5 o
vm 3.
:85. mm :35.
98> 8-: 2mm; 96

ncN
cmN
36
mm.“
wed
was
3.6
km;
00.“
mm.»
0V.»
um.“
EN
mvd
Nvd
mad
0»;
mad

2 .mm mode:
noxo 36mg.
{.mNmN

mm

Fmdm
mmdm
and
omdm
vmdw
36m
9....
mm. FN
$03.
om.m_.
cad
tdm
Nvdv
rod
5::
mmém
NmN
mmdm

5.3
Fm?
cams.

.98.

A53

3;

E

3.;

E;

g

3;

3;

E;

3,;

A5

St

E
Améé
a8. .oxaav
A3933
$9285
$593
EQmEV
A9295
:9):
3;

£90: b06895 3994.
550 b05595 .98.
292.. 925m 3903.
.900 £55 .muoh

Ego: b07435. ommB><
530 >562: 9.63280
.050 >532: .350
.050 >562: .90...
:56: E2990 mom.m><
B>oo b230>0 2532000
850 b07530 52:00
850 bo~m8>0 .92.
.960 0863

mm? .88

882 .8628”...

32w

mum 52mm

mm< $96".

$301 9 8535
>m>>umom 9 8.535
cozm>m_m

wmofi 53> Ememn.
Emucoomw + meE: 2

$259 Smmmgml

.AEmEE 3.3335 8263 2358:

S; mosmtuuofimzu 3:9:wa was $28 :8 $55 mus—g :38 new “moon; £3» :oanoE .mBE 292318 59:32 .mm ficcuaalx

154

155

8.... 8... 9.8. 8.8 .88 8.8 .88. 88. 8.. 8.8 89 299.. 82998.5 8899;.
8... .88 8.... 8.8 ...v 8... ...v 8... 8... 88. $9 .960 8298.5 :98
88 m... 88 88 88 8.... 8. 8 8.... 88 88 85 .88: 828 88983.
88 888 3.8 8... .88 88. 8.8. 8.8 88 .8... 8.... .960 858 as.
8.. 88. 88 88. 88 8.8 8. .v 8... .8. 8... .8 .86: 82.0.82 899;.
88 8.8 8. .v 8.8 .m... 88. .N 8.. 8.. 88 8... 8.... 960 .8988 88280
8.. 88 88 88.8 88 88 8.. m... 8.8 88 39 .960 .8283 8.80
88 .88 8. 8 8.8 .8.» mm. .8 .88 mm... 8.8 8. .N 8.9 .960 82.0.2.2 as.
8.. 8.8 8.. 88. V. .m 88. 8. .. 888. 8.8 .88. 85 .86: 82996 8899;.
88 88 88... 8.8 88 mm. .8 88 88.8 8... 88. 39 .960 82996 88888
8. 8 8.8 88 88 88 88 .8... 88 .. ..V 8.8. 8.... 960 8.996 8.80
8. 8 .888 8... 8. .8 .88 mm. .8 .8... 88 .8... 8.8 8.9 .960 82996 .98.
8.. 8.8 8.... 88.. .88. 8.8 8.. 8.8 88 8.8 8.9 .98 88588
88 8.8. 8.. 88 88 88.. 8... 8.. 8.. 88 .288 89.. 888
8.8 .... 8.8 88. .. 8.. 8.... 88... 8... 8.. 88. .98. .959 .882 8282on
88 88. .88 8.8 .8. 8.. .N m. .w 8.8 8.. 88. .8983 988
88 ...N .88 8.8 .. 8 88 88 88 8.. .88 @9985 $8 :28“.
.. .8 888 88 88.. 88 8.8 8.. 88 8... 8.8 .9898 88,. 898“.
.98. .88.... 8.8. 8.88.. 88. 8.88 .888 8.8.. 8.8. 883.. .9988 895: 2 8:998
8 8.. .888. 8.8 ....... «8.8 .888 8.8. .888 8.8. 8.8.. .9988 .8288”. 9 8:998
8.8 8. .8. 8. ..8. 8888 m. ..8 8.88 88.. 888... 8.8 88.... :82. 88998
8.8 8 8 88. 8.8 8.... 889 5;. .899.
m m m w m. Emccoomm + 866:8 2

mm cmmE Wm. cam—2 Wm. ram—2 mm cam—>— mm. :92)—
£265.20 98> co .-.m 9.88.» cm... 988.» 3-0 Eek 89.2.8 258%.;

Atatzfi 8.2.38.3V 009889 2:28.;

.0. 82.95.88.828 oEquOuw 8:8 .888 ..o 35.3 32.; :82: 8:8 888.. .23 82:088.. 8.28 29:8 .0 53:52 .Um £2.23...

00.8 :00
E .2 8.00
00.0 00.0
00.0 00.8
3.0 {.0
$0 $0.
2 .0 3..
00.0 :0
k .0 :0
2 .0 00.2
v0.2 00.00
3.2 $.00
3: m).
.3 {.0
00.0 2.0
00.0 030
03 3.0
00.: 5.02
00.000 :28
0.0: 8.3;
o
N.
Mm. .505.
539920

V50
0V6
nmd
vm.»
hm...
8N
NF .N
2.1m
..v...
mmd
m. .m
5.».
30
mm...
mm.»
NQN
kmNN
mm...

mm .mm

mm

v... 5
mmhh
afim
mndm
mos
3K
mmd
mud—
3.5..
:30
mném
36m
NF.
mod
36
Nmér
m—dw
mmsm

©VS©V wmévow
9.600
31.5.. mwdmmw NV; mm.momm

o

NN
000.2

0000> 03.5

5.0.
09v
mud
k9».
Ed
mm.“
mm.»
SYN
“N...
m»;
vmé
m9».
MEL
mm.“
mad
om.»
R...“
and
56k
@060
mva

mm

mm...»
vm.mo
mvN
wmdm
vmd
mad
omN
mv. I.
mfov
had
Nm.mN
mmdm
EC
mmN
mwd
QEFN
50.0
wrNN

wmdnm
vmdmm
modem

0

mm
000.).

0000.» 8-:

KM.
n06.
omd
mmN
nvd
ovd
um...
um...
sud
vmd
m» .N
mmN
m};
36
end
NF.“
N50
omd
vam
mmdn

m0

omdm
..vdn
woé
vod
Div
mad
.VmN
an...”
mmN
3.0
006
mod
NE
9.;
mm.o_.
nmdp
vaw
00...

mwémm
Nvfiom
mmdm wwdo ..N

0

mm
000.).

0000.. 9.0

EN
EN
2.0
um...
mvd
med
00.“
mm.“
mud
and
E. .N
NNN
m\:
«6.0
wvd
be.»
mnw
mvd

N93
00.;
05?...

mm

dev
ofmo
hm...
mmdw
mm...”
36
and
..NN
va
NmN
00.5.
dev
NE
0N4V
wed
mmfiw
Nw.mm
NNNN

mm.wmw
006mm
Edna

o
03
000.2

.90...

.50.
.0...
.5.
so.
.5.
.0...
3°.
.0...
:5
3...
3°.
3...
$0.

.205.
@000 09:00.
300503
0000.005
.9003
0000.00:
$00.00..

30.2.
3..

.090: 00.0.0003 00000><
00>00 00.00000: .80...
.090... 0200 0mm00><
.050 0200 .000...

.090... 00.00:). 0mm..0><
.050 00.00:). 0:30.000
.050 00.00:). 000.000
.030 00.02.). .08..
0.90... 020.90 090024
.050 0000090 03032000
.030 00.990 000.000
050 00.990 .000...
.050 000.08

002 Emmm

00000.0. .00_m0.00w

000.0

0N_m 00.0.“.

0m< .0000“.

00:0... 0. 0000.05
0.02603. 0. 0000.05
00_.0>0_m.

0000.. 5.2, 0000000
000000000 + 00.000. 2

0000000 02

000800 00 5.3 00000

00. 0000000000000 0.0000w00w .000 00000.. 00 Ammi 30.? 0008 000 0000.. 5...: 000.0003 .307. 0.900000 000502 .QN 0.000%?

l56

LITERATURE CITED

Allan, D. G., J. A. Harrison, R. A. Navarro, B. W. vanWilgen, and M. W. Thompson.
1997. The impact of commercial afforestation on bird populations in Mpumalanga

Province, South Africa - Insights from bird-atlas data. Biological Conservation
79:173-185.

An, I... J. Liu, Z. Ouyang, M. Linderman, S. Zhou, and H. Zhang. 2001. Simulating
demographic and socioeconomic processes on household level and implications
for giant panda habitats. Ecological Modelling 140:31-49.

An, L., F. Lupi. J. G. Liu. M. A. Linderman, and J. Y. Huang. 2002. Modeling the choice
to switch from fuelwood to electricity Implications for giant panda habitat
conservation. Ecological Economics 42:445-457.

An. L., A. G. Mertig, and J. G. Liu. 2003. Adolescents leaving parental home:

Psychosocial correlates and implications for conservation. Population and
Environment 24:415-444.

Anderson, S. H., and B. J. Crompton. 2002. The effects of shelterwood logging on bird
community composition in the Black Hills, Wyoming. Forest Science 48:365-372.

Armstrong, A. J., G. Benn, A. E. Bowland, P. S. Goodman, D. N. Johnson, A. H.
Maddock. and C. R. Scott-Shaw. I998. Plantation forestry in South Africa and its
impact on biodiversity. South African Forestry Journal 182:59-65.

Arnold, M., G. Kohlin, R. Persson, and G. Shepherd. 2003. F uelwood revisited: What
has changed in the last decade? Center for International Forestry Research,
Jakarta, Indonesia.

Aust, W. M., and R. Lea. 1991. Soil-Temperature and Organic-Matter in a Disturbed
Forested Wetland. Soil Science Society of America Journal 55:1741-1746.

Caldecott. J. 0.. M. D. Jenkins. T. H. Johnson, and B. Groombridge. 1996. Priorities for
conserving global species richness and endemism. Biodiversity and Conservation
52699-727.

C ao, K. I995. Fagus dominance in Chinese montane forests: natural regeneration of
Fagus lucida and F agus hayatae var. pashanica. Wageningen Agricultural
University, Wageningen, The Netherlands.

Carey. A. B., and M. L. Johnson. 1995. Small Mammals in Managed, Naturally Young,
and Old-Growth Forests. Ecological Applications 5:336-352.

157

Carmosini, N., K. J. Devito, and E. E. Prepas. 2003. Net nitrogen mineralization and
nitrification in trembling aspen forest soils on the Boreal Plain. Canadian Journal
of Forest Research-Revue Canadienne De Recherche Forestiere 33:2262-2268.

Chamberlain, M. J ., L. M. Conner, B. D. Leopold, and K. M. Hodges. 2003. Space use
and multi-scale habitat selection of adult raccoons in central Mississippi. Journal
of Wildlife Management 67:334-340.

China State Forestry Administration. 2004. Steadily increasing giant panda populations:
results from the third giant panda
census.:http://www.forestry/gov.cn/DB/news/content.asp?table type=news&id=l
4784&pgid= l 4781.

China-Daily. 2004. Fighting illegal logging. Pages
http://www.china.org.cn/english/govemment/99832.htm in China-Daily, Beijing.

China's Ministry of Forestry and World Wildlife Fund. 1989. Conservation and
management plan for giant pandas and their habitat. Chinese MOF and WWF..
Beijing, PR China.

Choudhury, A. 2001. An overview of the status and conservation of the red panda
Ailurus fulgens in India, with reference to its global status. Oryx 35:250-259.

Cochran, W. G. 2002. Sampling Techniques. Third edition. John Wiley and Sons, New
York.

Cole, E. C.. W. C. McComb, M. Newton, J. P. Leeming, and C. L. Chambers. 1998.
Response of small mammals to clearcutting, burning, and glyphosate application
in the Oregon Coast Range. Journal of Wildlife Management 62: 1207-1216.

Convention on Biological Diversity. 200]. The value of forest ecosystems. CBD
Technical Series No. 4, Secretariat of the Convention on Biological Diversity,
Montreal, Quebec.

Costanza, R., R. d'Arge, R. de Groot, S. Farber, M. Grasso, B. Hannon, K. Limburg, S.
Naeem, R. V. O'Neill, J. Paruelo, R. G. Raskin, P. Sutton, and M. van den Belt.

I997. The value of the world's ecosystem services and natural capital. Nature
387 :253-260.

C ramer, P. C .. and K. M. Portier. 2001. Modeling Florida panther movements in response
to human attributes of the landscape and ecological settings. Ecological
Modelling 140251-80.

Danks, F. S., and D. R. Klein. 2002. Using GIS to predict potential wildlife habitat: a

case study of muskoxen in northern Alaska. International Journal of Remote
Sensing 23:4611-4632.

l58

De Grandpre, L., L. Archambault, and J. Morissette. 2000. Early understory successional

changes following clearcutting in the balsam fir-yellow birch forest. Ecoscience
7 :92-100.

De Grandpre, L., and Y. Bergeron. 1997. Diversity and stability of understorey
communities following disturbance in the southern boreal forest. Journal of
Ecology 85:777-784.

De'ath, G., and K. E. Fabricius. 2000. Classification and regression trees: A powerful yet
simple technique for ecological data analysis. Ecology 81:3178-3192.

deBellefeuille, S., L. Belanger, J. Huot, and A. Cimon. 2001. Clear-cutting and
regeneration practices in Quebec boreal balsam fir forest: effects on snowshoe
hare. Canadian Journal of Forest Research 31:41-51.

DeWulf, R. R., R. E. Goossens, J. R. MacKinnon, and S. C. Wu. 1988. Remote sensing

for wildlife management: giant panda habitat mapping from LANDSAT MSS
images. Geocarto International 1:41-50.

Dryver, A. L., and S. K. Thompson. 2005. Improved unbiased estimators in adaptive
cluster sampling. Journal of the Royal Statistical Society Series B- Statistical
Methodology 67:157-166.

Duffy, D. C., and A. J. Meier. 1992. Do Appalachian Herbaceous Understories Ever
Recover from Clearcutting. Conservation Biology 6:196-201.

Duguay, J. P., and P. B. Wood. 2002. Salamander abundance in regenerating forest

stands on the Monongahela National Forest, West Virginia. Forest Science
48:331-335.

Duguay, J. P., P. B. Wood, and J. V. Nichols. 2001. Songbird abundance and avian nest
survival rates in forests fragmented by different silvicultural treatments.
Conservation Biology 15:1405-1415.

Durst, P. B., T. R. Waggener, T. Enters, and T. L. Cheng, editors. 2001. Forests out of
bounds: Impacts and effectiveness of logging bans in natural forests in Asia-
Pacific. Food and Agricultural Organization of the United Nations, Regional
Office for Asia and the Pacific.

Emlen, J. M., D. C. Freeman, M. D. Kirchhoff, C. L. Alados, J. Escos, and J. J. Duda.
2003. Fitting population models from field data. Ecological Modelling 162:] I9-
I43.

Ericsson, S., L. Ostlund, and A. L. Axelsson. 2000. A forest of grazing and logging:

Deforestation and reforestation history of a boreal landscape in central Sweden.
New Forests 19:227-240.

159

FAO. 2000. Global Forest Resources Assessment (FRA 2000). Food and Agriculture
Organization of the United Nations, Global Forest Resources Assessment
Programme,, Rome.

FAO. 2001. Forests out of bounds: Impacts and effectiveness of logging bans in natural
forests in Asia-Pacific. Food and Agriculture Organization of the United Nations,
Asia-Pacific Forestry Commission,, Rome.

FAO. 2003. State of the world's forests: Part I- The situation and developments in the
forest sector. Food and Agriculture Organization of the United Nations, Forestry
Department,, Rome.

FAO. 2005. State of the World's Forests. Food and Agriculture Organization, United
Nations, Rome.

Fertig, W., and W. Reiners. 2002. Predicting presence/absence of plant species for range
mapping: A case study from Wyoming. Pages 483-490 in M. Morrison, editor.
Predicting species occurrences: Issues of accuracy and scale. Island Press,
Washington, DC.

Fidgett, A. L., A. T. C. F eistner, and H. Galbraith. 1996. Dietary intake, food
composition and nutrient intake in captive Alaotran gentle lemurs Hapalemur
griseus alaotrensis. Dodo-Joumal of the Wildlife Preservation Trusts 32:44-62.

Fleishman, E., R. Mac Nally, J. P. Fay, and D. D. Murphy. 2001. Modeling and
predicting species occurrence using broad-scale environmental variables: an
example with butterflies of the Great Basin. Conservation Biology 15:1674-1685.

Forman, R. I995. Land mosaics : the ecology of landscapes and regions. Cambridge
University Press, New York.

Franklin, S. E., D. R. Peddle, J. A. Dechka, and G. B. Stenhouse. 2002. Evidential
reasoning with Landsat TM, DEM and GIS data for landcover classification in

support of grizzly bear habitat mapping. International Journal of Remote Sensing
23:4633-4652.

Fredericksen, T. S., B. D. Ross, W. Hoffman, M. L. Morrison, J. Beyea, B. N. Johnson,
M. B. Lester, and E. Ross. 1999. Short-term understory plant community
responses to timber-harvesting intensity on non-industrial private forestlands in
Pennsylvania. Forest Ecology and Management 116:]29-139.

Fredericksen, T. S., B. D. Ross, W. Hoffman, E. Ross, M. L. Morrison, J. Beyea, M. B.
Lester, and B. N. Johnson. 2000. The impact of logging on wildlife - A study in
northeastern Pennsylvania. Journal of Forestry 9824-10.

Frey, B. R., V. J. Lieffers, A. D. Munson, and P. V. Blenis. 2003. The influence of partial
harvesting and forest floor disturbance on nutrient availability and understory

I60

vegetation in boreal mixedwoods. Canadian Journal of Forest Research-Revue
Canadienne De Recherche Forestiere 33:1 180-] 188.

Gardner-Outlaw, T., and R. Engelman. 1999. Forest futures: Population, consumption,
and wood resources. Population Action International, Washington, DC.

Ghazoul, J. 2002. Impact of logging on the richness and diversity of forest butterflies in a
tropical dry forest in Thailand. Biodiversity and Conservation 11:521-541.

Gilmour, D. 1999. Rehabilitation of degreaded forest ecosystems in Vietnam, Lao PDR
and Cambodia. Regional Overview. IUCN-The World Conservation Union,
Gland, Switzerland.

Goodale, C. L., and J. D. Aber. 2001. The long-term effects of land-use history on
nitrogen cycling in northern hardwood forests. Ecological Applications 11:253-
267.

Goodale, C. L., J. D. Aber, and W. H. McDowell. 2000. The long-term effects of
disturbance on organic and inorganic nitrogen export in the White Mountains,
New Hampshire. Ecosystems 32433-450.

Gratzer, G., P. B. Rai, and G. Glatzel. I999. The influence of the bamboo Yushania
microphylla on regeneration of Abies densa in central Bhutan. Canadian Journal
of Forest Research-Revue Canadienne De Recherche F orestiere 29:1518-1527.

Green. R. J ., and C. P. Catterall. 1998. The effects of forest clearing and regeneration on
the fauna of Wivenhoe Park, south-east Queensland. Wildlife Research 25:677-
690.

Gross, J. E., M. C. Kneeland, D. F. Reed, and R. M. Reich. 2002. GIS-based habitat
models for mountain goats. Journal of Mammalogy 83:218-228.

Grossman, B. C., M. A. Gold, and D. C. Dey. 2003. Restoration of hard mast species for
wildlife in Missouri using precocious flowering oak in the Missouri River
floodplain, USA. Agroforestry Systems 59:3-10.

Hamer, K. C., J. K. Hill, S. Benedick, N. Mustaffa, T. N. Sherratt, M. Maryati, and V. K.
Chey. 2003. Ecology of butterflies in natural and selectively logged forests of

northern Borneo: the importance of habitat heterogeneity. Journal of Applied
Ecology 40:150-162.

Harper, C. A., and D. C. Guynn. 1999. Factors affecting salamander density and
distribution within four forest types in the Southern Appalachian Mountains.
Forest Ecology and Management 114:245-252.

Harpole, D. N., and C. A. Haas. 1999. Effects of seven silvicultural treatments on
terrestrial salamanders. Forest Ecology and Management 114:349-356.

I61

Hartley, M. J. 2002. Rationale and methods for conserving biodiversity in plantation
forests. Forest Ecology and Management 155281-95.

Hawks, M. M., J. S. Stanovick, and M. L. Caldwell. 2000. Demonstration of GIS
capabilities for fisheries management decisions: Aanalysis of acquisition potential
within the Meramec River Basin. Environmental Management 26:25-34.

He. D., and C. Barr. 2004. China's pulp and paper sector: an analysis of supply-demand
and medium term projections. International Forestry Review 61254-266.

Hemami, M. R., A. R. Watkinson, and P. M. Dolman. 2004. Habitat selection by
sympatric muntjac (Muntiacus reevesi) and roe deer (Capreolus capreolus) in a
lowland commercial pine forest. Forest Ecology and Management 194:49-60.

Heydon, M., and P. Bulloh. 1996. The impact of selective logging on sympatric civet
species in Borneo. Oryx 30:311-316.

Heydon, M., and P. Bulloh. I997. Mousedeer densities in a tropical rainforest: The
impact of selective logging. Journal of Applied Ecology 34:484-496.

Heydon, M. J ., and J. C. Reynolds. 2000. Demography of rural foxes (Vulpes vulpes) in
relation to cull intensity in three contrasting regions of Britain. Journal of Zoology
251:265-276.

Hill. J. K. 1999. Butterfly spatial distribution and habitat requirements in a tropical forest:
impacts of selective logging. Journal of Applied Ecology 36:564-572.

Hill. J. K., K. C. Hamer, M. M. Dawood, J. Tangah, and V. K. Chey. 2003. Rainfall but
not selective logging affect changes in abundance of tropical forest butterfly in
Sabah, Borneo. Journal of Tropical Ecology 19:35-42.

Howard, E. A., S. T. Gower. J. A. Foley, and C. J. Kucharik. 2004. Effects of logging on
carbon dynamics of a jack pine forest in Saskatchewan, Canada. Global Change
Biology 10:1267-1284.

Hulbert, I. A. R., G. R. Iason, D. A. Elston, and P. A. Racey. I996a. Home-range sizes in
a stratified upland landscape of two lagomorphs with different feeding strategies.
Journal of Applied Ecology 33: 1479-1488.

Hulbert, I. A. R., G. R. Iason, and P. A. Racey. I996b. Habitat utilization in a stratified
upland landscape by two lagomorphs with different feeding strategies. Journal of
Applied Ecology 33:315-324.

Insightful Corp. 2003. S-Plus Academic Site Edition. in, Seattle, WA, USA.
Ito, H., and T. Hino. 2004. Effects of deer, mice and dwarf bamboo on the emergence,

survival and growth of Abies homolepis (Piceaceae) seedlings. Ecological
Research 19:217-223.

I62

IUCN: The World Conservation Union. 2005. Forest environmental services. IUCN:
Forest Conservation Programme, Gland, Switzerland.

Jenkins, M. 2003. Prospects for biodiversity. Science 302:1175-1 I77.

Johns, A. G., S. Eltringham, J. Harwood, D. Pimentel, A. Sinclair, and M. Sissenwine,
editors. 1997. Timber Production and Biodiversity Conservation in Tropical Rain
Forests. Cambridge University Press, Cambridge.

Johnson, K. G., W. Wang, D. G. Reid, and J. C. Hu. 1993. Food-Habits of Asiatic
Leopards (Panthera-Pardus Fusea) in Wolong Reserve, Sichuan, China. Journal of
Mammalogy 74:646-650.

Kanowski, J ., C. P. Catterall, G. W. Wardell-Johnson, H. Proctor, and T. Reis. 2003.
Development of forest structure on cleared rainforest land in eastern Australia

under different styles of reforestation. Forest Ecology and Management 183:265-
280.

Kim, C., T. L. Sharik, and M. F. Jurgensen. 1995. Canopy Cover Effects on Soil-
Nitrogen Mineralization in Northern Red Oak (Quercus-Rubra) Stands in
Northern Lower Michigan. Forest Ecology and Management 76:21-28.

King, D. l., and R. M. DeGraaf. 2004. Effects of group-selection opening size on the
distribution and reproductive success of an early-successional shrubland bird.
Forest Ecology and Management 190:179-185.

Kirkland, G. L. I990. Patterns of Initial Small Mammal Community Change after
Clearcutting of Temperate North-American Forests. Oikos 59:313-320.

Klute, D. S., M. J. Lovallo, and W. M. Tzilkowski. 2002. Autologistic regression
modeling of American woodcock habitat use with spatially dependent data. Pages
335-343 in J. M. Scott, P. J. Heglund, M. L. Morrison, J. B. Haufler, M. G.
Raphael, W. A. Wall, and F. B. Samson, editors. Predicting species occurrences:
Issues of accuracy and scale. Island Press, Washington, DC.

Knapp, S. M., C. A. Haas, D. N. Harpole, and R. L. Kirkpatrick. 2003. Initial effects of
clearcutting and alternative silvicultural practices on terrestrial salamander
abundance. Conservation Biology 17:752-762.

Kristof, N. 1993. The rise of China. Foreign Affairs 72:59-74.

Kunshan, S., L. Fengming, M. Yongqing, S. Feng, and Z. Zuofeng. 1997. China's forest
product consumption and its demand for the market of international tropical forest
products by 2010. ITTO PD 25/96 Rev. 2(M), International Tropical Timber
Organization and Chinese Academy of Forestry, Beijing.

Laporte, M. F ., L. C. Duchesne, and I. K. Morrison. 2003. Effect of clearcutting,
selection cutting, shelterwood cutting and microsites on soil surface C02 efflux in

163

a tolerant hardwood ecosystem of northern Ontario. Forest Ecology and
Management 174:565-575.

Latham, J ., B. W. Staines, and M. L. Gorman. I996. The relative densities of red (Cervus
elaphus) and roe (Capreolus capreolus) deer and their relationship in Scottish
plantation forests. Journal of Zoology 240:285-299.

Lechowicz, M. J. 1982. The sampling characteristics of electivity indices. Oecologia
52:22-30.

Leger, C.. and N. Altman. 1993. Assessing Influence in Variable Selection-Problems.
Journal of the American Statistical Association 88:547-556.

Levin, S. A. 1992. The Problem of Pattern and Scale in Ecology. Ecology 73: 1943-1967.

Li, C. B., editor. 1990. The ecological study of Sichuan forests. Sichuan Publishing
House of Science and Technology, Sichuan Academy of Forestry, Chengdu,
Sichuan, PRC.

Li, W. 2004. Degradation and restoration of forest ecosystems in China. Forest Ecology
and Management 201 :33-4 1 .

Li, Y., and D. Wilcove. 2005. Threats to vertebrate species in China and the United
States. BioScience 55:147-153.

Lindenmayer, D. B., and R. C. Lacy. 1995. Metapopulation Viability of Arboreal
Marsupials in Fragmented Old-Growth Forests - Comparison among Species.
Ecological Applications 5:183-199.

Linderman, M., L. An, S. Bearer, G. He, Z. Ouyang, and J. Liu. 2005a. Modeling the
spatio-temporal dynamics and interactions of households, landscapes, and giant
panda habitat. Ecological Modelling 183:47-65.

Linderman, M., S. Bearer, L. An, Y. Tan, Z. Ouyang, and J. Liu. 2005b. The effects of
understory bamboo on broad-scale estimates of giant panda habitat. Biological
Conservation 121:383-390.

Linderman, M., J. Liu, J. Qi, L. An, Z. Ouyang, J. Yang, and Y. Tan. 2004. Using
artificial neural networks to map the spatial distribution of understorey bamboo
from remote sensing data. International Journal of Remote Sensing 25:1685-1700.

Litvaitis, J. A. 2001. Importance of early successional habitats to mammals in eastern
forests. Wildlife Society Bulletin 29:466-473.

Litvaitis, J. A., and R. Villafuerte. I996a. Factors affecting the persistence of New

England cottontail metapopulations: The role of habitat management. Wildlife
Society Bulletin 24:686-693.

I64

Litvaitis, J. A., and R. Villafuerte. l996b. Intraguild predation, mesopredator release, and
prey stability. Conservation Biology 10:676-677.

Liu. J ., and J. Diamond. 2005. China's environment in a globalizing world. Nature
435:1179-1186.

Liu, J., M. Linderman, Z. Ouyang, L. An, and e. al. 2001. Ecological degradation in
protected areas: The case of Wolong Nature Reserve for giant pandas. Science
292:98.

Liu, J., Z. Ouyang, Y. Tan, J. Yang, and H. Zhang. 1999a. Changes in human population
structure: Implications for biodiversity conservation. Population and Environment
21:45-58.

Liu, J ., Z. Ouyang, W. W. Taylor, R. Groop, K. C. Tan, and H. M. Zhang. 199%. A
framework for evaluating the effects of human factors on wildlife habitat: the case
of giant pandas. Conservation Biology 13:1360-1370.

Liu. J .. and W. W. Taylor, editors. 2002. Integrating landscape ecology into natural
resource management. Cambridge University Press, New York, New York.

Liu, S. L., B. J. Fu, Y. H. Lu, and L. D. Chen. 2002. Effects of reforestation and
deforestation on soil properties in humid mountainous areas: a case study in

Wolong Nature Reserve, Sichuan province, China. Soil Use and Management
18:376-3 80.

Lo. N. C. H., D. Griffith, and J. R. Hunter. 1997. Using a restricted adaptive cluster
sampling to estimate Pacific hake larval abundance. California Cooperatic
Oceanic Fisheries Investigations 38:103-1 l3.

Maddock, A. H. 1988. Resource partitioning in a viverrid assemblage. Ph.D. Thesis.
University of Natal, Pietermaritsburg.

Malcolm, J. R., and J. C. Ray. 2000. Influence of timber extraction routes on central
African small-mammal communities, forest structure, and tree diversity.
Conservation Biology 14:1623-1638.

Matlock, R. B., D. Rogers, P. J. Edwards, and S. G. Martin. 2002. Avian communities in
forest fragments and reforestation areas associated with banana plantations in
Costa Rica. Agriculture Ecosystems & Environment 91:199-215.

Matsubayashi, H., E. Bosi, and S. Kohshima. 2003. Activity and habitat use of lesser
mouse-deer (Tragulus javanicus). Journal of Mammalogy 84:234-242.

Matthews. E., R. Payne, M. Rohweder, and S. Murray. 2000. Pilot analysis of global
ecosystems: Forest ecosystems. World Resources Institute, Washington, DC.

165

 

Mattson, D. J. 1997. Use of lodgepole pine cover types by Yellowstone grizzly bears.
Journal of Wildlife Management 61:480-496.

McLellan, B. N., and D. M. Shackleton. 1989. Grizzly Bears and Resource-Extraction
Industries - Habitat Displacement in Response to Seismic Exploration, Timber
Harvesting and Road Maintenance. Journal of Applied Ecology 26:371-380.

Morales, J. M., and S. P. Ellner. 2002. Scaling up animal movements in heterogeneous
landscapes: The importance of behavior. Ecology 83:2240-2247.

Nagaike, T., A. Hayashi, M. Abe, and N. Arai. 2003. Differences in plant species
diversity in Larix kaempferi plantations of different ages in central Japan. Forest
Ecology and Management 183:177-193.

Niemuth, N. D. 2003. Identifying landscapes for greater prairie chicken translocation
using habitat models and GIS: a case study. Wildlife Society Bulletin 31:145-155.

NOAA. 2004. Climate monitoring and diagnostic laboratory carbon cycle greenhouse
gases.

Noguchi, M., and T. Yoshida. 2004. Tree regeneration in partially cut conifer-hardwood
mixed forests in northern Japan: roles of establishment substrate and dwarf
bamboo. Forest Ecology and Management 190:335-344.

Oliver, C., and B. Larson. 1996. Forest stand dynamics. John Wiley and Sons, Inc., New
York.

Ott. L. 1993. An introduction to statistical methods and data analysis., 4 edition. PWS-
Kent, Boston.

PAl. 1999. One in three people lives in forest-scarce countries. Population Action
International, Washington, DC.

Palmer, S. C. F., and A. M. Truscott. 2003. Seasonal habitat use and browsing by deer in
Caledonian pinewoods. Forest Ecology and Management 174:149-166.

Parker, A. 1982. The topographic relative moisture index: an approach to soil-moisture
assessment in mountain terrain. Physical Geography 3:160-168.

Peng. X., and Z. Guo, editors. 2000. The changing population of China. Blackwell,
Oxford.

Petranka, J. W. 1994. Response to Impact of Timber Harvesting on Salamanders.
Conservation Biology 8:302-304.

Piirainen, S., L. Finer, H. Mannerkoski, and M. Starr. 2002. Effects of forest clear-cutting

on the carbon and nitrogen fluxes through podzolic soil horizons. Plant and Soil
239:301-311.

I66

Pottinger, A. 2003. Editorial: Special Issue: Illegal Logging. International Forestry
Review 5:ii-iii.

Potvin, F., R. Courtois, and L. Belanger. 1999. Short-term response of wildlife to clear-
cutting in Quebec boreal forest: multiscale effects and management implications.
Canadian Journal of Forest Research-Revue Canadienne De Recherche Forestiere
29:1120-1127.

Pradhan, S., G. K. Saha, and J. A. Khan. 2001. Ecology of the red panda Ailurus fulgens
in the Singhalila National Park, Darjeeling, India. Biological Conservation 98:11-
18.

Prior, K. A., and P. J. Weatherhead. 1994. Response of Free-Ranging Eastern
Massasauga Rattlesnakes to Human Disturbance. Journal of Herpetology 28:255-
257.

Quinn, G., and M. Keough. 2002. Experimental design and data analysis for biologists.
Cambridge University Press, New York.

Rao, K. S., and R. Pant. 2001. Land use dynamics and landscape change pattern in a
typical micro watershed in the mid elevation zone of central Himalaya, India.
Agriculture Ecosystems & Environment 86:] l3-123.

Rawat, G. S. 1997. Conservation status of forests and wildlife in the Eastern Ghats, India.
Environmental Conservation 24:307-315.

Ray. N., A. Lehmann, and P. Joly. 2002. Modeling spatial distribution of amphibian
populations: a GIS approach based on habitat matrix permeability. Biodiversity
and Conservation 11:2143-2165.

Reid, D. G., and J. Hu. 1991. Giant Panda Selection between Bashania-Fangiana Bamboo
Habitats in Wolong Reserve Sichuan China. Journal of Applied Ecology 28:228-
243.

Reid, D. G., J. Hu, D. Sai, W. Wei, and H. Yan. 1989. Giant Panda Ailuropoda-
Melanoleuca Behavior and Carrying Capacity Following a Bamboo Die-Off.
Biological Conservation 49285-104.

Reid, D. G., M. Jiang, Q. Teng, Z. Qin, and J. Hu. 1991a. Ecology ofthe Asiatic Black
Bear (Ursus-Thibetanus) in Sichuan, China. Mammalia 55:221-237.

Reid, D. G., A. H. Taylor, J. C. Hu. and Z. S. Qin. l99lb. Environmental Influences on
Bamboo Bashania-Fangiana Growth and Implications for Giant Panda
Conservation. Journal of Applied Ecology 28:855-868.

Reid, D. G., A. H. Taylor, J. C. Hu. and Z. S. Qin. 1991c. Environmental-Influences on
Bamboo Bashania-Fangiana Growth and Implications for Giant Panda
Conservation. Journal of Applied Ecology 28:855-868.

I67

Roberts, M. R. 2002. Effects of forest plantation management on herbaceous-layer
composition and diversity. Canadian Journal of Botany-Revue Canadienne De
Botanique 80:378-389.

Roberts, M. R., and F. S. Gilliam. I995a. Disturbance effects on herbaceous layer
vegetation and soil nutrients in Populus forests of northern lower Michigan.
Journal of Vegetation Science 6:903-912.

Roberts, M. R., and F. S. Gilliam. l995b. Patterns and Mechanisms of Plant Diversity in
Forested Ecosystems - Implications for Forest Management. Ecological
Applications 5:969-977.

Roberts. M. R., and L. X. Zhu. 2002. Early response of the herbaceous layer to harvesting
in a mixed coniferous-deciduous forest in New Brunswick, Canada. Forest
Ecology and Management 155:17-31.

Robinson, J. G., K. H. Redford, and E. L. Bennett. 1999. Conservation - Wildlife harvest
in logged tropical forests. Science 284:595-596.

Rodrigues, A. S. L., S. J. Andelman, M. I. Bakarr, L. Boltani, T. M. Brooks, R. Cowling.
L. Fishpool, G. da Fonseca, K. Gaston, M. Hoffmann, J. Long, P. Marquet, J.
Pilgrim, R. Pressey, J. Schipper, W. Sechrest, S. Stuart, L. Underhill, R. Waller,
M. Watts, and X. Yan. 2004. Effectiveness of the global protected area network in
representing species diversity. Nature 428:640-643.

Ruiz-Perez. M., F. Maoyi, Y. Xiaosheng, and B. Belcher. 2001. Bamboo forestry in
China. Journal of Forestry 99:14-20.

Saenz, G. P., and M. R. Guariguata. 2001. Demographic response of treejuveniles to
reduced-impact logging in a Costa Rican montane forest. Forest Ecology and
Management 140:75-84.

Sakuragi. M., H. Igota, H. Uno, K. Kaji, M. Kaneko, R. Akamatsu, and K. Maekawa.
2003. Seasonal habitat selection of an expanding sika deer Cervus nippon
population in eastern Hokkaido, Japan. Wildlife Biology 9:141-153.

Sallabanks, R., E. B. Arnett, and J. M. Marzluff. 2000. An evaluation of research on the
effects of timber harvest on bird populations. Wildlife Society Bulletin 28:] 144-
l 155.

SAS Institute. 1985. SAS/STAT guide for personal computers. in. SAS Institute.

Sayer, J ., U. Chokkalingam, and J. Poulsen. 2004. The restoration of forest biodiversity
and ecological values. Forest Ecology and Management 201:3-11.

Schaller, G. B., J. Hu, W. Pan, and J. Zhu. I985. The giant pandas of Wolong. University
of Chicago Press, Chicago.

I68

Scherer, G., D. Zabowski, B. Java, and R. Everett. 2000. Timber harvesting residue

treatment. Part II. Understory vegetation response. Forest Ecology and
Management 126235-50.

Shi. P.. and J. Xu. 2000. Deforestation in China. Center for Chinese Agricultural
Sciences, Chinese Academy of Sciences, Beijing, PR China.

Simard, J. R., and J. M. Fryxell. 2003. Effects of selective logging on terrestrial small
mammals and arthropods. Canadian Journal of Zoology-Revue Canadienne De
Zoologie 81: 13 18-1326.

Smil, V. 1984. The bad earth: environmental degradation in China. Zed Press, London.

Smith, A. P., and D. B. Lindenmayer. 1992. Forest Succession and Habitat Management
for Leadbeaters Possum in the State of Victoria, Australia. Forest Ecology and
Management 49:31 1-332.

Smith, D. F ., and J. A. Litvaitis. 2000. Foraging strategies of sympatric lagomorphs:
implications for differential success in fragmented landscapes. Canadian Journal
of Zoology-Revue Canadienne De Zoologie 78:2134-2141.

Stanturf, J. A., S. H. Schoenholtz, C. J. Schweitzer, and J. P. Shepard. 2001. Achieving

. restoration success: Myths in bottomland hardwood forests. Restoration Ecology
9: 189-200.

State Environmental Protection Administration of China. 1998. China's biodiversity: A
country study. China Environmental Science Press, Beijing. _

Stier, S. C., and S. F. Siebert. 2002. The Kyoto Protocol: an opportunity for biodiversity
restoration forestry. Conservation Biology 16:575-576.

Store, R., and J. Jokimaki. 2003. A GIS-based multi-scale approach to habitat suitability
modeling. Ecological Modelling 169:1-15.

Stott, T., and N. Mount. 2004. Plantation forestry impacts on sediment yields and
downstream channel dynamics in the UK: a review. Progress in Physical
Geography 28: I97-240.

Su, Z., and T. J. Quinn. 2003. Estimator bias and efficiency for adaptive cluster sampling

with order statistics and a stopping rule. Environmental and Ecological Statistics
10:17-41.

Sullivan, T. P., R. A. Lautenschlager, and R. G. Wagner. 1999. Clearcutting and burning

of northern spruce-fir forests: implications for small mammal communities.
Journal of Applied Ecology 36:327-344.

I69

Sullivan, T. P., and D. S. Sullivan. 2001. Influence of variable retention harvests on forest

ecosystems. 11. Diversity and population dynamics of small mammals. Journal of
Applied Ecology 38:1234-1252.

Sullivan, T. P., D. S. Sullivan, and P. M. F. Lindgren. 2000. Small mammals and stand
structure in young pine, seed-tree, and old-growth forest, southwest Canada.
Ecological Applications 10:1367-1383.

Sun. X. F., E. Katsigris, and A. White. 2004. Meeting China's demand for forest
products: an overview of import trends, ports of entry, and supplying countries,

with emphasis on te Asia-Pacific region. International Forestry Review 6:227-
236.

Sun, X. F ., L. Q. Wang, and Z. B. Gu. 2005. China and forest trade in the Asia-Pacific
region: Implications for forests and livelihoods. A brief overview of China's
timber market. Forest Trends.

Sykes, J. M., V. P. W. Lowe, and D. R. Briggs. 1989. Some Effects of Afforestation on
the Flora and Fauna of an Upland Sheepwalk During 12 Years after Planting.
Journal of Applied Ecology 26:299-320.

Takahashi, K., and T. Kohyama. I999. Size-structure dynamics of two conifers in

relation to understorey dwarf bamboo: A simulation study. Journal of Vegetation
Science 10:833-842.

Takahashi, K., S. Uemura, J. I. Suzuki, and T. Hara. 2003. Effects of understory dwarf
bamboo on soil water and the growth of overstory trees in a dense secondary
Betula ermanii forest, northern Japan. Ecological Research 18:767-774.

Tanaka, N. I988. Tree invasion into patchy dwarf-bamboo thickets within a climax
beech-fir forest in Japan. Pages 253-261 in J. H. Willems, editor. Diversity and
pattern in plant communities. SPB Academic Publishing, The Hague, The
Netherlands.

Taylor, A. H., J. Y. Huang, and S. Q. Zhou. 2004. Canopy tree development and
undergrowth bamboo dynamics in old-growth A bies-Betula forests in
southwestern China: a 12 year study. Forest Ecology and Management 200:347-
360.

Taylor, A. H., and Z. Qin. I987. Culm Dynamics and Dry Matter Production of Bamboos
in the Wolong and Tangjiahe Giant Panda Reserves Sichuan China. Journal of
Applied Ecology 24:4]9-434.

Taylor, A. H., and Z. Qin. 1988a. Regeneration from Seed of Sinarundinaria-Fangiana a

Bamboo in the Wolong Giant Panda Reserve Sichuan China. American Journal of
Botany 75:1065-1073.

I70

Taylor. A. H., and Z. Qin. I988b. Tree replacement patterns in subalpine Abies-Betula
forests, Wolong Natural Reserve, China. Vegetatio 78: 141-149.

Taylor, A. H., and Z. H. Qin. 1989. Structure and composition of selectively cut and
uncut Abies-Tsuga forest in Wolong Natural Reserve and implications for panda
conservation in China. Biological Conservation 47283-108.

Taylor, A. H., and Z. S. Qin. l988c. Regeneration patterns in old-growth Abies-Betula
forests in the Wolong Natural Reserve, Sichuan, China. Journal of Ecology
76:1204-1218.

Taylor, A. H., and Z. S. Qin. 1993a. Aging Bamboo Culms to Assess Bamboo
Population-Dynamics in Panda Habitat. Environmental Conservation 20:76-79.

Taylor, A. H., and Z. S. Qin. I993b. Bamboo Regeneration after Flowering in the
Wolong Giant Panda Reserve, China. Biological Conservation 63:231-234.

Taylor, A. H., and Z. S. Qin. 1997. The dynamics of temperate bamboo forests and panda
conservation in China. Pages 370 pp. in G. P. Chapman, editor. The Bamboos.
Academic Press, San Diego.

Taylor, A. H., Z. S. Qin, and J. Liu. 1995. Tree regeneration in an Abies faxoniana forest
after bamboo dieback, Wang Lang Natural Reserve, China. Canadian Journal of
Forest Research-Revue Canadienne De Recherche Forestiere 25:2034-2039.

Taylor, A. H., D. G. Reid, Z. S. Qin, and J. C. Hu. I991. Bamboo dieback - an
opportunity to restore panda habitat. Environmental Conservation 18:166-168.

Taylor, C. R., V. J. Rowntree, and S. L. Caldwell. 1972. Running up and Down Hills -
Some Consequences of Size. Science l78:lO96-&.

Taylor, S., and M. R. Perrin. 1996. Identification of the rodent species involved in tree

damage in commercial forestry in the Natal midlands, South Africa. Mammalia
60:767-773.

Thompson, F. R., and R. M. DeGraaf. 2001. Conservation approaches for woody, early
successional communities in the eastern United States. Wildlife Society Bulletin
29:483-494.

Thompson, S. K. 1991a. Adaptive cluster sampling. Biometrika 78:389-397.

Thompson, S. K. 1991 b. Adaptive cluster sampling - designs with primary and secondary
units. Biometrics 47:] 103-1 I 15.

Thompson, S. K., F. L. Ramsey, and G. A. F. Seber. 1992. An Adaptive Procedure for
Sampling Animal Populations. Biometrics 48:] 195-1 199.

171

Thompson. S. K., and G. Seber. 1996. Adaptive sampling. John Wiley and Sons, Inc.,
New York, New York.

Turk, P., and J. J. Borkowski. 2005. A review of adaptive cluster sampling.
Environmental and Ecological Statistics 12:55-94.

Turner, M. G., G. J. Arthaud, R. T. Engstrom, S. J. Hejl, J. G. Liu, S. Loeb, and K.
McKerey. I995. Usefulness of Spatially Explicit Population-Models in Land
Management. Ecological Applications 5:12-16.

Twedt, D. J ., and J. Portwood. I997. Bottomland hardwood reforestation for Neotropical
migratory birds: Are we missing the forest for the trees. Wildlife Society Bulletin
25:647-652.

Twedt, D. J ., and R. R. Wilson. 2002. Development of oak plantations established for
wildlife. Forest Ecology and Management 162:287-298.

Twedt, D. J ., R. R. Wilson, J. L. Henne-Kerr, and D. A. Grosshuesch. 2002. Avian
response to bottomland hardwood reforestation: The first 10 years. Restoration
Ecology 10:645-655.

Urban. D., S. Goslee, K. Pierce, and T. Lookingbill. 2002. Extending community ecology
to landscapes. Ecoscience 92200-212.

Venables, W., and B. Ripley. 1999. Modern applied statistics with S-PLUS, 3 edition.
Springer-Verlag, New York.

Viljoen, P. 1982. The distribution and population status of the larger mammals in
Kaokoland, South West Africa/Namibia. Cimbebasia 7 :7-33.

Waller, J. S., and R. D. Mace. 1997. Grizzly bear habitat selection in the Swan
Mountains, Montana. Journal of Wildlife Management 61:1032-1039.

Wang, S., and L. Chen. 1992. Biodiversity in China: Status and conservation needs.
Biodiversity Commission of the Chinese Academy of Sciences, Beijing, China.

Ward, D. H., R. A. Stehn, and D. V. Derksen. 1994. Response of Staging Brant to
Disturbance at the lzembek Lagoon, Alaska. Wildlife Society Bulletin 22:220-
228.

WEC. 2004. Energy efficiency: a worldwide review. World Energy Council, London.
Wenger, K., editor. 1984. Forestry Handbook. Wiley Interscience, New York, NY.
White, L. 1992. Vegetation history and logging disturbance: effects on rain forest

mammals in Lope Reserve, Gabon. Ph.D. Dissertation. University of Edinburgh,
Edinburgh.

I72

Wielgus, R. B., and P. R. Vernier. 2003. Grizzly bear selection of managed and
unmanaged forests in the Selkirk Mountains. Canadian Journal of Forest
Research-Revue Canadienne De Recherche Forestiere 33:822-829.

Wolong Nature Reserve. 2003. Gray literature documentation on extant of reforestation.
Wolong Nature Reserve Administration, Wolong, Sichuan, PRC.

World Wildlife Fund. 2004. Giant panda survey 2004: Questions and answers. Beijing.

Wormington, K. R., D. Lamb, H. I. McCallum, and D. J. Moloney. 2002. Habitat
requirements for the conservation of arboreal marsupials in dry sclerophyll forests
of Southeast Queensland, Australia. Forest Science 48:217-227.

Xinhua. 2004. Survey: China has over 1,750 giant pandas. in China Daily, Beijing.

Xu. J., and A. Wilkes. 2004. Biodiversity impact analysis in northwest Yunnan,
southwest China. Biodiversity and Conservation 13:959-983.

Xu. Z., M. Bennett, R. Tao, and J. Xu. 2004. China's Sloping Land Conservation
Programme four years on: current situation and pending issues. International
Forestry Review 6:317-326.

Zar, J. H. 1996. Biostatistical analysis. Prentice Hall, Upper Saddle River, New Jersey.

Zhao, G.. and G. F. Shao. 2002. Logging restrictions in China - A turning point for forest
sustainability. Journal of Forestry 100:34-37.

Zhao, M., and G. S. Zhou. 2005. Estimation of biomass and net primary productivity of
major planted forests in China based on forest inventory data. Forest Ecology and
Management 207 :295-3 1 3.

Zheng, S., Y. Yu, H. Zuo, Y. Cao, W. Yuan, and W. Fu. I995. The present status,

distribution, and conservation measure of the black bear in Hanzhong prefecture.
Shaanxi province. Acta Theriologica Sinica 15:93-97.

I73

 

 

   
 

lll‘l Hll

3 1293 02736

 

‘