viryli‘t};

.39“... 2g

, .Kc
32$. . . . .
a. 9...!”me : . . . a . . .
. mfifi
aaumflmm.

....

1L: . .n . . . z x .. . . . . . . .

a... [Chil- . .7 3 . . . . . . . . I

v . l. . . . . . . . . .

h@%fie .fififi .,. . . . , . .fifififl
. u. .. n . , . . y . . ‘3.

I.) . thug. . . , . . . . h..il‘UUHAuI-i( n
«5? 3.252, . a. .n..».:.:..:.: . . _ . . . . find»... 1%..»
I§PAI§FV11§1 .1 ....r altar?! . . . . n
.c ......;..1... 5i ... :32. . . .

“fawn, f :3»? n..{ . I}... . . . .
. % ,w
J:
.
5.1.5.3....
. r . ‘
.33... Law.»
. .9. .
“ulna:
...;

Lri
It.

....LICI. 5. :5
51:17)): 9.2.9.1.?

..\ .1! s3.
.. .x (Phil; (

...A

 

..Nunuwiufinmflwfi... . . . . ..., . ,. .. . . . . .. . 111.211....Q H.

.I .

 

L.

 

LIBRARY
Michigan State
University

 

 

 

This is to certify that the

dissertation entitled

MODELING THE DRIVING FORCES OF
LAND USE AND LAND COVER CHANGES
ALONG THE UPPER YANGTZE RIVER IN CHINA

presented by

QING XIANG

has been accepted towards fulfillment
of the requirements for the

Ph. D degree in Forestry

 

 

-

W

Major Professor’s Signature

{/o?/(O

Date

MSU is an Affirmative Action/Equal Opportunity Employer

 

 

 

_ -l-.---.-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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5108 KrlPrq/AcaPreuCIRCIDateDue.indd

 

MODELING THE DRIVING FORCES OF LAND USE AND LAND COVER
CHANGES ALONG THE UPPER YANGTZE RIVER IN CHINA

By

QING XIANG

A DISSERTATION

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

DOCTOR OF PHILOSOPHY
Forestry

2009

ABSTACT

MODELING THE DRIVING FORCES OF LAND USE AND LAND COVER
CHANGES ALONG THE UPPER YANGTZE RIVER IN CHINA

By

QING XIANG

Induced by high population density, rapid but uneven economic growth, and historic
resource exploitation, China’s upper Yangtze basin has witnessed remarkable changes in
land use and cover, which have resulted in severe environmental consequences, such as
flooding, soil erosion, and habitat loss. This research depicts the land use pattern and
examines the driving forces of land use and land cover change (LUCC) along the Jinsha
River, one primary section of the upper Yangtze, aiming at better understanding the
complex human and nature processes underlying the LUCC and providing policy
implications on sustainable land use and environmental protection.

Using land use data derived from remotely sensed images and second-hand survey
and statistics covering 31 counties over 5 time points from 1975 to 2000, the research
first constructs a fractional logit model to empirically examine the effects of
socioeconomic and institutional factors on changes for the share of cropland, forestland,
and grassland. Next, the research develops a cropland structural model and a forestland
structural model respectively to flirther investigate the complex human and natural
processes underlying the LUCC. Cropland structural model illustrates that cropland use
interacts with grain production and agricultural technological progress and meanwhile the

environmental consequence of land uses also imposes significance feedbacks. Forestland

structural model analyzes the forestland change in consideration of its interrelationships
with stocking volume and timber production. Results illustrate the critical role of
agricultural technological change in supplying food on a limited cropland, highlight
important impacts of institutions and policies — such as forest ownership, food self-
sufficiency policy, forest and soil conservation projects — on land use and resource status,
and also show significant effects of population expansion, industry development, and
better market access and natural factor on the LUCC.

The research not only improves our understanding of the complex human and natural
connections in the LUCC process, but also implies that understanding multiple driving
forces and their different impacts on land uses will help prioritize and balance various
land use policies and actions. The research provides some suggestions on supporting the
agricultural technological progress, reforming the collective forestland tenure,
encouraging market development and expanding environmental protection programs. In
observation of limitations of this research, the future research directions are also

discussed.

ACKNOWLEDGMENT

I owe my gratitude to all those people who have made this dissertation possible and
because of whom my graduate experience has been one that I will cherish forever.

My deepest gratitude is to my major advisor, Dr. Runsheng Yin. I would like to
sincerely thank Dr. Yin for his intellectual guidance and strong support whenever and for
whatever I need. His extensive experience in resources economics and insightful
understanding on international forestry exhibit me an exciting world for exploring. I am
always impressed by his enthusiasm on research, his hard working, and his visions on
new research fields. I am grateful to him for providing me opportunities to touch on
various projects. His instructions help me enrich but focus my ideas at different stages. I
also appreciate his patience all the time, and his care about my living at school. Thanks
from my heart also go to Dr.Yin’s great wife, Xiu Du.

I would like to extend my special thanks to my committee members, Dr. Karen
Potter-Witter, Dr. Larry Leefers and Dr. J iaguo Qi. Their insightful critics, constructive
comments, guidance, and careful reviews on my dissertation help me overcome
difficulties and finalize my research. I am also grateful for their kind understanding,
patience and encouragement when I was slow of going forward.

Grateful and sincere thanks also go to Dr. Jikun Huang, Dr. Linxiu Zhang, Dr. Jitao
Xu, and Dr.Xiangzeng Deng. I am indebted to them for their support on data collection,
field surveys, and comments on the research. Dr. Jikun Huang, Dr. Linxiu Zhang and Dr.

Jintao Xu have been advising me since my pursuit of MS. in China, and I am thankful

iv

for their continuous encouragement, support and inspiration. Thanks are also given to
Ping Zhao and his colleagues who worked hard and helped a lot in field surveys.

I am also grateful to faculty and staff at the Department of Forestry. I learned a lot
from classes, seminars and discussions; and helps from staff make it so comfortable
studying at the Department. I would also like to express gratitude to my graduate
colleagues and friends at MSU. Their care and friendships make my staying and working
at MSU an enjoyable and awarding experience.

Finally, and the most importantly, I would like to thank my dearest — my husband
Honglin Wang, my parents Yixin Xiang and Zhongqing Wei, my brother Wei Xiang and
his family. They are there for me all the time and home is always the place I most want to
go. My husband and my family accompanied me going through all ups and downs, and I
would not ask more than they always give me: love, care, encouragement, strength and

happiness.

TABLE OF CONTENTS

LIST OF TABLES ........................................................................................................... viii
LIST OF FIGURES ............................................................................................................ x
Chapter 1 Introduction ................................. 1
1.1 Background ............................................................................................................... 1
1.2 Research Questions and Objectives .......................................................................... 5
1.3 Significance of the Study .......................................................................................... 6
1.4 Dissertation Outline .................................................................................................. 7
Chapter 2 Literature Review ............................................................................................... 9
2.1 . Introduction .............................................................................................................. 9
2.2 Literature on the Driving Forces of LUCC ........................................................... 10
2.2.1 Scales and Structures of the LUCC Models .................................................... 10
2.2.2 Variables Used in the LUCC Driving Forces Models ..................................... 13
2.2.3 Studies on China’s LUCC ................................................................................ 18

2.3 Synthesis and Future Research ............................................................................... 20
Chapter 3 Land Use Changes in the Study Site ................................................................ 23
3.1 Study Site ................................................................................................................ 23
3.2 Data from Remotely Sensed Images ....................................................................... 28
3.3 Temporal and Spatial Land Uses Patterns .............................................................. 30
3.3.1 Temporal Variation of Land Uses/Covers ....................................................... 30
3.3.2 Spatial Variation of Land Use Patterns...................... ...................................... 33

3.4 Land Use Changes over Time ................................................................................. 38
3.4.1 Land Conversions between Primary Land Use Categories .............................. 38
3.4.2 Forestland Modification ................................................................................... 44

3.5 Comparing Cropland Data from Different Sources ................................................ 50
Chapter 4 Analyzing the LUCC Drivers Using Discrete Choice Models ........................ 52
4.] Conceptual Framework ........................................................................................... 53
4.2 Empirical Model and Estimation Method ............................................................... 56
4.3 Variable Selection and Data Description ................................................................ 59
4.3 Estimation Results .................................................................................................. 68
4.4 Conclusion and Discussion ..................................................................................... 74
Chapter 5 Structural Models for Cropland and Forestland Change .................................. 76
5.1 Introduction ............................................................................................................. 76
5.2 Structural Model for Cropland Change ................................................................... 77
5.2.1 Conceptual Framework for Cropland Change Structural Model ..................... 77
5.2.2 Empirical Model and Variable selection .......................................................... 82
5.2.3 Data Description .............................................................................................. 88

vi

5.2.4 Estimation Method for the Cropland Structural Model .................................... 95

5.2.5 Estimation Results for the Cropland Structural Model ..................................... 96
5.3 Structural Model for Forestland Change ............................................................... 102
5.3.1 Conceptual Framework for Forestland Structural Model ............................... 102
5.3.2 Empirical Model for the F orestland Structural Model ................................... 104
5.3.3 Estimation Method and Data Description ....................................................... 107
5.3.4 Estimation Results for the Forestland Structural Model ................................. 111
5.4 Discussion of Model Results ................................................................................. 115
Chapter 6 Conclusion and Discussion ............................................................................. 117
6.1 Overview ................................................................................................................ 117
6.2 Summary of Research Findings ............................................................................. 117
6.3 Policy Implications ................................................................................................ 122
6.4 Future Research Discussion ................................................................................... 125
Appendices ...................................................................................................................... 128
Appendix 3.1: Technical Notes on Classification of Remotely Sensed Images .......... 128
Appendix 4.1: Testing Estimation of the Fractional Logit Model for Primary Land
Uses .............................................................................................................................. 129
Appendix 5.1: Year and Province Dummy Estimate for the Cropland Structural Model
..................................................................................................................................... 131
Appendix 5 .2: Coefficients and SD. for the Forestland Structural Model ................. 132
BIBLIOGRAPHY ............................................................................................................ 1 33

vii

LIST OF TABLES

Table 2-1 Summary of Human—Dimension Variables in previous LUCC Models ........... 17
Table 3-1 Definition of Land Classification ..................................................................... 29
Table 3-2 Land Use Pattern in the Study Region over 1975-2000 ................................... 30
Table 3-3 Land Uses/Covers from 1975 to 2000 in Sichuan and Yunnan ....................... 34
Table 3-4 Shares for each Land Use at Three Elevation Zone ......................................... 36
Table 3-5 Area (and Percent) of Land Converted from One Use to Another between 1975
and 1985 .................................................................................................................... 42
Table 36 Area (and Percent) of Land Converted from One Use to Another between 1985
and 1990 .................................................................................................................... 42
Table 3-7 Area (and Percent) of Land Converted from One Use to Another between 1990
and 1995 .................................................................................................................... 43
Table 3-8 Area (and Percent) of Land Converted from One Use to Another between 1995
and 2000 .................................................................................................................... 43
Table 3-9 Area (and Percent) of Forestland Modified from One Sub-category to Another
between 1975 and 1985 ............................................................................................ 46
Table 3-10 Area (and Percent) of Forestland Modified from One Sub-category to Another
between 1985 and 1990 ............................................................................................ 46
Table 3-11 Area (and Percent) of Forestland Modified from One Sub-category to Another
between 1990 and 1995 ............................................................................................ 47
Table 3-12 Area (and Percent) of Forestland Modified from One Sub-category to Another
between 1995 and 2000 ............................................................................................ 47
Table 3-13 The Ratio of Cropland from the Government Statistics to That Derived from
the Remotely Sensed Images .................................................................................... 51
Table 4-1 Summary of Variables Used in the Fractional Logit Model ............................ 63
Table 4-2 Expected Signs for Variables in Fractional Logit Model of ............................. 67

viii

Table 4-3 Estimated Results of Fractional Logit Model for Primary Land Use ............... 69

Table 4-4 Estimated Results of Fractional Logit Model for Secondary Forest Classes 73

Table 5-1 Summary of Additional Variables used in Cropland Structural Model ........... 89
Table 5-2 Expected Signs for Variables in the Cropland Structural Model ..................... 94
Table 5-3 Estimated Results for the Cropland Structural Model ...................................... 98
Table 54 Additional Variables used in Forestland Structural Model ........ 107
Table 5-5 Expected Signs for Variables in the Forestland Structural Model ................. 111
Table 5-6 Estimated Results for the Forestland Structural Model .................................. 112
Table A4] Testing Estimated Results of the Fractional Logit Model .................. 130

ix

Chapter 1 Introduction

Studying land use/cover change (LUCC) has been recognized as a great challenge in
global change studies. A better understanding of the causes and consequences of LUCC
is essential for global change studies, because of its tremendous effects on climate
change, ecosystem functions, and human welfare. Over the past three decades, China has
witnessed substantial changes in land uses and covers, and these changes have induced
serious environmental consequences and even affected its sustainable development.
Taking China’s upper Yangtze — the region well known for its importance on ecological
functioning and biodiversity — as the study site, this research intends to improve the
knowledge of various driving forces of the regional LUCC.
1.1 Background

LUCC has attracted tremendous attention within the past two decades. It has been
widely recognized that past LUCC has adversely affected ecosystem functions, carbon
and water cycles, and human welfare (Miiller and Zeller 2002; Geoghegan et a1. 2001;
Turner et a1. 1994); and potential LUCC in the coming decades will further intensify the
trends of climate change, species extinction, groundwater depletion, desertification, and
soil nutrient losses (IPCC 2001). The National Research Council (NRC 2001) identified
land-use dynamics as one of the great challenges for environmental research. Thus, as the
US. Global Change Research Program (USGCRP 2004) has declared, it is essential for
the global change research to better understand the processes, rates, causes, and
consequences of land use change and land management practices. Achieving such an
understanding will certainly lead us into the realm of socioeconomic and biophysical

sciences because changes in land covers and uses are governed by human and natural

driving forces. Comprehensively and accurately modeling the causes of LUCC will better
illustrate human-environmental interactions at multiple scales, more reliably project the
future trend of land uses and covers, and provide viable policy suggestions on planning
and designing more sustainable land uses.

Within the past three decades, China has witnessed substantial changes in land uses
and covers, due to continuous population growth, rapid but uneven economic
development, and long-time resource exploitation, as well as poor regulations and
inadequate institutions. Such changes have [induced serious environmental consequences
and even affected its long-run economic development. How to control land degradation
and allocate limited land resources in order to simultaneously satisfy the demands for
food production, raw materials, urban expansion, and quality environment for a long term
(so called sustainable land use), has become an unprecedented challenge. Therefore, it is
imperative and important to examine how the land uses/covers have been affected by
various human and natural factors. Doing so will help create solutions that incorporate
stable economic development with sustainable resource utilization and environmental
services in China, as well as in other developing countries.

In China, the upper Yangtze River basin constitutes a great site for LUCC research.
The Yangtze is China’s longest and the world’s third largest river, which starts out from
the Tibetan Plateau, courses 6,300 kilometers (km), and finally flows to the Eastern
China Sea. The Yangtze boasts about 40% of the country’s potential hydro-power and
nurtures around 420 million people (in 2001), with a total catchment area of 1.8 million

km2 (19% of the whole country’s area). The Yangtze Delta has been leading China’s

economic growth since the early 19808. However, the development of the whole basin is
threatened by the environmental deterioration in the upper reaches.

The upper reaches of the Yangtze River refer to the vast area west to Yichang of
Hubei, with a total land area of over 1.05 million kmz. The regional population is around
163 million, or 160 persons per kmz, and it is projected to increase to 240 million by
2030 (Du 2001). The upper Yangtze features a wide variety of ecosystems, including
broad-leaved and coniferous forests, bamboo groves, shrub communities, freshwater
wetlands, meadows, and highland scrub. These ecosystems carry the richest biodiversity
in China and form the most biologically diverse temperate region on earth (UNESCO
2003), and they have been recognized as a major biodiversity hotspot (Conservation
International 2002). Accompanying the diverse ecosystems is the extreme fluctuations in
topography with elevations ranging from more than 6,000 meters (m) to less than 500 m.
Landscapes also vary tremendously, ranging from high mountains and deep gorges in the
west to hills and lowlands in the east. In addition, affected by sub-tropical monsoons,
precipitation in the upper Yangtze is quite uneven, concentrated in June to August.
Therefore, this is a very fragile eco-region due to its geological structure and climate
conditions.

The extensive human activities have changed the land covers in the upper Yangtze to
a large extent. One of China’s two largest natural forest regions is mainly distributed
within Yunnan and Sichuan provinces, through which the River flows. In the 1950S, over
forty forest bureaus were set up in the region to produce timber to fuel the economy;
large tracts of primary natural forests were exploited. Unfortunately, under the old

planning-economy system forest resources were utilized inefficiently, and regeneration

and management were neglected (Yin 1998). In the rural sector, the deprivation of
individual forest ownership in the 19508 discouraged farmers from planting and
managing trees. In the early 1980, the government attempted to devolve the forestland
back to farmers, but the policy was later reversed because of over cutting caused by lack
of confidence in the policy. In many places, farming on steep slopes became common due
to the combined effect of demographic expansion and poor regulation (Xu et a1. 2002;
Loucks et a1. 2001). Moreover, the national strategy of food self-sufficiency tacitly
reinforced the forestland conversion and agricultural land reclamation. As a result, not
only the regional forest cover declined significantly during the period of 1960 to 2000,
but extensive degradation also occurred to the remaining forests (Du 2001). Likewise,
extensive grazing and poor maintenance of rangeland also caused grassland degradation.
Coupled with vulnerable natural conditions, these malpractices have weakened the
ecosystems’ capacity of regulating water and soil, and aggravated water runoff and soil
erosion as well as sediments of the whole region. These factors were deemed to be the
primary sources of the widespread flooding in the Yangtze basin in 1998. The Second
National Soil Erosion Remote Sensing Survey (Li 2001; Yang and Liang 2004) estimated
that the eroded area in the upper Yangtze region reached 0.35 million km2 in the late
19908, accounting for 62% of the total eroded area of the Yangtze River basin and 35%
of the total area of the upper River. Annual soil loss in the upper Yangtze basin reached
1.57 billion tons, amounting to 71% of total soil loss of the whole Yangtze River basin.
The deteriorated environment has not only ruined land productivity in the upper Yangtze
and constituted a threat to the lifespan and the effectiveness of the Three-Gorges Darn

(Lu et a1. 2002), but it has also imposed more risks on economic development and people

livelihoods in the middle and lower reaches of the Yangtze. As such, it has triggered
many concerns about the sustainable development along the river.

Recognizing the seriousness of the current situation, Chinese government has
implemented the Shelterbelt Development Project and the Soil Conservation Project
along the upper Yangtze River since the late 19808, and initiated two of the world largest
ecological rehabilitation projects, the Natural Forest Protection Program (NFPP) and the
Sloping Land Conversion Program (SLCP) since late 19908. The latter two projects are
heavily concentrated in the upper Yangtze as well as the upper and middle Yellow basin.
These projects attempt to protect forest resources, rehabilitate vegetations, control erosion
and flooding, and address other environmental problems such as species extinction and
climate change (Xu et a1. 2007; Loucks et a1. 2001).

1.2 Research Questions and Objectives

While the recent efforts of ecological rehabilitation and land use management have
been unprecedented, more can be done in order to achieve a more sustainable land use
and ecosystem management. There are few rigorous studies about the human and nature
driving forces of LUCC in the upper Yangtze region, while such knowledge will help
decision makers at various levels implement measures to change land use patterns. Thus,
this research will address the following questions:

1. To what extent (magnitude and location) the regional LUCC has happened in the
upper Yangtze River during the past three decades?

2. How has the regional LUCC happened? What are the roles of demographic,

market, technological, economic, institutional and biophysical factors in influencing land

use/cover changes? What is the dynamic LUCC process, considering the interactions of
its driving forces including environmental feedbacks?

3. What policies can be taken to control resources degradation and change land uses?

The overall objective of the study is to gain a scientific understanding of the LUCC
driving forces in the upper Yangtze River. To address the above questions, the specific
tasks of this dissertation research include: (1) measure the extent of the LUCC in the
study site; and (2) develop effective modeling systems to examine the driving forces of
the LUCC, by taking into account the interactions among driving forces, resource
endowments, and environmental feedbacks. The accomplishment of these tasks will lead
to a better understanding of human and natural components in the LUCC process and
some policy implications for sustainable land use.
1.3 Significance of the Study

It is of great academic interest and broad policy significance to study LUCC in the
upper Yangtze basin. First, few rigorous studies have been done in this region. The upper
Yangtze basin is an interesting site for the LUCC study, due to its complex topographic
features, critical environmental and ecological role, and extensive human activities there.
Some descriptive studies have been done for this region; however, the nature and relative
magnitude of the regional LUCC remain largely unknown. Little empirical analysis like
this research has been carried out that employs both remotely sensed information and
statistics to model the LUCC driving forces in a systematic way.

Second, this study will capture the dynamics and human-nature processes underlying
the LUCC process. Specifically, cropland structural model takes into account interactions

among cropland use, grain production, agricultural technological change and the

environmental consequence in analyzing the cropland change, while forestland structural
model illustrates the interrelationships of forestland use with forest stocking volume and
timber harvest. Systematic modeling of the LUCC is not uncommon, but little has been
done to treat the agricultural technological progress endogenous and combine
environmental impacts into the analysis.

Third, the study examines a number of socioeconomic and institutional variables,
including demographic change, market signals, industry development, infrastructural
construction, agricultural technology progress, land tenure, food policy and
environmental project implementations. Doing so will lead to a better understanding of
the LUCC components, provide us the opportunity to examine a driver’s various impacts
on different land uses, and help identify tradeoffs of alternative drivers. It is also expected
that this dissertation research will provide useful information to other countries that face
challenges similar to China’s.

1.4 Dissertation Outline

In addition to this introduction, the dissertation contains the following five chapters:

After highlighting the importance of LUCC studies, Chapter 2 reviews the literature
of previous LUCC driving forces analyses. It closes with a discussion of the contribution
of this dissertation research to the literature.

Chapter 3 contains an introduction of the study site, a description of land use/cover
data derived from remotely sensed images, and an overview of land use changes in the
study region. Chapter 3 will accomplish the first task of this dissertation research, giving
a better understanding of what happened in the study site in terms of its temporal and

spatial features of land use patterns. Moreover, statistics of cropland are compared with

those obtained from remotely sensed images, to justify the utilization of land use/cover
data derived from remotely sensed images.

Chapter 4 analyzes the LUCC driving forces with a fractional logit model. The model
explains how the share of each land use type is affected by various human and natural
factors. The fractional logit model guarantees that the sum of the estimated shares for
each land use type equals unity. After specifying the theoretical framework and empirical
model, the chapter discusses data used for estimation before presenting the results and
elaborating the weakness of the fractional logit model.

Chapter 5 is devoted to structural modeling of cropland and forestland uses. The
cropland structural model explains cropland changes by incorporating interactions among
cropland use, grain production, agricultural technology, and soil erosion (as an
environmental consequence of the land use change). The forestland structural model
examines how forestland area, stocking volume, and timber production are affected by
human and natural factors, and how they interact with each other. For each structural
model, its conceptual framework and relevant variables are described, and then
estimation results are discussed.

Chapter 6 summarizes the findings of this dissertation research. The chapter recasts
the major LUCC in the study region from 1975 to 2000, highlights the empirical results
of different modeling approaches, and then discusses the policy implications for more

sustainable land use. Future research directions are discussed at last.

Chapter 2 Literature Review
2.1. Introduction

Lambin and others (2003), Irwin and Geoghegan (2001), and Kaimowitz and
Angelsen (1998) have provided thorough reviews of the LUCC literature. They all
pointed out that, modeling, especially if done in a spatially explicit and multi-scale
manner, is an important technique for the reliable projection of future LUCC. Early
LUCC work focused on measuring the LUCC and identifying the relationship between
LUCC and geographical characteristics (Irwin and Geoghegan 2001). Data used in the
models generally come from satellite images and topographic maps, using pixel as the
observation unit. The land use pattern is certainly constrained by biophysical factors such
as soil characteristics or topography, but it is also the result of human activities (Turner et
al. 1993). To improve the understanding of the complex LUCC, recently researches have
not only included both geographical and socioeconomic factors as the driving forces of
LUCC, but also attempted to construct more systematic models to capture the complex
processes embedded in LUCC.

Since several researchers have provided comprehensive reviews on the subject, the
literature review in this study focuses on summarizing model structures and variables
selected as the driving forces in the previous studies. Additionally, it will summarize the
LUCC research done for China. The final part of this chapter will highlight certain

important research directions to be pursued by this and other research.

2.2 Literature on the Driving Forces of LUCC
2.2.1 Scales and Structures of the LUCC Models

The relations between land use and its driving factors are dependent on the scale of
observations (Veldkamp and Fresco 1996). Studies indicate that relationships between
the biophysical and social variables differ when the scale at which one examines these
relationships changes (Walsh et al. 1999). Scale usually consists of two aspects — extent,
which refers to the size of the study region, and the resolution, which refers to the size of
an individual (observation) unit (Verburg and Chen 2000). The ‘scale’ in this section
mainly means the spatial size of the study unit, although the temporal scale is also one
critical aspect in LUCC studies (Agarwal et al. 2002). Most often, coarse scales are
useful to reveal the general trends and relations between land use and its determining
factors (Veldkamp and Fresco 1996). However, the broad level of aggregation can
obscure the variability of units and processes at finer level, and is therefore considered
inaccurate for local assessments. On the other hand, change patterns that emerge at the
broad scale result from the environmental, economic or land use changes at the fine scale
(Parker et al. 2002). Therefore, analyses at different scales nright find different
correlations between the land use pattern and its driving forces. Lambin et al. (2003)
pointed out that the proximate (direct) causes of LUCC are those human activities that
usually operate at the local level (e. g., households) and directly affect land covers. Causes
that operate at the regional (e.g., county or province) or even global level are called the
underlying causes, and they include a complex set of social, political, economic,

demographic, technological, cultural, and biophysical variables.

10

In a theoretical context, Kaimowitz and Angelsen (1998) classified the economic
models for deforestation into three categories — micro-level model (household and firm
level), regional-level model, and macro-level model. Models for each category are based
on different underlying assumptions, methodology, and variables. For example, the
micro-level models are grouped as open economy model, subsistence and Chayanovian
model, empirical and simulation model, while macro-level models are grouped as
analytical model, computable general equilibrium model, trade and commodity model,
and multi-country regression model. The micro-level models are better used to
investigate how the land use decision is affected by the emerging labor market, poverty,
long-term investment vs. short term incomes and individual differentiations, while the
regional models are more appropriate to answer where land use changes happened thanks
to the availability of the remote sensing and spatially explicit information.

Empirical research has been done at various scales. Pixel has been widely used in
studies of either monitoring the land use changes (Irwin and Geoghegan, 2001; Pontius Jr
et al. 2001)) or evaluating the causes of LUCC (Nelson et al. 1999; Walsh et al. 1999).
Recognizing the problem that pixel is not directly associated to any social unit of
observation (Mertens et al. 2000), many studies have been based on the unit of decision
making. The specific scale for study ranges from plot (Chomitz and Gray 1996; Evans et
a1. 2001) to individual farm property (McCrachen et al. 1999), household (Walker et al.
2000), village (Mertens et al. 2000), and county (Miller and Plantinga 1999; Parks et al.
2000). And some of them feature multi-scale analyses (Geoghegan et al. 2001; Verburg

et a1 2002).

11

The structure of the LUCC models has also evolved to incorporate the LUCC
dynamics. Agarwal and others (2002) reviewed and characterized 19 land use models,
based on spatial complexity (spatial interactions, dynamic systems modeling or single
module), temporal complexity (short term vs. long term interval decision making
process), and involvement of human activities. For example, the CLUE (Conversion of
Land Use and its Effects) model (Veldkamp and Fresco 1996) was described as a
‘discrete finite state model’ with multiple modules (e,g., biophysical, land-use objective,
and land allocation modules) and it considers various biophysical and human drivers at
different scales. The General Ecosystem Model (Fitz et al. 1996) was viewed as a
‘dynamic systems model,’ which reflected the feedbacks and interactions between units
and across time, but without adequate considerations of human decision-making. Some
studies like Chomitz and Gray (1996) used econometric models with multiple equations
for various land uses, where each equation represents a different land use. Nelson et al.
(1999) and Miller and Plantinga (1999) used multinomial logit models to estimate the
changes of various land uses. Walker and others (2000) developed a multi-equations
model to capture the connection between deforestation and cattle production. Other
studies adopt a single-equation to model the change for one land use, such as
deforestation (Parks et al. 1998; McCracken et a1. 1999; Geoghegan et al. 2001).
Moreover, household data have been used to fit the system model (Evans et al. 2001) or
the agent-based model (Manson and Evans 2007). Such models consider the constraints
of labors and other inputs, and the learning or evolving decision-making in the land use

change process.

12

2.2.2 Variables Used in the LUCC Driving Forces Models

LUCC is driven by complex interactions between various human and biophysical
factors that act over a wide rage of temporal and spatial scales (Verburg et a1. 2002).
Turner (1993) summarized possible human driving forces for land use change into six
elements: population, affluence, technology, political economy (systems of exchange,
ownership and control), political structure (institutions and organizations of governance)
and attitudes and values of individuals and groups. Similarly, Lambin and others (2003)
grouped the LUCC causes as natural and biophysical environment, economic and
technological factors, demographic factors, institutional and policy context, cultural
factors, and globalization. Kaimowitz and Angelsen (1998) classified the causes of
deforestation as parameters that affect agent decisions and macro-level underlying
factors. Included in the former are crop and timber prices, wages and input prices,
technology application, and property regimes; included in the latter are population,
income, external distribution and trade environment, and political factors. Usually, an
LUCC model tends to incorporate as many relevant factors as possible because they act
together in impacting the land use patterns. The main drivers used in studies, including
demographic, market, technological, and institutional factors, will be highlighted below.

Given that population affects land use (Lambin et al. 2003), demographic variables
can be found in almost every LUCC study. Increasing population density reflects greater
pressures on linrited land resources for food and/or living spaces (Mertens et al. 2000),
leading to higher demand for farmland and built-up land. Demographic features refer to
not only the total population change, but also other aspects like the rural labor growth,

labor transfer, or household development in a life cycle that also impose great effects on

13

land use (Lambin et al. 2003). And these factors have implications to labor availability
for extensive farming, deforestation dynamics, or frontier exploration (McCracken et al.
1999). Pfaff (1999) found that the first immigrants to a county have greater impact than
the later ones, while population density per se does not have a significant influence;
Shively and Fisher (2004) showed that off-farm wages attract labor to non-farming
sectors and thus reduce forest clearing, while the number of on-farming laborers and
forest clearing has a positive correlation. Walker et al. (2000) provided an account of the
link between cattle ranching and deforestation in the Amazon. They showed that for areas
with major in-migration of small producers, forest clearing at the household level is
mainly attributable to the availability of hired labor.

According to the quality-based land rent theory, variables that affect benefits/returns
from land, such as output prices or input costs, determine the land use patterns and
changes (Barlowe 1958). This theory also suggests that variations in location and
transportation cost give rise to differential land rents and land use patterns around the
settlement center. Thus, market signals or their proxies are widely used in the LUCC
models. Parks and others (2000) explicitly examined the effects of market value of crops
and livestock and costs of crop production on forestland use. Mertens and others (2000)
investigated the linkage between declining cocoa/coffee prices and deforestation in
Eastern Cameroon; and Cropper and Griffiths (1994) included timber price directly in
their model. Pfaff (1999) employed soil suitability as a proximate to potential land
revenue, and showed that more fertile land received higher economic benefits, and thus
was claimed before inferior land was brought into use. Walker and others (2000) showed

that the market signal for beef production motivates the pasture formation.

14

Rooted in the Von Thiinen land use theory (Barlowe 1958), distance is widely used as
a proximate variable for transportation costs or market access as well, and thus included
as an explanatory variable for the farmland and forestland conversion. It is found that
some landholders are most likely to convert forest to farmland, where good access to
markets and favorable farming conditions make agriculture more profitable. Chomitz and
Gray (1996) reported that locations closer to urban markets have less remaining forest in
Belize, consistent with the findings that deforestation, or forest clearing, declines with the
distance to road (Nelson and Hellerstein 1997; Kaimowitz and Angelsen 1998), although
significant depletion may also be associated with greater distances in certain cases
(Mamingi et al. 1996). On the other hand, roads appear to diminish the negative impact of
high poverty levels on forest in southern Mexico, where poverty and deforestation are
closely correlated, and the relationship is more pronounced in more isolated areas
(Deininger and Minten 1997).

Technological factors not only are inputs in the grain production function, but also
affect land use changes (Shively 2001; Miiller and Zeller 2002). Shively (2001) tested the
effects of irrigation development in lowlands on forest clearing in adjacent uplands in
Philippines, while Miiller and Zeller (2002) examines the effects of the time (number of
years since the introduction of a technology) and scale of using a new agricultural
technology. To improve food supply for an expanding population on a stabilized land
base, productivity and production increase must come from technical changes, such as
increased intensity of farming and use of modern inputs (Agricultural and Rural
Development Taskforce 2004). Such effects, however, will be different between small

and large farms, and in short and long run (Cattaneo 2001). In the long run, agricultural

15

technology will make large farms, which have less capital and labor constraints, expand
their farmlands because higher productivity will induce higher profits from farming.
Angelsen and Kaimowitz (2001) summarized the studies of relationships between
agricultural technology and deforestation. They viewed and grouped technology-
deforestation links in different agricultural system and economic development contexts,
such as commodity booms (technology: new varieties), shifting cultivation (technology:
intensive farming, new varieties, improved fallows), permanent rain-fed cultivation
(technology: fertilizer, pest control, new varieties), irrigated intensive farming and
intensified cattle ranching (improved pasture technology).

All land use decisions are made within an institutional framework. Institutions
include property-right regimes, economic institutions such as subsidy or tax,
governmental polices, social and cultural context (Lambin et al. 2003). Miiller and Zeller
(2001) found that during 1975-2000, the rapid and more labor- and capital-intensive
agricultural growth in the central highlands of Vietnam, combined with the establishment
of protected areas and the discouragement of shifting cultivation, reduced the pressure on
forests, while at the same time increased crop productivity and incomes for a growing
population. Some (Mendelsohn 1994; Nelson et al. 1999; Angelsen 1999) emphasized the
importance of secure property rights on forest management. Other studies of
deforestation incorporate the effect of credit use and possession of forest title (Caldas et
al. 2002), or the effect of the property tax (Evans et a1. 2001). Table 2.1 summarizes the

variables discussed above.

16

Table 2-1 Summary of Human-Dimension Variables in previous LUCC Models

 

Category

Specific Variable

 

Demographic

I Population density (Mertens et a1. 2003);

I Rural labor growth, labor transfer, household change in a
life cycle (Lambin et al. 2003; McCracken et al. 1999);

I Frontier immigrants (Pfaff 1999; Walker et al. 2000);

I Migration to non-farming sector (Shively and Fisher
2004);

 

Economic

I Affluence (Turner 1993)

I Poverty (Deininger and Minten 1997 );

I Credit (Caldas et al. 2002);

I Crops & livestock market values
(Parks et al. 2000; Walker et al. 2000);

I Timber (Cropper and Griffiths 1994) and forest product
prices (Mertens et al. 2000) ;

I Crop production costs (Parks et al. 2000);

I Wage and input costs, income
(Kaimowitz and Angelsen 1998);

I Distance to urban market, distance to road as market
access and transportation cost proximate
(Chomitz and Gray 1996; Mamingi et al. 1996; Nelson
and Hellerstein 1997; Kaimowitz and Angelsen 1998;)

I Soil Suitability as land revenugroximate (Pfaff 1999);

 

Technological

I Time and scale of a new technology
(Miiller and Zeller 2002)

I Irrigation development (Shively 2001)

I Farming intensity and modern inputs (Agricultural and
Rural Development Taskforce 2004);

I New varieties, fertilizer, pest control, ranching technology
(Angelsen and Kaimowitz 2001)

I Application scale (Cattaneo 2001)

 

 

Institutional

 

I Property-right regimes, economic institutions such as
subsidy or tax, governmental policies, social and cultural
context (Lambin et al. 2003);

I Environmental protection policy (Miiller and Zeller
2001); -Secure property rights (Mendelsohn 1994; Nelson
et al. 1999; Angelsen 1999; Caldas et al. 2002);

I Property tax (Evans et a1. 2001);

I External distribution and trade environment (Kaimowitz
and Angelsen 1998)

I Political economy (systems of exchange, ownership and
controlmolitical Structure (Turner 1994);

 

17

 

2.2.3 Studies on China’s LUCC

The LUCC studies of China are mostly concerned with its long-term food security
because of arable land 1088 that is undermining China’s food production capacities. Some
are focused only on monitoring LUCC. For example, Liu et al. (2003) reported that
China’s primary forests in the northeast and southwest declined by 0.11 million ha in the
19908 due to commercial logging and conversion to farmland. Urban and rural built-up
area, driven mostly by economic growth, population increase, and changes in land-use
regulation, consumed 1.86 million ha of primarily fertile suburban farmland. Similarly,
Yeh and Li (1997) analyzed regional arable land losses induced by rapid urbanization and
infrastructure or industrial expansion, especially in metropolitan areas like Shanghai and
the Pearl River Delta in southern China. Others combined remotely sensed information
with hydrological model to evaluate the change of land suitability (Li et a1. 2001) and the
connection of runoff change with land use change (Weng 2001), or applied statistical
models to examine the effects of urbanization and industrialization on the change of
arable land area and land use intensity (Zhang et al. 2003).

Meanwhile, scholars have applied complex models to analyze the driving forces of
LUCC, despite the fact that most of their works are preoccupied with food production.
Taking one of the most important grain production provinces in Northeast China as their
study site, Wang and Otsubo (2001) examined how the agricultural input and investment,
and grain prices and exports have affected the cropland conversion to urban uses. Zhang
and others (2000) used county-level data to address the effects of grain and timber prices,
population growth, forest ownership, and cropland tenure on land allocation between

forest and other uses. Using different grid sizes (smaller grids are aggregated to larger

18

one) as the study units, Verburg and Chen (2000) developed multi-scale models to
identify explanatory variables for land use changes at the region and national extent.
Based on an input-output model, Fischer and Sun (2001) examined how human activities
affect land suitability for cropping and land productivity as well as the land requirements
under various scenarios of socioeconomic development in the future. Their work appeals
for sustaining efforts to furnish technological progress in land-based sectors and more
grain imports so as to reduce the pressure on cropland and water resources, especially in
the north. In addition, Verburg and Veldkamp (2001) used the CLUE model to analyze
the spatial and temporal dynamics of land uses at the regional level, and simulate the
near-future changes for land use pattern and crop distributions. Both the CLUE model
and the study by Verburg and Chen (2000) included several modules to illustrate how the
cropland changes in China are inter-related to both biophysical and human drivers, and to
predict the future cropland demand under various scenarios. However, these models are
built for analyzing the national cropland changes, and they do not include explicit
regional policy or institutional variables. Also, these models do not consider the effects of
environmental feedbacks on LUCC.

So far, little rigorous work has been done on the driving forces of the LUCC in the
upper Yangtze basin. LUCC studies along the Yangtze River are either descriptive, or
from the hydrologic perspective. While human disturbances on land use changes are
discussed more or less in these studies, few socioeconomic statistics and policy variables
are used to quantify how human activities affect the LUCC. Some describe the severe
deforestation and forest degradation, which are claimed to induce serious soil erosion and

be one of the major causes for the massive flooding in 1998 (Ma 1998; Yu et al. 1999;

19

Yin and Li 2001). Others investigate the effects of vegetation loss on the river runoff and
sediments (Higgitt and Lu 1999), lake shrink along the river (Du et al. 2001), and soil
loss from different land uses (Yang and Liang 2004). Remotely sensed data are widely
used in monitoring the land use changes, and variables representing natural conditions

such as geographic and climatic variables are considered.

2.3 Synthesis and Future Research

It is seen that dynamic or systematic models have been applied in studying the LUCC
driving forces of which various socioeconomic and institutional factors have been
incorporated one way or another. Causes of the LUCC have been traced down to
demographical factors (e. g., population structure and migration), economic development
(e.g., infrastructure, income, urbanization), market conditions (e.g., costs, prices, trade),
technological factors, institutional and political structure (e.g., property regimes and
policies), and biophysical environment. Research to model the LUCC has been carried
out at various scales, using data from statistics, field surveys and remotely sensed images.
Models vary from a single equation, which illustrates the relationship between one land
use change and its driving forces, to a dynamic system, which examines the interactions
of various factors in the land use process. Table 2.1 summarizes the international

literature reviewed in this study.

Nonetheless, a lot remains to be learned before scientists can fully assess and project
future role of LUCC in the functioning of the earth system and identify conditions for
sustainable land use (USGCRP 2003). For instance, limited efforts have been made to
examine the effects of technical and institutional innovation on mitigating the pressures

of population growth and ecological degradation. To overcome this challenge, future

20

studies must incorporate the relevant variables and account for their possible endogeneity
through developing more effective structural models of the LUCC within a spatially
explicit framework (Irwin and Geoghegan 2001). To that end, however, longer series data
are a prerequisite. It is also crucial to combine a realistic description of household and
community behavior with regional dynamics (Kaimowitz and Angelsen 1998) and to
couple land use change with biophysical factors so as to consider the feedback of the

LUCC biophysical responses on the drivers of land use (Veldkamp and Lambin 2001).

It is also necessary to advance the knowledge of the LUCC causes and consequences
in China. As stated by Liu et al. (2003), “. . .we have constructed a long-term monitoring
system on land-use change. In the future, we need to study thoroughly the impact of
human social and economic activities on land-use change at regional scales, and further,
study the effects of land-use change on the environment (p. 384).” Improved knowledge
of the complex dynamic processes underlying land-use change will allow more reliable

projections and more realistic scenarios of future development.

Using China’s upper Yangtze basin as the study site, this research intends to advance
the knowledge of the LUCC driving forces by taking the following two innovative steps.
First, it will construct a set of discrete choice models to examine how the potential
driving forces have affected the primary land uses as well as the various forest uses. It
will do so by incorporating more socioeconomic, policy, and institutional variables into
the models so that it can generate richer empirical evidence. concerning what factors have
driven the LUCC. Second, it will include human factors, agricultural technological
progress, environmental feedbacks and various resource facets into structural land use

models to illuminate the interrelationships underlying the land use process. In view of the

21

particular attention to both food supply and environmental protection, this research will
develop structural models to investigate the changes in cropland and forestland over more
than two decades. It is expected that the findings from these efforts will give rise to a
more comprehensive understanding of the LUCC in the study region and thus make a
valuable contribution to the making and execution of land management policy in China

and elsewhere.

22

Chapter 3 Land Use Changes in the Study Site

This chapter is devoted to describing the land use changes in the study site from the
mid 19708 to the late 19908. To be thorough, both spatial and temporal changes will be
discussed. The land use/cover data, derived from satellite images and land use maps,
cover five time periods - the mid 19708, mid 19808, late 19808, mid 19908, and late
19908. The chapter starts with a study site introduction, followed by a description of data
sources, data processing procedure and some data problems. Then, it depicts the regional
LUCC in detail. At the end of this chapter, cropland area derived from the remotely
sensed images is compared to official statistics to justify the use of data derived from
remotely sensed images in this study.
3.1 Study Site

The total area of the upper Yangtze basin is about 1.05 million kmz, accounting for
58% of the total Yangtze basin area. Because of the expansiveness of the upper Yangtze
basin, the study site is selected along the Jinsha River, part of the upper Yangtze with a
length of 2,290 km. Specifically the Jinsha River refers to the section starting from Yushu
County in Qinghai province, flowing across Qinghia, Tibet, Yunnan, Sichuan, and ending
in Yibin city of Sichuan province. Most of the upper Jinsha River is located in Yunnan
province, and the lower reaches of the Jinsha River constitutes the boundary between
Yunnnan and Sichuan provinces. The total area of the Jinsha River catchment is about
340,000 km2. Included in the study area are 31 counties1 fully located inside the Jinsha
River catchment (97.7°-104.8°E, 25.4°—32.7°N), with a total area of 140,000 kmz.

Nineteen counties are in Yunnan province and twelve in Sichuan province (Figure 3.1).

 

I See discussion below regarding the selection of sample counties.

23

Because of the difficulty in systematically matching data from different sources, no more

counties were possible at the current time.

Figure 3-1 The Study Region in Sichuan and Yunnan of China

 

  

 

1
Legend for Study Region (right) Qinghar I" Gansu
/I/ River ‘ ’ ‘IIIIIIIII I
[IIJIIII] Counties in Sichuan‘ ' 1| IIIIII
l I“ Sichuan
Counties in Yunnan I“! Yangtzi: River
\ III:5:3 "I " III?"
/1/ Provincial boundary 1:" II. {I}?
2323‘ l'
I. ‘1“ng
fl III" (3111211011
3:5 girl "l"§
first?
Yunnan

 

 

 

 

 

Map of China

 

 

 

24

Located in the transition region from the Tibetan Plateau in the west to hills and
lowlands in the east, the Jinsha River basin is well known for its sharp descent, fragile
geological structure, and severe soil erosion. The main stream descends by 3,280m, while
the elevation of the whole basin ranges from 6,140 m to 300 m. Among the 31 counties,
six have 70% of their lands at altitudes higher than 3,000 m, and 21 have more than 50%
of their lands at altitudes between 1,500 m and 3,000 m. The west part of the Jinsha basin
is characterized by high mountains and deep valleys, and 13 counties have land slopes up
to 60 degrees; among them, 10 have more than 10% of their lands with slopes steeper
than 20 degree.

The lower reaches of the Jinsha basin are known for severe moisture deficiency, high
evaporation rate, thin soil layer, and low vegetation cover. Moreover, the geologic
structure of the lower Jinsha basin is dominated by Triassic shale and sandstone, with
small proportions of granite, limestone, and Quaternary deposits (Lu 2005), which
weather rapidly in the subtropical monsoon climate and yield soils that are susceptible to
erosion and coarsening through the loss of fine particles. Uneven but concentrated
rainfall in conjunction with the steep slopes and fragile soils accelerate the soil 1088 along
the river. The Jinsha River basin is thus identified as the major sediment source to the
Yangtze River. The lower Jinsha stream, which accounts for only 11% of the Jiasha
basin, contributes 57% of its sediment (Pan 1999).

In addition to geological variation, the study site is also known for its cultural
diversity and poor economic development. The study site in Sichuan province covers four

counties in its Ganzi Tibetan Autonomous Prefecture (Dege, Baiyu, Batang and Derong),

25

seven counties in its Liangshan Yi Autonomous Prefecture (Muli, Huili, Huidong, Butuo,
Ningnan, Yinyang and Leibo), and one county in its Yibin city (Pingshan).

Ganzi Tibetan Autonomous Prefecture, located in the west of Sichuan province and
bordered by Tibet, Yunnan and, Qinghai provinces, is one of the most populous Tibetan
regions. Ganzi is the largest prefecture in Sichuan, occupying around 31.5% of the total
provincial land; however, poor infrastructure and weak capacity make this region’s
economy far behind other regions in the province. In 2002, the Gross Domestic
Production (GDP) of Guanzi was ranked second to the last in Sichuan and the per capita
GDP ranked fourth to the last (Sichuan Statistics Yearbook 2002). Even worse, the net
per capita annual income of rural households (900 yuan) ranked last, less than half of the
provincial average (2,100 yuan). The population density in Ganzi is the lowest in Sichuan
province. In 2002, it was six persons/ kmz, much lower than the provincial level of 175
persons/ km2 (Sichuan Statistics Yearbook 2002).

Liangshan Yi Autonomous Prefecture, the third largest prefecture in Sichuan
province, is a major Yi ethnical region in China. This prefecture, located in the southwest
part of Sichuan, is famous for its affluent mineral resources and diverse ethnical cultures.
The overall economic situation of this prefecture is ranked in the middle of the province,
but the development level varies across counties. Counties included in the study site are
those comparatively poor with rough topographical and unfavorable natural conditions.
The population density in Liangshan is the third lowest in Sichuan, with 68 persons/ km2
in 2002 (Sichuan Statistics Yearbook 2002). In both prefectures agricultural population

accounts for more than 85% of the total population, meaning that most of the local people

26

depend primarily on agriculture for living, which partially explains the low income and
poor development situation in these counties.

The 19 counties in Yunnan province belong to seven cities and prefectures. They are
Zhaotong City (Yongshan, Shuijiang, Jiaojia, Zhaoyang), Qujing City (Huize), Kunming
City (Luquan), Chuxiong Yi Autonomous Prefecture (Dayao, Yongren, Yuanmou,
Wuding), Dali Bai Autonomous Prefecture (Binchuan, Heqing), Lijiang City (Lijiang,
Yongsheng, Huaping, Ninglang), and Deqing Tibetan Autonomous Prefecture (Deqin,
Zhongdian, Weixi). Unfavorable natural conditions and poor transportation infrastructure
make this region isolated and suffer from a high incidence of poverty. The Jinsha basin
within Yunnan province occupies an area of 109,000 kmz, amounting to 28.6% of
Yunnan’s territory. But this segment of the Jinsha is the most eroded place in Yunnan
province, and the eroded area accounts for 39% of its total land base in the late 19808
(Wang 2003). The per capita cropland holding is less than 0.1 ha on average, with almost
a half on slopes. Of the 48 counties along Yunnan’s Jinsha section, 27 are national
poverty counties2 (Wang 2003), and 31.5% of the rural households had a per capita
income lower than 1,000 yuan in 2002 (Yunnan Statistics Yearbook 2003).

In short, sharp topographic variations and fragile natural conditions make the study
region difficult for infrastructural construction and poverty alleviation. Meanwhile, this
region plays a critical environmental role along the whole Yangtze River, because of its
location in the upper river basin and its high proportion of forests (see detail in the next

section) that serves important ecological functions. However, human activities have

 

2 The Chinese government initially defined the poverty line as per capita income below 200 yuan in 1985.
Based on inflation and other considerations, the figure has been adjusted upwards over time, reaching 1067
yuan in 2007 (China State Statistics Bureau 2008). A national poverty county is declared if a majority, but
not necessarily all, of the local population lives below the poverty line.

27

affected the land use patterns, resulting in resource overexploitation and poverty. Thus, it
is crucial to understand the human and natural causes of land use changes to benefit
sustainable resource management, environmental protection, and economic growth.
3.2 Data from Remotely Sensed Images

The land use data covering the mid-19708, mid-19808, late 19808, mid-19908, and
late 19908 are employed to investigate the regional LUCC from 1975 to 2000 in the study
site. Specifically, data for the mid-19708 (hereafter, 1975 or 1975 time point) were
derived from 1973-1977 Landsat Multi-Spectral Scanner (MSS) images, data for the late
19808 (1990) from 1988 and 1989 Thematic Mapper (TM) images, data for the mid-
19908 (1995) from 1995 TM images, and data for the late 19908 (2000) from 1999—2000
TM and Enhanced Thematic Mapper (ETM) images. A8 a common practice, in case
certain images were missing or of poor quality, those from adjacent years were used to
obtain the information for a given time point. The data were provided by the Chinese
Academy of Sciences (CAS)3. Due to the unavailability of quality images for the mid-
19808, the CAS scanned and digitized a land cover map for 1985 to generate the needed
data". The classification accuracy rate is 88% for 1975, 92.9% for 1990, 98.4% for 1995,
and 97.5% for 2000, respectively (CAS 2006)5. To unify the resolution from different
images (resolution for MSS, TM and ETM images is 80m, 30m and 15m, respectively),

the classified land use/cover data at each time point were re-sampled to a 100m-

 

3 Classification, interpretation and re-sampling of the remotely sensed images were all carried out by the
collaborator from the Institute of Geographic Science and Natural Resources, Chinese Academy of
Sciences. Land use classification and raster-format land use dataset were then provided by the collaborator.

4 It is acknowledged while the map has a fairly high resolution, this can cause problems of accuracy and
consistency.

5 Please see Appendix 3.1 for technical details of the classification method.

28

resolution raster—format dataset.6 This study then overlaid a county boundary map on land

use data to generate the corresponding county-level data, and the detected land use

conversion and modification were derived from overlaid land use data for adjacent

points.

Following the CAS protocol (CAS 2006), land uses/covers were classified

hierarchically into four primary categories (cropland, forestland, grassland, other land).

Forestland is further classified into four secondary categories (closed forest, shrub, open

forest, and other forest). Table 3.1 shows the definition of each primary and secondary

land categories.

Table 3-1 Definition of Land Classification

 

Primary
category

Secondary
category

Description

 

Cropland

Land for crops, including cultivated, newly plowed, fallow
and grass-crop rotated land; land for orchards, mulberry, and
agro-forestry but mainly for crop production; tide/beach land
with at least 3-year cultivation.

 

Forestland

Closed

Natural or artificially planted forest with canopy > 30%,
including timber, economic, and Shelterbelt trees.

 

Shrub

Woody vegetation less than 2 m and with canopy> 40%.

 

Open

Forests with canopy between 10 and 30%.

 

Other

Newly planted, or newly regenerated forest; nursery; and
orchard, mulberry field, tea field, and tropical evergreen
forest/orchard.

 

Grassland

Land with herbaceous types with canopy > 5%, including
shrub grassland for grazing and open-forest grassland with
canopy < 10%.

 

 

Other land

 

 

Including bodies of water, built-up, and unusable lands

 

Source: the Institute of Geographic Science and Natural Resources (IGSNR), CAS.

Notably, data based on such a classification facilitate the examination of both land

cover conversion and modification. The discrete representation of land cover has the

advantages of conciseness and clarity. It, however, has led to an overemphasis of land-

 

6 . . . .
Inforrnatron on resamplrng method and errors can be obtained from our collaborator of thrs research.

29

 

cover conversions (i.e., shifts from one primary land use/cover type to another), and a
neglect of land-cover modifications (i.e., more subtle changes that affect the character of
the land cover without changing its overall classification, reflected in the changes
between secondary categories) (Lambin et al. 2003). In reality, cover modifications may

be even more relevant to understanding changes for resource quality change and

environmental effects.

3.3 Temporal and Spatial Land Uses Patterns

3.3.1 Temporal Variation of Land Uses/Covers

Table 3-2 Land Use Pattern in the Study Region over 1975-2000

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. Changes
Lafiflffi?“ 1975 1985 1990 1995 2000 $5236
2000
Cropland 16,0369 16,146.2 15,617.0 15,6343 15,6072 -4297
Forestland 74,8737 75,3078 74,9898 75,5155 74,6645 -2092
-closed 39,748.2 41,858.7 40,147.3 41,815.4 39,606.4 -141.8
-shrub 20,9725 19,3797 20,6245 19,6564 20,9109 -61.6
-open 13,2579 13545.3 13,9008 13,7405 13,8454 587.5
- other 895.1 524.1 317.3 303.2 301.8 593.3
Grassland 42,4275 41,178.4 43,108.8 42,249.8 43,882.3 1,454.7
Other land 7,078.2 7,783.9 6,700.7 7,016.8 6,262.3 -815.9
Share
Cropland 0.114 0.115 0.111 0.111 0.111 -0003
Forestland 0.533 0.536 0.534 0.538 0.532 -0001
- closed 0.531 0.556 0.535 0.554 0.530 -0.001
- shrub 0.280 0.257 0.275 0.260 0.280 0
- open 0.177 0.180 0.185 0.182 0.185 0.008
- other 0.012 0.007 0.004 0.004 0.004 -0.008
Grassland 0.302 0.293 0.307 0.301 0.313 0.011
Other land 0.050 0.055 0.048 0.050 0.045 -0005

 

Note: The share of secondary forest classes is its share in the forestland.
Data Source: IGSNR, CAS.

 

Forestland, grassland, and cropland are the major land use types in the Jinsha basin.

From 1975 to 2000 these three land uses fluctuated around 53%, 30%, and 11%,

30

respectively. All other lands accounted for roughly 5% of the total land area. For each
time point during 1975-2000, the share of each primary land use varied by less than 1%.

Compared to 1975, the areas of cropland, forestland, and other land decreased in
2000, while grassland area increased. Moreover, the trends and magnitudes of variation
differed for different land uses. Cropland area declined by about 400 km2 from 1975 to
1990 and by 10 km2 from 1990 to 2000, but it witnessed a temporary rise in 1985 and
1995. In comparison, grassland area changed in the opposite direction; it increased by
about 700 m2 from 1975 to 1990 and 770 km2 from 1990 to 2000, but its temporary
decrease occurred in 1985 and 1995. Forestland area increased continuously from 74,900
lrrn2 to 75,500 lrrn2 between 1975 and 1995, but dropped significantly in 2000 to a level
even lower than that of 1975.

Closed forest is a major component of forestland, accounting for more than 50% of
the total forestland area, and shrub is ranked second with a share of around 30%. Closed
forest increased by 2,070 lrrn2 from 1975 to 1995, but decreased by 2,200 km2 from 1995
to 2000, leading to a net loss from 1975 to 2000. The trend of closed forest change was
similar to that of total forestland, but the former experienced a greater fluctuation than the
latter. The direction of change for other forest types was opposite to that of closed forest,
and thus the total forestland area exhibited only a mild change. Shrub declined slightly,
and its change for each point was always in the opposite direction to the change of closed
forest. In comparison, the area of open forest increased by about 600 km2 over the years.
Finally, other forest declined successively during the study period, ending with an area

that was only 36% of the 1975 level. Such findings validate the importance of studies on

31

land use modification, because changes in the total forestland area conceal the variations
of different forest components.

As indicated in Table 3.1, other land, including bodies of water, built-up land, and
unusable land, only accounted for 5% of the total land area. It decreased by 800 km2 from
1975 to 2000, equivalent to 11.5% of the 1975 level. Although this change constitutes a
small share of the total land, it might have significant environmental consequences. For
example, if such a change resulted from wetland encroachment, it would severely
deteriorate the wetland functions. On the other hand, the decrease of other land implies
that built-up land such as urban uses in the study region did not expand a great deal.

Generally speaking, the share (or area) for each land use (either primary category or
secondary category) did not vary much over the years. It is thus not surprising that
someone might doubt that misclassification or the absence of substantial land use changes
contributed to the slight variation of the share for each land use from 1975 to 2000. It
should be mentioned that in this section we only focus on the ‘static’ change of land use
patterns over the study period — whether the area of a land use varied over time. If we
look at the ‘dynamic’ change of land use patterns, however, a different conclusion might
emerge. For example, if the size of cropland converted to forestland is similar to the size
of forestland converted to cropland, the share of cropland or forestland at the adjacent
time points can be very close. However, land use changes did happen. Thus, the fact that
the share of each land use did not change significantly from 1975 to 2000 does not
necessarily imply that the regional LUCC did not occur. That is, the aggregate picture can
obscure the temporal and spatial variations of the land use/cover patterns. Section 3.4 will

discuss in details the land use conversion and modification.

32

3.3.2 Spatial Variation of Land Use Patterns

This section compares land use patterns between the two provinces and shows how
the land use patterns vary with altitude. Such a spatial analysis will reveal what cannot be

observed from the aggregate land use data.

3.3.2.1 Land Use Pattern Variation between the Two Provinces

The study region covers 61,600 km2 in Sichuan province and 78,800 km2 in Yunnan
province. Despite the fact that in both provinces forestland dominates the land uses,
followed by grassland and cropland, the share for each land use differed to a certain
extent. In Sichuan, forestland accounted for 48% of its total land area, while it amounted
to 58% in Yunnan. Grassland in Sichuan stayed around 37.5%, and it was 25% in
Yunnan. Compared to its 1975 level, forestland in Sichuan decreased by 270 km2 in
2000, while it increased by 70 km2 in Yunnan. Thus, the forestland decline in the study
region is mainly attributable to the decrease in Sichuan. From 1975 to 2000, grassland
increased by 620 km2 in Sichuan and 830 km2 in Yunnan. One should be warned, though,
that the increase of grassland area may not be a sign of environmental improvement,

because it could come at the expense of forestland or cropland.

33

.95 £me3 550m 585 some 5 55 3: 552 eoo 8.50m

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

emoo mvoo omoo #moo goo omoo omoo omoo 5wo.o mooo 5:2 .550
5mm.o mvmo _mm.o ommo ovmo vwmo m5m.o w5m.o 5omo V5m.o 5.5.580
wood wood moo.o mooo oooo wooo wooo wooo oooo foo .550 ..
vomo 55o oofio momo 32o _5~.o womo oo_.o o5_.o 52o come-
ammo vvmo omNo ommo Gmo micro vwmo Qmo vomo 5om.o 5:5 .
memo memo Nmmo mwmo Sumo 5om.o ono oomo Smo oomo come—o -
N5m.o 55m.o N5m.o omo _5m.o o5v.o owvo mmvo 59nd mwvo Hexagon
5m_.o wm_.o 52o memo Sumo w5o.o w5o.o w5o.o o5o.o o5o.o 555%er

chasm
oévorm 5._ov.m odwoxm owmvam Nae—NM 029m 529m ”Sore. o.wmm.m odwwfi 5.5— 550
mammrom 953.2 0505.2 ogwmfifi m.mom.o_ o.omo.mm mNofimN m; 5.3 o.oom.mm m.¢mo.mm 5:57.980
05¢ com 5.Nm w.wom ”mow mémm Rmmm Odom m.mmm mowv .55o .
5.mw5.w ~.oo5.w odmwrw o.mov.w mdomd wgooam vovoam o.~5o.m m.mmo.m m.moo.v some .
o.omo.: _.owo.: woomr: w.vmo.o_ m.m55.: oémmd m.o5m.w c.5mm.o o.mm5.w 053$ 5:5 -
oooosvm N.5om.mm odmonN o. ~N15N 5.Nmm.vm mooori vaNS 5.58.2 5.55.: ©6052 name—o .
_.5oo.mv o.wom.mv 5w5ormv 5.03.9» :3on m.5om.om v.2 _.om w; Sdm N. 5de ©6323 tangy—om
w.5 _ w.o~ w.¢mw.o_ oomwre Q53; _ o. :1; _ v.ow5.v m .35..“o m.oo5.v m.w5w.v N.mow.v c.5380
ooom moo. ooo~ mwg m5o_ oooN moofi ooe moi m5o_ mad—q no.2.

5535. 525%

 

 

5355 one 53:95 E .53 3 m53 So...“ £96053: 555 min 933—.

34

The land resource for crop production is more limited in Sichuan than in Yunnan.
Cropland accounted for only about 8% of the total land area in Sichuan, and it declined
by 100 km2 from 1975 to 2000. In Yunnan, cropland occupied about 14% of its total land
area, and it also declined by 330 km2 over the whole period of study. So, it seems that
both provinces did not witness an aggregate cropland expansion as assumed. Again,
provincial level data can obscure what happened at a finer scale like county, and this is
partly why models based on county-level data is more appropriate to identify the LUCC
processes and their causes.

Closed forest was a major component of forestland in both provinces, but Yunnan had
a higher proportion of closed forest than Sichuan, which implies a better quality of the
forest in Yunnan. In Sichuan, the increase of shrub and open forest did not offset the loss
of closed and open forest, which implies the conversion of forestland to other land uses.
In Yunnan, however, the increase of closed and Open forest surpassed the decrease of

shrub and open forest.

3.3.2.2 Variation of the Land Use Patterns with Altitude

The land use patterns also varied with altitude, which is not surprising because natural
conditions such as temperature, precipitation, and soil features change with altitude and
thus affect land uses. Here, the whole region is divided into three zones: low elevation
zone with an altitude less than 2,000 m, medium elevation zone with an altitude between
2,000 and 3,000 m, and high elevation zone with an altitude higher than 3000 m. The
land area amounts to 15,600 km2 in the low elevation zone, 61,600 km2 in the medium

elevation zone and 63,200 km2 in the high elevation zone.

35

The share for each land use in 1975 and 2000 at three elevation zones is reported in
Table 3.4. Figures in sections a and c are the shares of a certain land use in the total area
for that use, whereas figures in section b and d are the share of a certain land use in that
elevation zone. For example, section a shows that in 1975, 23.1% of the cropland was
distributed in the low elevation zone, 67.5% in the medium elevation zone, and only
9.5% in the high elevation zone. Likewise, section b indicates that in 1975 cropland
accounted for 23.6% of the total area of the low elevation zone, 22.7% of the medium

elevation zone, and only 2.4% of the high elevation zone.

Table 3-4 Shares for each Land Use at Three Elevation Zone

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Forestland Other

Cropland Grassland Total closed shrub open other Land

Low 0.231 0.084 0.110 0.116 0.122 0.077 0.076 0.020

; a Medium 0.675 0.341 0.466 0.456 0.378 0.644 0.310 0.207
7 High 0.095 0.576 0.424 0.428 0.500 0.279 0.614 0.773
5 Low 0.236 0.227 0.527 0.294 0.164 0.065 0.004 0.009
b Medium 0.176 0.235 0.566 0.294 0.129 0.139 0.005 0.024
High 0.024 0.387 0.503 0.269 0.166 0.058 0.009 0.087
Low 0.234 0.080 0.112 0.113 0.134 0.076 0.021 0.023

2 c Medium 0.671 0.350 0.466 0.462 0.359 0.648 0.127 0.158
0 High 0.095 0.571 0.422 0.425 0.507 0.276 0.852 0.819
0 Low 0.233 0.224 0.534 0.287 0.179 0.067 0.000 0.009
0 dMedium 0.170 0.249 0.565 0.297 0.122 0.146 0.001 0.016
High 0.023 0.396 0.499 0.266 0.168 0.061 0.004 0.081

 

 

Note: 1. The figures in sections a and c are the share of a land use in the total area for that
use, while the figures in b and d are the shares of a land use in the total land area in
that elevation zone.

2. Elevation is derived from the 500m-resolution DEM model.
3. Source for DEM: IGSNR, CAS.
The land use patterns in 1975 at different elevation zones were very similar to those
in 2000, but sections a and c show that at each time point land uses did vary at different

elevations. Most cropland was located in the medium elevation zone, grassland was

largely observed in the high elevation zone, forestland was mainly distributed in both the

36

medium and high elevation zones, and other land was concentrated in the high elevation
zone. In addition, it can be seen that closed forest and shrub were mainly found in both
the medium and high elevation zones, open forest was mainly distributed in the medium
elevation zone, while other forest was largely in the high elevation zone. That most of
other land use was located in the high elevation zone is probably related to the fact that
there is a large amount of unusable land (such as bare rock and barren soil) in that alpine
zone.

Sections b and d present the variation of the share for each land use in the three
elevation zones. It can be seen that forestland was the dominant land use in all three
elevation zones, and its share in the low and medium elevation zones was slightly higher
than that in the high elevation zone. The share of grassland was around 39% in the high
elevation zone, 16% higher than that in the low and medium elevation zones. Moreover,
the grassland share in the high elevation zone was 12% higher than the closed forest, and
also higher than the share of any other forest categories in all three zones.

Sections b and d can also help identify the regional LUCC over time. Cropland
decreased from 1975 to 2000 in all three elevation zones; grassland increased by 1% in
both the medium and high elevation zones, whereas it shrank in the low elevation zone.
In comparison, forestland increased in the low elevation zone, but decreased elsewhere. It
seems that in the low elevation zone, the major conversion was from grassland and
cropland to forestland, while in the medium and high elevation zones the conversion was
mainly from cropland and forestland to grassland. The increase of forestland area may
have resulted from a variety of ecological rehabilitation projects implemented along the

upper Yangtze in the past. These findings also implies that it is difficult to restore forest

37

in the highland once it is disturbed, once deforested land is often converted to sparse
vegetation.

While the land use patterns in Sichuan and Yunnan are different, these patterns did
not change much over time for either province or in any elevation zone. Once again, this
is simply a ‘static’ view of the land use patterns. The next sector will look into land
conversion and modification to illustrate the regional ‘dynamic’ LUCC.

3.4 Land Use Changes over Time
3.4.1 Land Conversions between Primary Land Use Categories

This section quantifies the temporal land conversions and modifications from 1975 to
2000. Land conversion is a two-directional process - for example, that conversion from
cropland to forestland and conversion from forestland to cropland may occur at the same
time, but at different locations. Because changes in opposite directions often offset each
other, the net change of a land use over time may not be that significant.

Temporal land use changes can be presented by a land conversion matrix that shows
in detail the changes for each land use during a certain period, including the magnitude
and destination of change. Table 3.5 is a land conversion matrix for the period of 1975-
1985. The first row indicates that 91.5% of cropland in 1975 was still cropland in 1985,
while 1.4% converted to grassland, 6.7% to forestland, and 0.3% to other land. In the
meantime, 1.8% of grassland was converted to cropland from 1975 to 1985, and 0.3% of
forestland and 7.2% of other land was converted to cropland as well. Tables 3.6, 3.7, and
3.8 list the land conversion matrixes for the periods of 1985-1990, 1990-1995, and 1995-

2000, respectively.

38

The extent of cropland conversion to forestland, grassland, and other land was larger
from 1975 to 1990 than thereafter. Between 1975 and 1990, about 10% of cropland was
converted, whereas between 1990 and 2000 only 4% of cropland was converted. The
destinations of converted cropland varied as well. More cropland was converted to
forestland than to grassland and other land from 1975 to 1985 and from 1990 to 1995,
while in the other two periods more cropland was converted to grassland. In terms of
conversions from forestland, grassland, and other land to cropland, forestland was the
largest source, except for the period of 1975-1985 when more grassland was converted to
cropland. Even though forestland converted to cropland accounted for only a small
amount of the total forestland area, such a conversion, if concentrated in ecologically
fragile counties, can still induce major environmental consequences. Unlike the coastal
region of China where urbanization and industrialization have encroached on a large
amount of cropland (Yeh and Li 1997), cropland converted to other land such as built-up
land was not that dramatic in the study region.

Grassland conversion to cropland, forestland, or other land was significant before
1990. Except for the period of 1975-1985 when a majority of grassland was converted to
other land, most of the grassland was converted to forestland. The conversion of
grassland to other land between 1975 and 1985 was comparatively large due to possible
overgrazing. Regarding the conversion to grassland, forestland was the largest source.
Poor regeneration and management after clear cutting might partially be responsible for
that conversion.

The conversion from forest to cropland, grassland, and other land accounted for only

a small amount of the total forestland. Meanwhile, the magnitude of change, either

39

conversion into or out of forestland, was larger prior to 1990, when it reached 2,400 km2
for each period. After 1990, the largest change was the conversion from forestland to
other uses at a magnitude of 1,400 km2 from 1990 to 1995. With the deepened economic
transformation, environmental protection was put on the agenda gradually and rapid
industry development lured rural labor to migrate out (Xu et al. 2007). These measures
alleviated pressures on forest resources. Moreover, substantial logging occurred in the
early 1980s when collective forestland was contracted to farmers who were afraid that
contracts would be rescinded soon (Xu et al. 2007). Later on, regulations were tightened
on the collective forest, and in some places contracts were even revoked. Thus, the
temporal forestland changes can be traced to the economic and institutional changes
taking place in the study region. Grassland was the largest source as well as destination of
the forestland conversion, and the forest reduction over time also largely resulted from its
conversion to grassland. Although other land took a very small share of the total land,
their changes were very significant, especially from 1975 to 1990. Only 83% of other
land stayed unchanged during 1975 —l985, and that figure dropped to 73% from 1985 to
1990. Conversion of grassland in both directions was the main reason for this change.
The fact that the amount of land converted into or out of a land use was similar for
each period explains why the static makeup of the land use was almost the same. The
conversion matrixes also indicate that the LUCC did take place for each period and often
in opposite directions simultaneously, despite the differences in major sources or
destinations for each period. Moreover, the land use pattern was relatively stable and the
scale of change was smaller in the 1990s. This might be partially related to the changes in

the overall macro-economic situation, as mentioned above. At the beginning of the

40

reform era, unstable and inadequate regulations caused uncertainties that might have
induced short-run land exploitation activities. Later on, the regional land use pattern

became more stable due to improved utilization and strict regulation.

41

.m<U £me: 558 5:5 :95 5 «55 5m: 55— 55 ooSom

 

0.35.5 m.wmo.m 5N2

w.5om.m5 mdvm
5555:“ mowm
353.2 0.2m

 

588.
m wo _

550 55555m 5:25.80 55380

 

2&2 .85 82

oofi 5.m5 N Nmm No 550

oofi mo Woo md 55 580»

oofi 5d 5m m.mo mo 555550

oofi Wm m; 5m woo 55380 32
530.5

33 550 55555.”: 52555.5 55380

 

€8.53 82

ooam 5:5 mag .55an .5533. 3 on: 2.9 Be...— 535280 553 .5 959::— 5:3 «0.5. 5-». 253.

.m<0 £me: 553 5:5 5:5 5 555 8: 5:3 55 850m

 

N.w5o.5 5.53% Wmvm

5.m5w.55 odfi

05%.? w.mm5._ 5.2m;

0.58.3 5.mm

 

530,—.
m 50 ~

5550 5538a 525—580 55380

 

2...; 3.5 m2:

2: 5% ....m no 2 5:5

2: mo 28 2 no .85.:

2: 3. 2m 28 w: @3225

2: no 3 E n; @5305 mg:
EB

53 550 5:5555m 555350 55380

 

2.833 32

mag 55. m53 .5935: .55:=< 3 3.: 5:0 8?... 53.5250 5:5..— .«e 9.595.— 555 «5.3. mum 25C.

42

.m<0 dZmQ ”ESQ 08:: :08 :m Saw 8: 2:2 :8 8::om

 

wbflofi «.mmmd ado

 

new? 3

22.3. 2o

38.: SN
as:

32

:35 28:88“: 95380 28380

 

me: .83 22K

2: w.ww v4 56 :0 550
2: o :3 .1 ed :38“:

2: o 5.0 mo md cam—$80

oo: _.o v: Nd 000 98380 moo:
:38.

550 93:88.: 98.3.80 38380
0:853 25m

39

 

2.3 can mag 5953 ..ofic:< 3 3.: 2:0 25.: 68.5280 ES.— .«c Gaugom 65:0 3.5: wé 033.

.m<0 .Mng :an BE: :03 2: 8% 3: 2:2 :8 venom

 

5856 Nwowd m6

 

”03.3. Nmo

Qwofimv 25mm

9206 _ v.2
3:8.

coo:

550 98:38“: 95380 238:0

 

$5: .8532

co: mda :0 Nd fio 850

2: _.o 33 md v.0 :88“:

cc: md m wdo :0 9:3380

OS _.o ad ad Nbo flaw—no.6 com:
ESP

32 550 98:38”: cam—$30 98380

 

€853 m8:

mg“ 65: .53 :8an .5533: 3 3.: 0:0 89¢ 63.3280 unaw— ua 359:5 .33 a9:< Wm 03:0.

43

3.4.2 Forestland Modification

As the major land use category in the study area, forestland is further classified into
four secondary categories. Since forest ecosystems play important environmental roles,
such a classification can better identify the changes in forest quality and direct attention
to possible forest degradation. Even though the conversion (in or out) for the whole
forestland class was not large, the modifications within the forest class show a different
picture.

The forest modifications for different periods are reported in Tables 3.9, 3.10, 3.11,
and 3.12. Modification varied for each period, and each secondary forest category had its
own features of modification. The modification between closed forest and shrub was the
dominant change within forestland, amounting to more than 1,000 km2 for each period.
Furthermore, changes from shrub to closed forest overwhelmed the opposite change for
the periods of 1975-1985 (Table 3.9) and 1990-1995 (Table 3.11). Overall, from 1975 to
2000 more shrubs were changed to closed forest than vice versa.

Modification from closed to open forest was a major factor for the closed forest
reduction. From 1975 to 2000, the net change between closed and open forest was a 450
km2 loss of the former, and the largest occurred between 1975 and 1990. The overall
closed forest reduction over the whole study period was just 140 m2, implying that a
conversion of about 300 km2 from other primary land uses to closed forest offset the
closed forest loss.

Other forest accounted for a small share of the total forestland, but its changes were
the most significant. From 1975 to 1985 (Table 3.9), 10% of it was changed to close

forest and 29% was converted to grassland, while no any other land was converted or

modified to other forest. From 1985 to 1990 (Table 3.10), about 85% of other forest was
changed to closed forest, while conversion and modification from other land to open
forest compensated for only 25% of the loss. Notably, the different source for the 1985
land use data might cause their consistency with the same data for other time points and

thus calls for caution in interpreting the land modifications.

45

.m<U £ng 558 095 some 3 8% 3: US: ..5 8.50m

 

 

 

 

m.moc.w_ S is 33; 33 0.8 o to 3 spam
3. E No 33. 2: M3 3; o 3% 550
M33 3. $3.2 So ”.3 no o moo to :25
was: mam :2; 2%.? 3a 3 3 3 moo 8.85 £2
£5 25 :25 886 use a Begum spam ago :25 886
CS. .85 82 €833 c9:

 

.53 can mwa :8an .3523. 5 rewoaaounsm ”EC :85 595—52 1558.8 m we Cuoouom 5:3 «91¢ 3A” 25¢.

.m<0 £ng qua 0:5 some 3 Saw 3.: 5:2 c5 8.50m

 

 

 

 

$3.: 3 c. a can? 3 ”.3 o to 3 seam
3 3% no 55 26 o 33 S no sec
9% 3 Qfifi 3o” 23 ....o o a. a o :25
OS} 3 me: $32 0.8 an o to is 886 m5
555 .550 5&0 @820 685 mm UofimEoM Edam 550 5&0 @820
es. .85 £2 €853 £2

 

mwam 6:: m3— 50.52. .3523. 5 bewouaoA—sm 2.0 895 35:52 ...—«53.8.,— ue Canyon .53 35¢ aim 25¢.

46

.m<U .Mme: 550m :28 «a Saw 3: was com venom

 

 

 

 

$8.: od 2w 3%; :0 ma o to 3 93m
3 33 no pm 23 o ”.3 3 MS 850
S no 338.2 ms 8 o o 23 S 8.5
8.82 mm as 28.2 3 G o No So 886 32
£55 850 88 885 8.3 8 Begum £55 850 88 886
£5 .83 82 9.883 82

 

.53 25 man 5953 .3523. 5 bewouaoésm oaO 295 35:52 2353...;— ue 3:098.”— 623 3.2.. ~7m 25d.

.m<U £me: 553 0:5 none 3 Saw 3.: 28— E.“ 850m

 

 

 

 

$8.: 3 2: 383 .33 8.3 o o m: 8%
mm 3% no 3: Q8 3 4.8 do 2. 850
2.2 mo 4.0%.: 3.2 :8 mo Q So 2 88
4.23 3 28 38.8 N8 3 o 3 M2 885 82
2:5 8:5 8.5 886 888 8 85588 3% 850 8.5 886
was. 8.3 32 2888 82

 

mag 23 33 :8an .3525. 5 .Cemfiauéam 2.0 29¢ 35:52 can—amoueh me @5935 2.3 no.3. :4” 352.

47

Examining the modification and conversion of secondary forest categories has
enabled a clear understanding of the forestland structure and its change, which explicates
potential forest degradation that might otherwise be ignored if only changes of aggregate
forestland were studied. Moreover, such an examination has also helped identify the
underlying sources of the overall forest changes. For example, the forest area decrease
during the period of 1995-2000 mainly resulted from the conversion of open forest out of
forestland, implying that marginal forests were more vulnerable to conversion.

Notably, aggregate land conversion and modification are moderate, given the
accuracy rate of land classification at each time point. It is possible that the local changes
would be obscured when just looking at the aggregate data. Thus, changes of cropland,
forestland and grassland for each county between 1975 and 2000 are shown in the Figure
3.2. The charts present the percentage changes of the land uses compared to their 1975
levels. Clearly, some counties experienced dramatic changes. Thus, it is meaningful to
investigate the driving forces of the LUCC at a scale as fine as possible, which is the

county in this case.

48

Figure 3- 2 Changes In Cropland, Forestland, and Grassland for Sample Counties

‘ Cropland Change % for Each County between 1975 and 2000

 

.

l 10%

l 0% .1. ,

E 40%

. m
-30%

l 40% if.

 

 

 

T -50% b "

 

 

 

 

Grassland Change % for Each County between 1975 and 2000

 

‘ 13% ~ 77-7

’ 14%

i 10% " "

l

1 61/0" #_,

mII I- I
--

;.%--EWI' "l  .22:I m-

 

 

 

I
,2

Source for land use data at 1975 and 2000:1GSNR,CAS

49

3.5 Comparing Cropland Data from Different Sources

Given the problem of farmland underreporting by local government entities in China
(Lambin et al. 2003), a comparison of data from different sources can help detect the
potential discrepancy and identify its causes. This comparison is also crucial to ensure the
credibility of an analysis of the agricultural economy, particularly the effects of technical
and institutional changes on productivity and the environmental consequences of past
land uses.

Table 3.13 lists the cropland ratios of the government statistics to those derived from
the remotely sensed images. A scientific consensus is that the cropland area in China is
underreported by around 30% (Crook 1991). Here in the upper Yangtze basin, the
government statistics were around 40% of those obtained from remotely sensed images.
The situation is similar for the two provinces, and the discrepancy varied with altitude.
Notably, some sloping farmland is not included as “cropland” in the government
statistics, but they are captured by the satellite images and interpreted accordingly.
Furthermore, cropland derived from the remotely sensed images also includes certain
orchards and agro-forest plots, which are not categorized as cropland in the government
statistics. In addition to explaining the discrepancy, these factors help highlight why the
cropland gap between the government statistics and data derived from the remotely
sensed images is so large in a mountainous region like the study site. of this research.
Cropland area derived from remotely sensed satellite images, instead of government

statistics, will be used in modeling the driving forces in the next two chapters.

50

from the Remotely Sensed Images

Table 3-13 The Ratio of Cropland from the Government Statistics to That Derived

 

 

 

 

 

 

 

 

. Altitude
Year Overall Sichuan Yunnan Low Middle High
1975 0.385 0.382 0.387 0.339 0.394 0.418
1985 0.385 0.404 0.373 0.348 0.375 0.450
1990 0.389 0.397 0.384 0.326 0.396 0.444
1995 0.393 0.402 0.388 0.340 0.394 0.452
2000 0.402 0.399 0.404 0.344 0.413 0.442

 

 

 

 

 

 

 

Data Source: IGSNR of CAS, and Sichan and Yunnan Statistics Yearbooks.

51

 

Chapter 4 Analyzing the LUCC Drivers Using Discrete Choice Models

As mentioned in the literature review, it is commonly hypothesized that land use
changes are determined by biophysical, socioeconomic, and institutional factors (Irwin
and Geoghegan 2001; Kaimowitz and Angelsen 1998), and the logistic regression is
widely employed to examine the effects of these driving forces (Irwin and Geoghegan
2001). Such a regression method, however, cannot capture interactions and feedbacks in

the LUCC processes.

The analytic framework of induced technical innovation, developed by Davis and
North (1971) and Ruttan and Hayami (1984), carries greater relevance and richer
substance in the land use/cover change analysis, especially for a developing country like
China. These scholars point out that technological innovation and resource status are
interrelated in determining the long-term economic development and social changes.
Limited land resources, population expansion, and rapid economic development impose
challenges on sustainable resource utilization to developing countries, and technological
progress might the key for increasing land productivity to satisfy both food demand and
natural resource conservation. Building on this framework and advances in spatial social
science, this study will examine the driving forces of LUCC in the Jinsha river basin
using two types of models: fractional logit models to examine the changes of each land
use/cover share in this chapter, and structural models to examine the cropland and forest
area changes in the next chapter, which incorporates the interaction of technical changes

and environmental feedbacks.

52

4.1 Conceptual Framework
LUCC in the study region is expected to be driven by a number of factors, including
market signals, demographic factor, economic development of non-farming sectors,

institutional and policy factor, and biophysical conditions. Economic models typically
assume that a landowner allocates a parcel of land of quality ] at time t to the use that can
generate the highest net return over time. For example, the net return for a parcel of

cropland of a certain quality at time t, W? , equals the net present value of a stream of

crop revenues minus cultivation costs, plus the present value of the land salvage value

(Ahn et a1. 2000). The net return for a tract of forestland of certain quality at time t, Wf ,

is the net present value of a series of timber harvest rotations, determined by stumpage

prices and yields by species (Munroe and York 2003). The land will be allocated to
forestry if W f 2 W? , otherwise to farming (Plantinga 1996). Thus, the market signals,

such as prices and costs of products from various land uses, are widely used as the LUCC
drivers.

However, the effects of price signals on the LUCC might be indeterminate in this
study. First, product prices in China were historically distorted. For example, the grain
procurement prices facing producers were controlled and depressed by the government,
and their effects on land use decisions might not be clear and significant (Lin 1992). For
forestland, the state-owned forest is managed by the government and its timber
production is mostly subject to an annual quota, instead of market price. The procurement
prices faced by farmers or the grass-roots administrative unit who manages the collective

forest were usually lower than market prices, and various taxes and fees greatly reduced

53

farmers’ earnings from timber harvest (Yin 1998). It is thus reasonable to expect that the
forestland changes were not significantly affected by price variable as well. Secondly,
since some lands in a county are still controlled and managed by the state or local
government, decisions on the use of such lands probably cannot be explained by the
market signals. Therefore, this research incorporates other factors to explain the LUCC in
the study region.

Land use decisions are usually constrained by the local and national institutional
environment (Lambin et al. 2003), and in China these decisions are even more influenced
by the government regulations and policies (Zhang et al. 2003). In general, cropland,
forestland, and grassland are not allowed to be converted to other uses, except in the case
of conversions induced by government projects such as the Sloping Land Conversion
Program that requires the conversion of sloping cropland to forestland or grassland, or
infrastructural construction. For cropland, farmers sign contracts with local governments
to manage the land, and these contracts are valid for decades and seldom adjusted in
order to provide “stable” use rights. In forestry, timber harvest and distribution used to be
controlled by the government, and the state-owned enterprises and collective forest
entities had little motivation for managing the forest so as to maximize the net returns
(Yin and Newman 1997). The food self-sufficiency policy and forest tenure arrangement
are two other major political-institutional factors that have affected the land use patterns
in the study region (Yin et al. 2003). The former, reflected in grain procurement quota,
encouraged cropland expansion on slopes previously covered by forest or grassland (Xu
et al. 2007); the latter, if clearly defined and enforced, forms the basis for forest

management that is at least stable.

54

Moreover, economic development, especially industrialization, in China plays an
important role in affecting the LUCC in multiple ways. First, industrial development can
induce more land conversion from non-industry use to industry use. Such a trend has
already been observed in the more deve10ped, coastal region of China (Yeh and Li 1997).
Second, industrial development may promote migration of surplus rural labor to off-farm
sectors, which will reduce natural resource exploitation such as deforestation and
cropland encroachment, and thus help conserve the environment. Third, Ioff-farm
opportunities will alter the opportunity cost for rural labor, which can influence labor
availability for production activities like extensive farming (McCracken et al. 1999) and
then change the land use pattern. Finally, off-farm opportunities can improve farmers‘
income and enhance their abilities to adopt better farming technology, which will
increase cropland productivity to meet food and livelihood demands and therefore reduce
resource overexploitation. All these considerations call for the inclusion of some
measure(s) of industrial deVelopment in the analysis. In addition to the afore-mentioned
institutional and economic factors, demographic and biophysical variables are also
incorporated into the LUCC models. Next section will provide a description of the
variables used for the modeling.

Modeling the aggregated data of land use/cover at the county level can be thought as
that the land in a county is allocated to different uses subject to land quality, availability,

and factors that would influence the land uses (Miller and Plantinga 1999; Ahn et al.

2000). The area for the use k in the land of quality class j at time t can be expressed as:

him = 113.4111an

55

where H it is the total land area with quality j at time t, and X it is a full set of
decision or explanatory variables that affect the decision for each land use type (k=l ,. . .).

If the total land area for a county is H (t ) = 2 H j(t ) , the sum of lands with different
j

quality j, and then the optimal share of land use k for this county at time t is defined as:

It: 1 at:
X,t =— h- H't,Xt
pk( ) HO)? jkt( J )

where p k is the optimal share of land use k in the county. First, note that the land with a

certain quality can be allocated to all different uses. But, the share of each land use differs

across lands with different qualities or land characteristics. Second, the share of land use
k is not only a function of variables that affect land use k, but also a function of variables
that affect the return function for each of the land uses. Third, the share for each land use
will change over time, which is actually caused by changes in those X variables.

4.2 Empirical Model and Estimation Method

Following the theoretical model for land allocation, the share of a land use can be
k _ k k
defined as: yit "' pit (Xit) + 8"
, k , . .
where y it and P it are the observed and expected share of land use k 1n county I at

trme t; and 8 it rs the independently and 1dent1cally distributed error term for county I at

. k .
time t. The sum of y it or pit equals one when the range of k covers each land use 1n

county 1' at time t.

56

It should be noted that either y or p is bounded between zero and one, implying that it

is not appropriate to express y or p as a linear combination of explanatory variables and
thus to estimate it with a conventional linear estimation method, as doing so could

generate a fitted value of y or p that falls outside the unit interval (Woodridge 2002). To

avoid this problem, p can be modeled as a logistic function:

= exp(X,-t .Bk)
1+ eXP(Xit ,Bk)

 

k k
E(yit/Xit):pit

and then the observed land share y for use k is just expressed as:

k: CXp(th,Bk) +8};
" 1+exp<xitflk> "

 

where ,6 is the coefficient vector, and X is a set of explanatory variables. The above
specification, called fractional logit model, ensures that predicted values for y ranges
between zero and one (Papke and Wooldridge 1996). It is assumed that 8 satisfies a

logistic distribution. Note that while ,3 gives the sign of the partial effect of an

explanatory variable on a land use, its magnitude cannot represent the partial effect of an
explanatory variable on the dependent variable (Woodridge 2002). The partial effect for

any explanatory variable X, can be evaluated using the estimated coefficient and a certain

value (such as the mean) of X], with the function of fl/g(x,3), where

g<xm=exp<xfl>/[1+exp(xfl>]2.
A popular approach to coefficient estimation is to transform the model so that the log-

odds of the dependent variable, log[y|(l-y)], has a conditional expectation on the linear

form of explanatory variables: E(log[y|(l-y)]/X)= Xfl. Using log-odds as a dependent

57

variable in a linear model ensures that the log-odds ranges over all real values as y ranges
between zero and one. Such a transformation leads to coefficient estimation by Ordinary
Least Squares (OLS) method. But such a transformation has two drawbacks. First, it
cannot be used directly if the dependent variable takes on boundary values, zero and one.
Second, it is difficult to interpret the coefficients, because without further assumptions it
is not possible to recover how y is expressed by explanatory variables (Wooldridge
2002). To avoid these two problems, the fractional logit model is directly estimated by

the quasi-Maximum Likelihood Estimation (MLE) method that ensures predicted values

for y in (0,1) and that the effect of any X on EQ/LX) diminishes as Xfl—>0. The quasi-MLE

approach can consistently estimate ,3 when E(y|x) is expressed as a logistic form.

Meanwhile, the fully robust variance operation of the econometrics software takes care of
heterokastacity and serial correlation in variance. Since this 'test discovered that only

contemporary exogeneity assumption is satisfied, as discussed later, which means that y,

has feedbacks on X,+ 1 so that X,’s are only uncorrelated with 8, but probably correlated

with 8,+1 or 8* 1, a pooled quasi-MLE method will be employed on a panel dataset,

instead of using the methods of unobserved effects.

Another issue for estimation is the identification problem. For example, in this study
the land is classified into four primary categories, which indicates that a system of four
equations will be estimated. The dependant variable for each equation is the share of one
land use, and explanatory variables are the same for each equation. The link among four
equations is that the sum of dependent variables equals to one. To ensure the
identification of these equations, only three equations are estimated and the partial effects

of explanatory variables on the fourth land use equals to unity minus the sum of the

58

in the model to assess the effects of explanatory variables on land degradation, but the
sum of dependant variables should be restricted to unity. For example, if the share for
close forest and the share for shrubs are taken as dependant variables as well as farmland
and grassland shares, shares of the other two forests are then combined with the share for
other lands as the fifth dependent variable. Such an operation guarantees that the
summation of dependent variables equals one.

4.3 Variable Selection and Data Description

The dataset, covering 31 counties over five time points from the mid 19705 to the late
1990s, is used in the fractional logit model to elucidate the LUCC drivers. As described
in the conceptual framework, dependant variables are the share for each land use, and
hypothesized driving forces include demographic variable, market signals, non-farming
industry development, institutional factor and biophysical conditions. It should be made
clear that in China, counties are the basic administrative units where national policies are
executed and socioeconomic statistics are gathered. Additionally, the county-level
administrative boundary maps are readily available. As such, the disaggregate
observations at the county level are taken for this study. This section will first explain
why specific variables are selected, and then describe how each variable changed over
time.

The theoretical model indicates that land use decisions at time t are affected by the
net economic returns from various land uses at that time point. Calculating the net return
for each land use requires a lot of information, including the yield, price, cost, and
discount rate. In this study, the provincial procurement price indices for grain, timber,

and livestock are used approximate economic returns from cropland, forestland, and

59

grassland. At each time point, counties from the same province have the same price
indices. Although provincial data obscures variations across counties, they are still used
in the model due to the following considerations. First, provincial statistics are the only
available source for price information in this study. At least, these price indices reflect the
overall trend of price change for each product. Second, procurement prices faced by land
users or managers were determined by the government. Procurement price for the same
product (say, grain) probably differed across counties, but the price variation from one
year to another is similar for all counties in a province.

Moreover, no lagged price indices are used as explanatory variables because it is
assumed that the contemporary, instead of the long-term, economic return affects land
use decisions. That is, cropland reclamation could be induced by contemporary price
increase, but not by the long—term benefit consideration because of the insecure tenure for
the newly reclaimed cropland. For forestland, both the state-owned forest enterprises
(they managed the state forests) and the collect forest managers (either community
organizations or farmers who signed contracts to manage the collective forests) had little
incentive to consider the long-term forest management or harvest decisions, due to the
governmental control over timber production, distribution, and procurement prices. It did
happen that the harvested timber was more than the planned amount when market prices
went up. Still, regeneration for the future benefits was largely ignored. Grassland
accounts for a considerable share in the study region, and grazing plays an important role,
especially in the ethnic communities, as a means of providing necessary protein and
improving income. Similar to the cases of cropland and forestland, decisions based on the

long-term return are not quite feasible for analyzing the grassland changes. This is

60

because grassland was mostly controlled by the local collectives, making it hard for
herders to use the land efficiently. In short, given that the prices tended to be controlled
and distorted by the government, they could not lead producers to make long—term land
use decisions and thus their effects may not be significant.

Costs for farming, forest, and livestock production at the county level are not easily
identifiable in existing statistics. Thus, the approach used by Chomitz and Gray (1996) is
modified and applied in this study. These scholars pointed out that the distance of a
parcel of land to roads, representing market access, affects both output and input costs
and thus the land use pattern. In this study, road density in a county is used to reflect
transportation cost and market access. It is assumed that farmers in a county with higher
road density have better market access and thus more land for farming.

As explained earlier, the local economic development also has great impact on land
use. Industrial development, represented by the ratio of industrial output to agricultural
output, is defined as one of the likely driving forces. In addition, decisions on land use
are also impacted by social and institutional factors. Demographic variable is widely used
as a LUCC determinant (Mertens et al. 2000). Population increase in China called for
broad attention on issues of food safety and resource protection, and is always deemed as
the main reason for cropland expansion on to forestland or grassland and for forest and
vegetation degradation (Yin and Li 2001). This study uses population density to examine
the demographic effects on the LUCC. It is expected that higher population density will
lead to more cropland at the expense of forestland and/or grassland.

The food sufficiency policy and forest tenure arrangement are regarded as two major

political—institutional factors affecting the land use patterns. It is believed that the food

61

sufficiency policy tacitly encourages cropland expansion on slopes previously covered by
forest or pasture. For a long time, satisfying the grain demand with local production was
one goal of the local governments. Grain procurement quota is used in this study to
represent this policy. It is assumed that a decreasing grain procurement quota is a sign of
relaxing food sufficiency policy and can reduce cropland use and encourage restoration
of vegetation covers.

The land tenure system plays an important role in land use decisions. Secure land
tenure guarantees the long-term returns from lands, and encourages a land user to invest
more and conduct sustainable management in order to gain the long-term benefits. In
other words, clearly defined land ownership is the basis for efficient and stable land uses.
In the study area around 30% of the forestland is owned by the state, and 70% by the
collective. Forest that is far from the village residences and of good quality was
commonly allocated to the state-owned forest enterprises for management. These forests
were seldom converted to other uses due to their stable and clear ownership, although
forest quality degradation happened to some extent because of over-harvesting and poor
management. However, for some sloping lands that belong to the collectives or those
without clear ownership, cropland reclamation and forestland or grassland loss occurred
more often. Thus, the share Of state-owned forest is employed in this study to represent
forest tenure stability. A higher percentage of state-owned forest represents more stable
and clearer tenure arrangement, which is more likely to maintain the forest cover and
reduce forest depletion or degradation. It is not suggested, though, that the state-owned
tenure arrangement is superior to the collective, and it is only assumed that stable and

clearly specified land tenure benefits sustainable forest management.

62

Natural factors affect land allocation (Lambin et al. 2003), and many variables have
been used in LUCC model (Verburg and Chen 2000; Miiller and Zeller 2002), including
soil suitability, elevation and slope. Due to a combination the huge variation of
biophysical conditions in the study region and limited information, the average elevation
for a county is selected to represent soil features, temperature, and other biophysical
conditions that have effects on land use pattern. Table 4.1 summarizes all variables used

in the fractional logit model.

Table 4-1 Summary of Variables Used in the Fractional Logit Model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Measurement 1975 1985 1990 1995 2000
Grain Price Index Index(1990=1) 0.587 0.669 0.972 1.619 2.035
Log Price Index Index( 1 990:1) 0.589 0.531 0.927 1.046 1.294
Livestock Price Index Index(l990=l) 0.860 0.863 1.085 1.873 2.218
Industry/Agricultural
Output 0.320 0.403 0.443 0.768 0.762
Highway Rate Km/ha 0.0009 0.0014 0.0016 0.0021 0.0031
Population Density Person/km2 0.66 0.73 0.78 0.82 0.86
Per capita Grain Quota Ton/person 0.030 0.028 0.016 0.016 0.014
Share of
State-owned Forest 0.338 0.331 0.322 0.319 0.314
Elevation Meter 3070 3070 3070 3070 3070
Year Dummy l 1 1 l 0
Province Dummy =1 if the county is in Yunnan

 

Note: 1. Grain, log, and livestock price indices are provincial indices.
2. Industry/Agricultural Output is the ratio of industry output to agricultural

output.

 

3. The variable value at each time point is the average value from the adjacent
years whose range is the same as land use/cover data. For example, the land
use/cover data in 1975 is derived from the remote sensing images of 1973—
1977; so correspondingly, data for each explanatory variable in 1975 is the
averaged value of 1973—1977.

4. Highway data is missing for some counties at some time points. Specifically,
there are two counties without highway data, five counties with data at only one
time point, and one county with missing data at three time points. Altogether,
there are 33 missing values for highway data.

5. Grain quota data is from the local grain bureau, and the state forest share data is
from the local forest bureau. Elevation data is derived from DEM that is
provided by CAS. All other data come from local statistics.

63

All the provincial price indices increased continuously from 1975 to 2000, and the
trend of a price index should reflect that of the real product price. Grain and livestock
price indices rose sharply in the early 1990s at an annual rate of over 12%, while the late
1980s witnessed the fastest increase of log price index. Compared to their levels in 1975,
livestock and grain price indices in 2000 tripled, and log price index doubled. Given its
relatively low input cost, growing livestock price over time partially explains why animal
husbandry was preferred as a means of improving incomes. It seems that the relatively
low level and growth rate of log procurement price did not reflect the increasing lumber
demand. This is why it is assumed that farmers had no much incentive for engaging in
long-run forest investment and management; instead, they logged more when the timber
price went up or when they had access to the forest to capture the immediate profits.

Industry development is represented by a variable of the ratio of industry output to
gross agricultural output. Agriculture (including farming, forestry, livestock, and sideline
occupations) is the major economy component in the study region. Gross agricultural
output, an indicator of agricultural development, was always higher than industry output
from 1975 to 2000. The gross agriculture output (valued at the 1990 constant price)
increased at an average annual rate of 7% from 1975 to 2000, and the annual growth rate
in the early 1990s even reached 10%. However, the magnitude and growth rate of each
sub-sector of agriculture differed. Farming dominated the agricultural economy, with at
least 55% of the total agricultural output. Livestock husbandry developed more rapidly
than farming and forestry and its share increased continuously from 18% in 1975 to about
35% in 2000. Output from the forest sector was comparatively small, with about 8% of

the total agricultural output, and its share did not change a lot over time. However, output

64

from the forest sector was calculated based only on the market values of forest products,
excluding non-market products and services of the forest ecosystems. Despite the
dominant status of the agricultural sector in the whole economy, its share declined
because of the rapid growth of the industry sector. In 1975, the ratio of industry output to
the agricultural output was 0.3, and it increased to 0.4 in 1990 and further soared to 0.76
in 2000. Compared to that in 1975, industry output in 1990 more than doubled, and in
2000 increased by 10 times. It is obvious that the annual growth rate of industry output
was much higher than that of agricultural output.

Population density, total population divided by county area, is used in the model in
order to remove the effect of county size. Total population kept increasing over time, but
at a declining rate. The annual population growth rate was 1.3% between 1975 and 1985,
then down to 0.83% in the late 1990s. Correspondingly, the population density also rose
over time at the declining rate. Local statistics also indicate that the percentage of rural
population started at 92% in the mid 19708, down to 87% around the late 19803, but was
back up to 90.5% in the late 19908. Moreover, the share of agricultural labor among the
total population increased steadily from 36% in 19703, to 41% in the 1980s and then up
to 46% in the 19903.

Highway rate, the ratio of highway length to county area, is the variable used in the
model. China’s statistics provide a standard definition of highway, and thus there is no
discrepancy for highway data from different provinces. It is assumed that the higher the
highway rate, the lower transportation costs and the better market access for that county,
and the more cropland is likely allocated. Regarded as an effective way to alleviate

poverty and promote economic development, large scale road construction has been

65

taking place since the late 19703. As a result, highway length doubled from 1975 to 1995,
and doubled again within five years in the late 1990s. The highway rate tripled from 0.01
in the early 1980s to 0.03 in the late 19908.

Part of grain procurement quota was for paying the agricultural tax, and another part
was mandated to sell grain to the state at the procurement price. Grain procurement quota
declined gradually, and in 2000 it decreased to 60% of the 1975 level. Because of such a
large decline, per capita grain quota (grain quota per rural population) also decreased
dramatically, down from 30 kilograms (kg) in the mid 1970s to around 14 kg in the late
1990s. The decrease of grain quota over time implies that the food sufficiency
requirement became gradually out of date as the agricultural produce market became
more deve10ped. It is not necessary to meet food demand by local production for regions
like our study site where cropland is very limited. Instead, farmers can specialize in their
production with other less scare local resources and thus capture a comparative advantage
in the marketplace. Since 2003, China has abandoned the quota-based agricultural tax
nationwide.

The share of state-owned forests in Sichuan is around 50%, while in Yunnan province
it is just 20%. This share decreased slightly over time in both provinces due largely to
two reasons. First, disputes about land ownership were gradually resolved between local
government and communities, with the former having transferred some forests to the
latter. Second, afforestation on degraded collective mountains and conversion of sloping
cropland back to forestland resulted in an increase of collective forestland. Thus, the

percentage of the state forest declined.

66

Therefore, the empirical model incorporating specific explanatory variables can be
written explicitly as:

Y,-, = f (GPu, FP,-,, LP“, IND“, POP,-,, ROAD“, 0Q“, SF ,3, Eu) +8"

where Y represents cropland share, forest share, or grassland share; GP is the price
index for grain, FP is the price index for logs, LP is the price index for livestock; IND is
the ratio of industry output to gross agricultural output; POP denotes population density,
ROAD denotes the highway rate, GQ represents per capita grain quota, SF represents the
share of state-owned forests, and E denotes the elevation. Province and year dummy
variables are also included in the model to control the variation across provinces and over
time. The model is estimated with STATA software. Exogeneity test indicates that all
variables can be taken as exogenous ones, which simplifies the estimation procedure.

Table 4.2 lists the expected sign of coefficient for each equation in fractional logit model.

Table 4-2 Expected Signs for Variables in Fractional Logit Model of

 

 

Primary Land Uses
Cropland Forestland Grassland

Grain Price Index + _ _
Log Price Index — + _
Livestock Price Index + — +
Industry/Agricultural — + +
Output

Highway Rate + _ _
Population Density + _ _
Per Capita Grain Quota + — _
Share of State-Owned — + _
Forest

Elevation _ + ...

 

Note: ‘+’ (‘—’) means that the share of land use is expected to increase (decrease)
with the increase of explanatory variable.

67

4.3 Estimation Results

Table 4.3 lists the estimated results for the share equations of cropland, forestland,
and grassland, including coefficient estimates for province and year dummies. It should
be noted that the coefficient for each variable is the corresponding elasticity, but not ,8.
This way of presentation can clearly indicate the extent of the driving forces’ effects on
each of the land use/cover types.

In general, the signs of many coefficients are as expected and statistically significant.
Test of overall fit (Pearson Chi Square) for each model shows that we cannot reject the
hypothesis that the distribution of predicted value is the same as that of observed one. It
can be seen that the coefficient of the provincial dummy is highly significant, indicating a
sharp difference across the geopolitical boundaries. The results also indicate that the
effects of explanatory variables differ on different land use types. For example, grain
procurement quota significantly affects the share of cropland, but it is hard to reject
statistically that its effects on forestland and grassland are negligible.

Only two of the price index coefficients are statistically significant. First, grain price
index has a significant negative effect on the change of forest share; a 1% increase of
grain prices index can reduce the forestland share by 0.37%, holding other variables
constant. This shows that seeking short-term farming profits from the increased grain
prices indeed result in encroaching on forestland. Second, the effect of livestock price
index on cropland share is positive at the 10% significance level; a 1% increase of
livestock price index can result in a 0.94% increase of the cropland share when
controlling other variables. Given that some crops are grown for feeding domestic

animals, it is plausible that a higher livestock price drives more feedstock production. It is

68

Table 4-3 Estimated Results of Fractional Logit Model for Primary Land Use

Note:

 

 

 

Explanatory Variables Cropland Forestland Grassland
Grain Price Index -0.519 -0.367 0.357
(0.358) (0.180)“ (0.342)
Log Price Index 0.011 -0.087 0.067
(0.101) (0.058) (0.089)
Livestock Price Index 0.944 0.192 0.087
(0.565)* (0.368) (0.694)
Industry/Agricultural Output -0.047 0.068 -0.114
(0.01 1)*** (0.027)** (0.059)*
Highway Rate 0.215 0.018 0.105
(0.077)*** (0.054) (0.140)
Population Density 0.480 -0.335 0.157
(0.028)*** (0.082)*** (0.134)
Per Capita Grain Quota 0.037 -0.005 -0.015
(0.020)* (0.01 1) (0.034)
Share of State-Owned Forest -0.132 0.116 -0.133
(0.085) (0.072)* (0.107)
Elevation -1 .232 -0.633 1 .222
(0.128)*** (0.139)*** (0.335)***
Province Dummy -0.162 0.124 -0. 157
(0.071)** (0.039)*** (0.064)***
1975 Dummy 0.205 -0.043 0.110
(0.113)* (0.071) (0.142)
1985 Dummy 0.162 -0.039 0.097
(0.105) (0.065) (0.130)
1990 Dummy 0.132 -0.032 0.086
(0.081) (0.052) (0.104)
1995 Dummy 0.068 -0.014 0.040
(0.041)* (0.026) (0.050)
d.f. 107 107 107
Log-Likelihood -34.212 -54.903 -47.501
Pearson Chi Square 1.053 5.222 6.221

 

l. “*”, “**” and “***” represent 10, 5 and 1 percent significance level, respectively.

2. Numbers in parentheses are standard error of the coefficient.

3. Observations are 122, because of some missing values for highway variable.
4. p-values for Pearson Chi Squares are all 1 for each land use. When dependent
variables take categorical values, the Pearson statistic is used for measuring a

model’s goodness of fit.

69

also noticed that there exist high correlations between price indices, which affects their
estimated effects (see Appendix 4.1 for more detail) and caution is needed in interpreting
the effects of livestock price change. The insignificance of other price variables partially
proves my prior claim that the regional LUCC might not be driven much by the distorted
and depressed price signals, especially for forestland and grassland. On the other hand,
the provincial price data may have obscured the price variability across counties, making
the regression fail to capture the actual effects of the relevant prices.

Industrial development significantly reduces the pressures on resource exploitation.
The higher the ratio, the lower the shares of cropland and grassland, and the higher
forestland share. A 1% increase of industrial development can result in 0.05% decrease in
cropland, 0.11% decrease in grassland, and 0.07% increases in forestland. While the
magnitudes of these coefficients are very small, they confirm what we speculated — that
economic development affects the LUCC by altering labor opportunity costs and labor
migration. Average farmer’s net income was only around 15-25% of the annual wage for
urban employees in the study region (Sichuan Statistics Yearbook; Yunnan Statistics
Yearbook). Thus, industrial expansion lures labor out of the agricultural sector and
reduces the pressure for farmers to convert forestland to cropland. It might be argued that
since there are a large number of surplus laborers in rural area, labor transfer and the
associated migration should not affect the current labor supply for agriculture. However,
such an argument ignores the fact that surplus labor in the rural area depends on natural
resources for its livelihood, which disturbs the environment to some extent. For example,
grain produced on sloping farmland may be limited, but the close-to-zero marginal labor

cost makes such an activity appeal to farmers. Therefore, industrial development would

70

alleviate the pressure on cropland expansion and induce more labor-saving land use like
forestry.

Highway construction has a significant effect on cropland use. Holding other
variables constant, a 1% increase of the highway rate can result in 0.22% increase in the
cropland share. This is expected, and it is consistent with prior studies (Chomitz and Gray
1996; Nelson et al 1999). The insignificant effects of highway rate on forestland and
grassland changes imply that given other conditions, forestland or grassland share in a
county is not closely connected to the highway density in that county. Such findings are
not surprising, because the whole study area features high altitude and sharp topographic
variation on the one hand and greater abundance of forestland and grassland than
cropland on the other. The demographic factor, population density, significantly affects
cropland and forestland shares. A 1% increase in population density causes the cropland
share to grow by 0.48% and forestland to decrease by 0.33%.

Grain procurement quota has a significant effect on cropland expansion. A 1%
increase of per capita grain quota could induce 0.04% increase of the farmland share. The
results also suggest that a county with a higher grain procurement quota has a larger
farmland share. It implies that eliminating the grain quota system will reduce not only
farmers’ burden but also cropland area. Forest tenure has a significant influence on the
forest resource status. The share of forestland increases by 0.12% when the state-
ownership is 1% higher, holding other variables constant. The result is the consistent with
expectation, and implies that a stable, clearly defined and effectively enforced tenure

arrangement can help protect forest resources.

71

Recall that Chapter 3 already indicates that land allocation varies with altitude. With
an increase of altitude, the cropland and forestland shares decrease, while grassland share
increases. These results are further confirmed here, showing the significant effects of the
biophysical factor on land allocation and suggesting that effects of socioeconomic and
institutional factors on the LUCC are manifested in conjunction with the biophysical
conditions. The results also suggest that land use decisions should be made according to
the local biophysical conditions in different regions.

Because of the large magnitude of forestland modification in the study region, it is
also important to understand how the status of a secondary forest category is affected by
various driving forces. Therefore, this research also develops fractional logit models to
examine driving forces for secondary forest categories. Dependent variables are the
shares of closed forest and shrub (due to the small amount of open and other forest, they
are combined and called other forest); the explanatory variables are the same as those for
the primary land use model. Table 4.4 reports the estimated results.

Compared to the results of forestland share equation, the same five variables have
significant effects on the share of closed forest with the same signs. Specifically, the
share of closed forest decreases if grain price index increases, population density
increases, or altitude increases; the share of closed forest increases with an increase in the
ratio of industry output to agricultural output, or an increase in the share of state-owned
forest. As a major component of the forestland, it is not surprising to see that drivers for
closed forest share also affect the entire forestland share. No variables in the model have

significant effects on the share of shrub, even though the goodness of fit test is

72

Table 4-4 Estimated Results of Fractional Logit Model for Secondary Forest Classes

 

 

 

Explanatory Variables Closed Forest Shrubs
Grain Price Index -0.772 0.658
(0.368)** (0.734)
Log Price Index 0198 0.236
(0.175) (0.203)
Livestock Price Index 0.641 -0.590
(0.772) (1.307)
Industry/Agricultural Output 0.082 0.048
_ (0.042)* (0.052)
Highway Rate 0.068 -0.060
(0.103) (0.179)
Population Density -0.599 -0.046
(0.098) *** (0.183)
Per Capita Grain Quota -0.017 -0.035
(0.023) (0.096)
Share of State-Owned Forest 0.257 -0. 175
(0.132)** (0.170)
Elevation -1 . 171 0.348
(0.236)*** (0.332)
Province Dummy -0.216 0.116
(0.062)** (0.127)
1975 Dummy -0.032 0.030
(0.134) (0.203)
1985 Dummy 0.019 0.015
(0.123) (0.187)
1990 Dummy -0.023 0.018
(0.098) (0.155)
1995 Dummy 0.066 -0.018
(0.048) (0.073)
Observations 122 122
d.f. 107 107
Log-likelihood -47.764 -36.217
Pearson Chi Square 5.382 5.503

 

Note:1. “*”, “**” and “***” represent 10, 5 and 1 percent significance level, respectively.

2. Numbers in parentheses are standard error of the coefficient.

3. Observations are 122, because of some missing values for highway variable.

73

significant. Apparently, the driving forces can differ for the changes of different
secondary forest categories.
4.4 Conclusion and Discussion

This chapter develops and estimates fractional logit models to examine the driving
forces of changes for the shares of different land uses, and a majority of the variables
have significant effects as expected. Estimated results indicate that land use changes in
the study region were not broadly affected by market price signals, but industrial
development did have significant effects on reducing cropland expansion and conserving
forest resources. Population growth imposed pressures on resource uses and contributed
to deforestation. In addition, institutional and policy factors play critical roles in shaping
the land use patterns: lowering grain quota levied on farmers reduced cropland
expansion, and stable forest tenure led to a higher share of forestland.

These results have great policy implications. First, as noted, industrial development
can increase farmers’ income and lure them out of farming and rural economy and thus
reduce their disturbance to land resources. Local governments may thus provide job
information and training services to facilitate farmers’ pursuit of off-farm and off-village
opportunities. Second, the market mechanism should be encouraged wherever it can
provide clear and strong signals for guiding the resource allocation; likewise, the
government should play its role wherever the market is “thin” or “absent” by providing
public goods and regulating economic and environmental activities. Finally, land tenure
arrangement is critical for encouraging sustainable resource use and management. State-
owned forest represents a stable forest tenure arrangement, which has reduced the

possibility of forestland conversion. In contrast, the collective forestland was more likely

74

owned forest represents a stable forest tenure arrangement, which has reduced the
possibility of forestland conversion. In contrast, the collective forestland was more likely
to be degraded and converted to other uses. These results imply that actions must be
taken to reform the tenure system for the collective forestland, including clarification of
use and benefit rights, creation of a well-organized monitoring and enforcement system,
and communication of transparent and fair market information to the forestland users and
managers in order to encourage investment and management.

It should be pointed out that the fractional logit model has some limitations. First, as
proximate, the provincial price indices may have obscured the spatiotemporal price
variation a great deal and made it difficult to draw any firm conclusion of the price effect
on the LUCC. Also, only a county’s average elevation was included as a biophysical
variable, which might not be sufficient to capture the variation of natural conditions.
Additionally, the model did not work well for grassland or shrub as a secondary forest
category, and better defined variables and improved data should be explored.

Finally, the logistic form of the share equations is meant to restrict the estimated
shares to the unit simplex (Miller and Plantinga 1999). However, such a regression model
is ad hoc somehow. It does not treat relevant variables as endogenous nor account for the
feedback effects. Researchers have noticed possible endogeneity in population variables
and used lagged value as an instrument (Miiller and Zeller 2002). But this is far away
from illuminating the dynamic interactions between different variables. To overcome
these shortcomings, the next chapter will develop cropland and forestland structural

models to demonstrate interactions of various facets in the cropland and forestland uses.

75

Chapter 5 Structural Models for Cropland and Forestland Change

5.1 Introduction

This chapter attempts to develop structural models to examine the cropland change
and forestland change, in order to better understand the complex LUCC process. It has
been recognized widely (Lambin et al. 2003) that the LUCC process is a dynamic and
interactive process, and its driving forces and consequences simultaneously affect each
other to determine the long term land use pattern. The logit model developed in chapter 4
explains some of the causes of changes for different land uses, while it is unable to
illustrate the interactive relationships among various components in the land use process.
Therefore, by constructing structural models, this chapter aims to demonstrate the
interactions underlying the land use changes. BecauSe of complexity of the connections
among different land uses, this chapter only focuses on studying the cropland and
forestland changes, respectively. Cropland and forestland are directly associated with
food supply and environmental conservation. How to incorporate these two aspects into
the economic growth path has always been a challenge especially for a region like the
upper Yangtze River that suffers from poverty but also has ecological importance. Thus,
it is expected that studies on cropland and forestland will provide more policy insights on
developing a sustainable land use pattern in the study region.

The structural models are built for the cropland change and forestland change
respectively. In addition to the LUCC’s driving forces examined in the previous chapter,
the cropland structural model highlights the role of agricultural technology progress in
improving cropland productivity as well as in controlling cropland expansion, which has

policy significance especially for the country like China which needs to feed an

76

expanding population with limited land resources. Specifically, the cropland structural
model examines how cropland change is interrelated to agricultural technological
progress and crop production, and how the LUCC’s environmental consequence, soil
erosion, imposes feedbacks on cropland change. The forest structural model examines the
forestland change in the context of its interrelationships with timber production and
changes to the forest stocking volume.

This chapter is composed of three sections. The first section presents the structural
model for cropland change and the second one discusses the structural model for
forestland change. In each of these two sections, the conceptual framework, empirical
model and estimation results for the corresponding structural model will be shown.
Additional variables for each structural model other than those used in the fractional logit
model of chapter 4 will be explained. Final section of this chapter is a brief discussion of
findings from both structural models.

5.2 Structural Model for Cropland Change
5.2.1 Conceptual Framework for Cropland Change Structural Model

The cropland structural model is constructed mainly based on the induced
technological and institutional innovation theory. The theory states (Davis and North
1971; Ruttan 2001) that resource endowments, technology, and institutions are
interrelated with each other in determining the long-term economic growth and social
changes. Moreover, the Hicksian microeconomic foundation approach is employed here
to explain the induced technological change. The Hicksian microeconomic foundation
approach states that the relative price of factors in a production induces the technical

change that would economize the use of an expensive factor. Such an approach was

77

applied to understand the effects of differences in and changes of factor endowment on
the direction of technical change (Ruttan 2001).

Accordingly, the process for changes of resource status and technology can be briefly
described as the following. When changes in the socioeconomic as well as biophysical
environments take place, such as the changes in input costs or output prices, in order to
maximize their returns economic agents will be induced to make technological changes
or adopt technology that will save the use of expensive inputs, which in turn affects
resource status and economic growth. Meanwhile, the change of resource status implies
the change of resource scarcity, which will affect resource rents and therefore induce
another round of technological innovation. Technological innovations happen or
sometimes are driven by institutional changes, such as improved land tenure and price
liberalization (Lin 1992; Yin and Hyde 2000). This is why it is held that resource
endowment, technology and institutional innovations are interrelated with each other.
However, in this research, the technological innovation is of interest and the institutional
setting is regarded as exogenous. What is examined is how the land use interacts with
technological innovation given certain institutional environments.

Viewed from the perspective of such induced technological change theory, the
decisions of land uses can be explained within the context of agricultural technological
changes. Farmers make their land use decisions and agricultural technology practices in
response to the external economic conditions. And, at the same time land use decisions
and agricultural technology change also affect with each other in determining the long
run land uses. Specifically, it is assumed that major inputs for agricultural production are

just categorized as land, labor and capital. When cost of land expansion becomes lower

78

compared to capital or labor prices, land-extensive technology and production mode
would be employed instead of capital inputs. If land becomes scarcer, substitution of land
with labor or capital intensive technology would be adopted (Miiller and Zeller 2002).
Meanwhile, agricultural technology changes such as improved irrigation or crop
management will change cropland productivity as well as potential cr0pland returns,
which will in turn affect land use decisions. Changes in land uses can then alter the status
of resource scarcity, which has impacts on resource rents (prices) and affects technology
application behaviors.

How agricultural technological change interacts with land use decisions is also
subject to other factors such as the economy mode, product elasticity and demographic
change (Angelsen and Kaimowitz 2001). Under the self sufficient household economy,
the change on agricultural technology would halt cropland expansion because the
increasing productivity can satisfy the food demand. Under the open market economy
where the agricultural products are mainly sold as commodity and income sources for the
household, the technological change might promote the cropland expansion for more
profits from crop production, especially in the case of the improved marketing of non-
staple or economic crops (Mertens et al. 2000). With respect to the product elasticity,
products with elastic demands would promote cropland expansion, while products with
higher supplies from technological change but inelastic demand will dampen the
incentives on cropland expansion.

Moreover, land use decisions as well as the extent of technological change are both
subject to labor and capital constraints (Angelsen 1999). The labor opportunity cost thus

is a critical factor in determining land use decisions and technology changes. When

79

labor’s opportunity cost other than farming is lower, farmers would spend more time on
land, either expanding farmland extensively or farming intensively; but if opportunity
costs are higher (such as off-farm jobs), the farmer might prefer a labor-saving land use
pattern such as tree planting and allocate more household labor away from the land. With
regard to capital availability, a credit constraint might discourage capital-intensive
technology adoption; but on the other hand the accumulation of capital might enforce the
technological application on the currently available croplands or expanded croplands,
depending on the labor availability and market conditions.

Therefore, land use decisions are interrelated with technological change, and both of
them are affected by economic return, labor/capital constraints, and other external
socioeconomic, biophysical and institutional factors. Incorporating technological
measures into the LUCC carries great implications. To improve the food supply for an
expanding population on a limited land base, for instance, the productivity and
production increase must come from technical changes, such as increased intensity of
farming and use of modern inputs (Agricultural and Rural Development Taskforce 2004).

Not only does the land use change interact with technological change, but its
environmental consequence will also generate feedbacks on land use decisions. Extensive
farming on slopes can lead to lower productivity but more deforestation and soil erosion;
serious soil erosion may negatively affect land productivity, which would force farmers
to reclaim more cropland to satisfy their food demand, and result in more severe erosion
in a vicious cycle. However, the converse could occur as well. Intensive farming, induced
by technological improvement or relevant policies, will reduce the disturbances to

sensitive ecosystems. An improved environment, such as declining soil erosion intensity,

80

 

will gradually benefit agricultural production; then, increased agricultural productivity
will further relieve pressures on cropland expansion, encourage environmental
conservation and finally lead to a more sustainable land use pattern. Thus, incorporating
the feedback of the LUCC’s environmental consequences into the analysis will better
depict the dynamic LUCC process, and provide policy recommendations to control the
environmental deterioration and encourage sustainable land uses.

The conceptual framework for the cropland structural model is illustrated in the

following flowchart of Figure 5.1.

Figure 5-1 Flowchart of Structural Model for Cropland Change

1

Agricultural
Production

 

 

 

   

  
   

 

 
  

Technolog

 
   

y
Chan

  

\

Inputs Institution/ Markets/ Demographics Biophysics
Policy Economic Factors

It is assumed that crop production, technological change, cropland changes and soil

l .-" ,,,, \ I \ - \
-"'

 

 

 

 

erosion—the LUCC environmental consequence — interact with each other, so that they are
endogenous variables in the structural model; other socioeconomic, biophysical and
institutional factors listed in the second row of Figure 5.1 are exogenous variables in the
structural model. The arrow linking two factors indicates the effect direction. How the
interactive process works in this framework can be explained as below. The cropland

change such as land expansion as well as deforestation might worsen the situation of soil

81

erosion, and degraded soil condition will negatively affect grain production or land
productivity. In order to satisfy food demand, farmers will reclaim more cropland and
then result in more serious soil erosion, and this forms a vicious cycle that happens quite
often in some developing counties. As explained earlier, technological progress plays a
critical role in improving crop production. It is noticeable that technological innovation
also has an impact on soil erosion control through its effect on cropland changes. For
example, if farming technology can improve grain production to an extent such that a
certain amount of cropland will satisfy the food demand, cropland expansion will be
halted or some sloping or inferior croplands will be converted to forest or grassland. The
soil erosion situation may be changed as a consequence of technological changes because
of vegetation rehabilitation, and technological change in this sense contributes to
environmental protection. Thus, this cropland structural model aims to provide policy
suggestions that would incorporate both environmental protection and sufficient food

supply so as to obtain a sustainable land use pattern.

5.2.2 Empirical Model and Variable selection

The conceptual framework depicted in Figure 5.1 can be represented by a system of
four equations. Cropland production, agricultural technological change, cropland area and
soil erosion area are the respective dependent variables. They are also interrelated and
appear as independent variables in some other equations. Other factors listed in the
frameworks are taken as exogenous explanatory variables. This structural model is
expected to demonstrate interactions and socioeconomic causes underlying the cropland

change process. This section will illustrate with details how the empirical structural

82

CIC

V3

 

 

cropland model is constructed and what exogenous factors are included as explanatory
variables for each equation.

Based on the framework, a system of four equations can be generally specified as
blow:

Pit =fl(Cit’Iit’Sit’Xit)+€it

7}, = f2(Crr.Yrr)+ 5::

Cit = f3(7It’Pit-19},it’zit)+vit
Sir =f4(Cit’Xit’Yit)+7it

where f1 - f4 are functional forms of the four equations. Dependent variables are grain

production (P), agricultural technology measure (T), cropland use (C), and soil erosion
(S); I are farming input variables including labor and capital inputs; X, Y, and Z are other
socioeconomic, institutional and biophysical variables (see detail below); i and t denote

the spatial and temporal units of observations; and a, 6, v and r are error terms.

5.2.2.1 Crop Production Equation

The crop production equation is constructed based on a conventional agricultural
production function, according to which the production is determined by agricultural
inputs such as cropland, labor and capital inputs. The capital inputs in this research are
reflected by technology-embedded inputs including fertilizer (amount of application),
irrigation (irrigation area), and other technological elements captured by multiple
cropping index (see detail in data section). In addition, cropland productivity as well as
crop production is also inevitably impacted by soil quality and relevant environmental
conditions. Thus the soil erosion area, representing the environmental condition, is

selected as an explanatory variable of crop production equation. It is expected that crop

83

production will increase with more agricultural inputs, while the degraded environment,
shown as the increase in soil erosion area, will negatively affect crop production.

Grain production is taken in this study as a proximate variable for crop production.
Grain consists of cereals, beans and tubers in China’s statistics, which are major
agricultural products in the study region. The grain sown area accounted for more than
90% of total crop sown area in 1975, and its share was still as high as about 80% in 2000.
Moreover, expanded cropland is more likely for grain production, because such land is
not good enough for high-value economic crops. Thus, it is reasonable to use grain

production representing crop production.

5.2.2.2 Technological Change Equation

Technological innovation is mainly governed by two forms: technology adoption and
technology diffusion. Technology adoption can be measured by the timing and extent of
new technology utilized by individuals, while the technology diffusion focuses on the extent
of aggregate dissemination of a technology (Angelsen and Kaimowitz 2001). Considering
that county is taken as the study unit in this research, the technological innovation — also
called as technological change or progress hereafter — for a county employs the mixture
of the above two forms, and is measured as the increase of agricultural technology
application.

Usually a variety of agricultural technologies is applied simultaneously and they
complement each other in crop production. One technology itself, therefore, can not
represent the overall technological change situation. Thus an index is created in this study
that combines the change of fertilizer application, irrigation infrastructure and multiple

cropping indices. Fertilizer application and irrigation infrastructure are regarded as main

84

technological inputs in crop production, which also contribute to land productivity
increase. Fertilization in a county is represented by the fertilizer amount applied, while
the scale of irrigation infrastructure is reflected by the irrigation area. Multiple cropping
indices capture other technological elements, such as new cultivation practices.
Anecdotal evidence indicates that the application of agricultural film in mountainous
especially high elevation areas effectively increases soil temperature accumulation and
moisture maintenance. This results in a longer growing season as well as higher cropland
production. Such technological change is reflected by multiple cropping indices. The
technological change index at each time point is calculated by multiplying the growth
rate of each of three elements at that time. Three technologies are weighed equally when
creating the index, because farmers apply all of them and weigh them similarly. For
instance, the technological progress index in 1975 is obtained by multiplying the growth
rate of fertilizer application, the growth rate of irrigation area and growth rate of multiple
cropping indices in 1975. Thus, this index reflects the change of agricultural technology
application. The larger the number, the higher the rate of technological change.

As mentioned earlier, technological change is induced by changes in relative prices
(such as other input costs and output prices), and by resource endowment, labor
opportunity cost and capital constraints (Ruttan 2001, Angelsen and Kaimowitz 2001).
An economic agent whose objective is to maximize profits from production activities
adopts the technology that can save costs to the largest extent, or the technology that can
improve the output at a large scale if the output price increases. In addition, the roles of
distance and geography should be emphasized in the adoption of technological change.

Producers in locations farther away from a regional center are likely to adopt and diffuse

85

technologies more slowly. Thus, investment in infrastructure to reduce transportation
costs is expected to accelerate technological change (Sunding and Zilberman 2000). This
study uses the road length in a county to represent the infrastructure condition. Therefore,
agricultural technological change is the function of cropland area which represents the
resource endowment, grain (output) price, input cost, farmer net income (capital
constraints), road construction and the county’s average elevation that denotes the
biophysical environment of a county. It is expected that a decrease in input costs, fewer
capital constraints, or more road construction would promote technological changes. How
the increase of crop output price affects technological changes is empirically dependant
on how farmers weigh household food consumption or cash crops. In crop production
regarding the impact of cropland change, it is expected that more farmers would become
likely to adopt technologies to improve productivity in places where less cropland is
available. However, it is necessary to look at both this technological change equation and

the cropland equation (described below) to understand their interactions.

5.2.2.3 Cropland Use Equation

As indicated in the framework, the technological change interacts with land use
decisions and it might either stimulate or dampen cropland expansion. Labor and capital
intensive agricultural technology would halt the land expansion, but would also motivate
the land conversion to croplands if higher productivity on a larger scale means more
profits. Similarly, labor saving technology might either release labor for frontier land
encroachment or promote migration to out-of-farm industry employment and thus reduce
the disturbance on the environment, depending on the labor cost and capital constraints.

Thus, it is interesting to take technological change as one of explanatory variables in the

86

cropland use equation, and to empirically examine its effect on land use decisions in the
study site. One period lagged cropland production is included in the cropland use
equation as the explanatory variable. Higher crop production in the previous period
would discourage cropland expansion or even encourage the conversion of cropland to
other labor/capital-saving land uses, if the subsistence farming is the priority. But,
farmers might also value the practice that generated high grain production, extend it to
larger space in order to obtain profits from crops, and promote the cropland expansion.
Also, it is not necessary to worry about the causality relations between the cropland
production and technological changes because the production variable is in a lagged
form. In addition, some exogenous factors that are employed in the cropland equation of
the fractional logit model in chapter four are also included here as explanatory variables,
including various price indices, population pressure, industry development, policy and
biophysical factors. Reasons for selecting those variables and a description of each one

and their expected effects on cropland change can be found in Chapter 4.

5.2.2.4 Soil Erosion Equation

As indicated in the framework of the cropland structural model, the changes in land
uses will produce environmental consequences, and such environmental consequences
will generate feedback effects to land uses. In this research, soil erosion is selected to
represent the environment condition that can be altered by the land use activities, and at
the same time impact land use decisions. Soil erosion area for each time point is used to
represent the soil erosion condition. In addition to the land use variable, cropland area,

two policy variables (the upper Yangtze Soil Conservation Project, the upper Yangtze

87

Shelterbelt Development Project), volumes of timber production and biophysical
variables are included as explanatory variables.

Recognizing the environmental deterioration and degraded soil quality, the Chinese
governmental initiated soil control and a forest shelter project in the late 19803.
Implementing projects is anticipated to reduce the soil erosion area, and conversion from
cropland to other vegetation covers is also expected to alleviate soil erosion. The timber
production is expected to be positively associated to the soil erosion area, because slope
land conversion to cropland, over-harvest of timber but poor forest management are

deemed as major reasons for severe soil erosion in the Upper Yangtze River.

5.2.3 Data Description

As in Chapter 4, the panel dataset covering 31 counties over 5 time points (mid
19703, mid 19803, late 19803, mid 19903, and late 19903) is employed in the cropland
structural model to examine the complex cropland change process from 1975 to 2000.
Cropland data are derived from remotely sensed images7 that were classified and
interpreted by Chinese Academy of Sciences (CAS), and Chapter 3 gives details on the
land use data; socioeconomic data are from local statistics yearbooks (Sichuan Statistics
Yearbook 1975-2000; Yunnan Statistics Yearbook 1975-2000), and institutional
variables are mainly from local governmental documents. This section will give a
description of the variables that only appear in the cropland structural model (Table 5.1).
See Chapter 4 for descriptions of those that are already included in the fractional logit

model.

 

7 For the year of 1985, land use data were digitized from a land cover map.

88

Table 5-1 Summary of Additional Variawa used in Cropland Structural Model

 

 

 

 

Variables Measurement 1975 1985 1990 1995 2000
Endogenous
Cropland Area km2 16,037 16,146 15,617 15,634 15,607
Grain Production 1,000 Ton 1,623 1,954 1,942 2,633 2,786
Technology Index 0.073 0.044 0.054 0.065 0.075
Soil Erosion Area km2 20,330 23,850 28,940 30,110 27,430
Exogenous
A 'cultural
Lagl'iirs ”’00 Person 2,178 2,798 3,047 3,509 3,623
Irrigation Area km2 1,896 1,927 1,945 2,125 2,692
Fertilizer Ton 35,984 62,444 72,991 140,065 215,603
Multiple Cropping
indices 1 .397 1 .390 1 .448 1 .623 1.707
Farmer Income 100RMB 0.81 2.78 3.44 5.65 11.32
Cost Index1 Index (1975=1) 1.001 1.071 1.363 2.387 3.620
Timber Production Cubic Meters 259,925 833,007 850,862 854,845 380,868
Soil Project Dummy Variable =1, if project implemented; =0, otherwise

Shelterbelt Project Dummy Variable

=1, if project implemented; =0, otherwise

 

Note: 1. Cost index is the provincial average.

2. Timber production and Shelterbelt project implementation information are from
local forest bureaus; Soil erosion area and Soil project (Soil Conservation Project)
implementation information are from local water resources bureaus; cropland area
is from IGSNR, CAS; others from statistics.

Grain production increased by 70% from 1975 to 2000, despite a small drop in 1990.

The biggest jump in production occurred between 1990 and 1995 at a 7% annual growth

rate. This was also a time of significant progress in technology application. It is also

interesting to compare crop production growth with changes for cropland or labor input

over time, which helps identify the source of increased production. Agricultural labor in

2000 increased by 66% compared to 1975’s level, while cropland area decreased by 3%.

It is the technology-embodied inputs that resulted in the production growth not explained

by labor increase over time. Crop yield per unit of agricultural labor increased by 3%

from 1975 to 2000, while the yield per hectare of cropland increased by 76%. Such

productivity improvement carries significant implications for China. It not only

guarantees sufficient food supply from limited cropland resources to support an

89

expanding population and a booming economy, but also reduces the pressure on other
natural resources so that it is possible to maintain a more sustainable land use.

Fertilizer application and irrigated area both increased over the study period. The
increase in irrigation area accelerated, from an annual growth rate of 0.2% between 1975
and 1990, to 2% in the early 1990S and then up to 5% in the late 19908. The jump of
fertilizer application took place in the 19908. The amount of fertilizer application in 1995
was doubled compared to the 1990’s level, and the application in 2000 reached 1.5 times
that of the 1995’s level. Fertilizer effectiveness is restricted by water availability, so
growing fertilizer application is accompanied by increasing irrigation area. Notably,
fertilizer application is not only a capital intensive, but also a labor intensive technology.
Higher fertilizer input per unit of land increases the demand for labor, which will reduce
labor available for other activities like cropland reclamation. From 1975 to 1990, the
multiple cropping indices increased by 4%, and from 1990 to 2000 increased by 17%.

As mentioned earlier, the technological progress index for each time point is
calculated by multiplying the growth rate of each of three elements at that time. The
baseline for calculating the growth rate of each technological element is the previous
year, but not the previous time point in the study series. For instance, the growth rate of

fertilizer application in mid 19708 is calculated in two steps. First, the yearly growth rate
for 1976 and 1977 is expressed as (Z, - Z,-1)/Z,, where t indicates 1976 or 1977, and Z

denotes the fertilizer application. Second, the growth rate of fertilizer for 1975 is the
average value 0f the (21976- Z]975)/Z}975 and (21977- 21976)/ZI976- The growth rate for each

technological component is calculated in the same way, and the technological index of

90

1975 is a product of themg. For 1995 (when only the year 1995 is included), the baseline
for calculating the growth rate is 1994. Such a technological progress index takes various
technological components into consideration. The rate of technological progress
accelerated from 1985 to 1995, but slowed down by 2000.

Farmer’s net income represents their abilities for applying technology. It is assumed
that farmer with a higher income has the ability to afford more technological inputs.
Farmer’s per capita net income in a county, the variable used in the model, reflects the
average rural income level in that county. Farmer’s annual per capita net income
increased continuously over time. From 1975 to 1985 and from 1995 to 2000, the average
annual growth rate was more than 20%. Farmer’s annual net income in 2000 was about
15 times that of the 1975’s level. However, it should be cautious to state that farmers’
living condition has not been close to that of the urban residents. Annual wage of urban
employees also increased dramatically, and the gap between farmer’s income and urban
earning became even bigger over time. In 2002, the national per capita disposable income
for urban population in Sichuan and Yunnan was RMB 6,611 and RMB 7,241
respectively; farmer’s per capita net income was just RMB2,107 and RMB1,608 in
Sichuan and Yunnan, respectively (Sichuan Statistics Yearbook 2002; Yunnan Statistics
Yearbook 2002).

The production cost index is a provincial average that is a comprehensive cost
indicator. Its calculation takes into account various cash expense for agricultural inputs
including fertilizer, machinery, feeds, pesticide and other inputs. This cost index reflects

the trend of production cost over time. It is seen that the cost index more than tripled in

 

8 Technological change index is not calculated by comparing the technological change between two
adjacent time points (e.g., 1985 vs. 1975). This is done to: first, avoid missing values for the year of 1975;
second, keep consistent with time dimension of other explanatory variables.

91

2000 compared to that of 1975. The cost level increased by about 36% from 1975 to
1990, and then by 75% in the early 19908 and by another 50% in the late 1990s. The
average annual growth rate of the cost level was as high as 16%. Since technological
progress in this study means more application of technology-embodied inputs, the
escalating cost level might reduce farmers’ incentives to adopt technology.

The soil erosion area for each county comes from the local water resources Bureaus.
Two national soil erosion surveys were conducted in China, the first in the late 1980s and
the second at the end of the 1990s. Before the late 1980s, the soil erosion area for each
county was estimated by the local water resource bureaus based on their own field
surveys. Data for 1990 and 2000 are based on nationwide surveys. The total erosion area
in the study region continuously increased from the mid 19703 to the mid 1990s, and
declined around 10% in the late 1990s.

Over the study period, the government implemented a series of projects to control soil
erosion. The Soil Conservation Project along the upper Yangtze (hereafter, soil project)
carried out since 1989 in many counties is the biggest project to control erosion along the
J insha River. The soil project combines bio-control, engineering and cultivation measures
to mitigate current soil erosion and prevent future soil erosion. It also incorporates the
livelihood improvement for local people, by means of planting income producing trees
and improving cultivation methods. Until 1999, the soil control area thanks to the soil
project along the Jinsha River in Yunnan accounted for 43% of the total soil control area
for the whole province (Wang 2003). In Sichuan province, the soil project harnessed

21,500 km2 eroded area by 2003. A dummy variable is created to represent the

92

implementation of the soil project. In 1990, the soil project was implemented in 10
counties, and in the late 19908 the coverage was expanded to 15 counties.

The Shelterbelt project refers to the upper Yangtze Shelterbelt Development Project
starting in 1989. By means of plantation, mountain closure, and forest regeneration, this
project aimed to control soil erosion, enhance forest ecological functioning, and improve
farming conditions. From 1989 to 2000, the Shelterbelt project was implemented in 145
counties along the Upper Yangtze River, and made noticeable achievements. In 2001, in
Yunnan forest plantation under the project reached 4,600 kmz, mountain closure covered
3,700 kmz, and area for tendered forest was about 760 km2 (Yunnan Forest Bureau 2002)
In Sichuan, forest plantation under the project reached 11,700 km2, mountain closure
covered 4,500 kmz, and area for tendered forest was about 730 km2. In Sichuan, the soil
erosion area in counties under the project decreased by 34% compared to the soil erosion
area prior to the project implementation (Sichuan Forest Bureau 2002). This project was
not implemented in each county along the upper Yangtze River, and the starting year of
forest shelter project also varied for different counties. Therefore, a dummy variable is
employed to indicate the project implementation status for a county: if the project was
under taken, this dummy variable is valued at one. Otherwise, it is zero. In 1990, nine
counties in the study site were included in this Shelterbelt project. In 1995, the number of
counties covered by the project increased to 19 and kept at 19 in the late 1990s. Since
then, the Shelterbelt project in the upper Yangtze River was replaced by the National
Forest Protection Program (NFPP) that was initiated in a large scale along the upper

Yangtze River and all counties in the study region are covered in the NFPP area.

93

Timber production data come from local forest bureaus. From 1985 to 1995, timber
production was stabilized at a high level to provide raw materials needed for rapid
economy development. The massive flooding along the Yangtze River in 1998 raised
concerns about environmental deterioration, partly resulting from the large scale of
timber production and forest degradation in the region. As a result, the NFPP was
implemented from 1998 along the upper Yangtze River to protect forest resources, and
even cutting commercial logging in some counties. That is why the total timber
production from the studied counties in 2000 decreased sharply to 44% of the 1995 level.

The field survey indicated that 14 counties had no commercial logging in 2000.

Table 5-2 Expected Signs for Variables in the Cropland Structural Model

 

Grain Production Cropland Technology Soil Erosion

 

Grain Production (lagged) +/—
Cropland + — +
Technology —

Soil Erosion Area
Agricultural Labors
Fertilizer
Irrigation

Multiple Cropping Index
Log Price Index
Livestock Price Index
Grain Price Index
Population Density
Per capita Grain Quota
Highway Rate

Farmer Income 4-

Cost Index —

Share of State-owned
Forest
Industry Development + +
Soil Project -
Shelterbelt Project -
Timber Production +
Elevation — - — +/—

++++1

+++++1

 

Note: ‘+’ (‘—’) means that the positive (negative) effects of explanatory variable on the
dependent variable.

94

Table 5.2 lists expected signs for variables in each equation of the cropland structural
model. As indicated earlier, the effect of grain production at the previous time point on
cropland could be positive or negative. And, the effect of elevation on soil erosion is also
uncertain, because natural conditions within a county may vary a lot and variables such

as slope or soil structure probably have more influence on soil erosion condition.

5.2.4 Estimation Method for the Cropland Structural Model

The cropland structural model is estimated by the 3 stage-least-squares (3SLS)
method. Appropriate estimation method and identification are two critical issues in
estimating the structural model. Since at least one of explanatory variables in each
equation is the dependent variable in other equations, such variables become endogenous
and correlated with the error term. This makes the separate ordinary least square (OLS)
estimation for each equation fail to provide the unbiased and consistent coefficient
estimates. Thus, the 3SLS method is applied to estimate the four equations
simultaneously. 3SLS is more efficient than the two-stage-least-square (ZSLS) method
(W ooldridge 2002); however, assumptions are needed for 3SLS estimation (W ooldridge
2002). First, it is assumed that all exogenous variables are uncorrelated with any error
term at each time, making it feasible to include lagged endogenous variables. Second, the
rank order condition needs to be satisfied for equation identification. The model is
constructed in a way that for each of four equations the number of excluded exogenous
variables from the equation is at least as many as the number of the endogenous variable
included in this equation (W ooldridge 2002). Thus, the variable selection for this study
provides the prerequisite for the identification (significance of the exclusive exogenous

variable at each equation will ensure the identification). Third, the covariance matrix of

95

the e

(1110‘

The

prt

1'8

 

 

E
:Illflll
Lt}

the error terms is systematically homoskedasticity. In this study, province and year
dummy variables are included for each equation to reduce the heterokedasticity.

Therefore, the model satisfies the conditions for the 3SLS.

5.2.5 Estimation Results for the Cropland Structural Model

Table 5.3 lists the estimation results for cropland structural model. Yearly and
province dummy variables are included in each equation to control the temporal trend
and regional heterogeneity (see Appendix 5.1 for coefficients of year and dummy
variables). Log transformations for variables of grain production, cropland, agricultural
labor, fertilizer and irrigation are used in the estimation.

It should be noted that the coefficient listed for each variable in grain production,
cropland and soil erosion equations is the corresponding elasticity, but not the partial
effect. This way of presentation can clearly indicate the effect of each explanatory
variable regardless of its scale or unit. For the technology equation, the coefficient value
is the estimated partial effect because technological progress itself is an index. In general,
Chi-squares test shows significance for each equation, and this means that there is
significant linear relationship between the independent variables and the dependent
variable. Adjusted R2 values further indicate that the crop production and cropland
equations fit very well. Several variables of our primary interest are statistically
significant with signs as expected (Table 5.3). In each equation, at least one exogenous
variable that does not appear in other equations is statistically significant, which implies
that the identification requirement for the 3SLS estimation is satisfied. When those four
dependent variables are used as independent variables in other functions, they are all

statistically significant. This provides empirical evidence that interactions and feedbacks

96

exist in the cropland land use processes. That is, crop production, cropland use change,
technological progress and environmental consequence do affect each other in
determining the long run cropland use pattern.

Results of the grain production equation are mostly as expected. Cropland area, labor
and technological input such as fertilizer and multiple cropping indices have significantly
positive effects on grain production, while deteriorated soil condition has a significantly
negative effect on production. Specifically, grain production increases by 0.6% and
0.27% with 1% increase of cropland area and labor, respectively, holding other variables
constant. 1% increase of fertilizer application leads to a 0.16% increase in grain
productions and 1% increase in multiple cropping indices will lead grain production to
increase by 0.26%. Irrigation’s effect is not significant in this multivariate equation,
perhaps because its effect is partially explained by the fertilizer variable. As mentioned
earlier, the construction of irrigation facilities promotes fertilizer application, because
fertilizer functions more effectively under appropriate soil moisture conditions. Soil
erosion negatively affects grain production. Holding other variables constant, 1%
increase of eroded area will make grain production reduce 0.1%.

Livestock price index, highway rate, and grain quota have significant positive effects
on cropland expansion, while cropland area decreases significantly with rising altitude
and technological progress. Except technological progress, the other variables mentioned
above are also significant in the cropland share model of Chapter 4. Cropland area is
expected to increase by 0.55% when highway increases by 1%. Similarly, 1% increase in
per capita grain quota leads cropland to increase by 0.06%, but cropland area decreases

with rising elevation. Also, 1% increase in livestock price index makes

97

Table 5-3 Estimated Results for the Cropland Structural Model

 

 

13:72:22}:er Prfdliracigon Cropland Technology Soil Erosion
Grain Production (oggzgfifl
Cropland 0.604 -0.l98 0.380
(0.062)*** (0.064)*** (135.085)***
Technology (035(31‘
Soil Erosion Area (0:31)? )9,”
Agri. Labors (0.85251“
Fertilizer (0.321%...4
Irrigation (8333)
Multiple Cropping Index (035232“
Log Price Index (839783)
Livestock Price Index 02.656878)”,
Grain Price Index (363352) (8 017388)*
Population Density (8233)
Per capita Grain Quota (195%,.
Highway Rate (Ogssglfl (0%???“
Farmer Income ((2 (3)018)”:
Cost Index (dggfiak
Share of State-owned 0.223
Forest (0-196)***
Industry Development (338) (£3932)
Soil Project (25:93)“
Shelterbelt Project (£63120)
Timber Production (0.385%...

98

Table 5-3 (cont’d) Estimated Results for the Cropland Structural Model

 

 

 

Grain Soil
Production Cropland Technology Erosion
Elevation 0.101 -1 .000 0.002 -0.1 14
(0.028)*** (0.055)*** (0.003) (0.097)
R2 0.98 0.80 0.17 0.49
Chi Square 4722.83 42.6 21.12 90.59

 

Notezl. “*”, “**” and “"‘**” represent 10, 5 and 1 percent significance level, respectively.
2. Numbers in parentheses are standard error of the coefficient.
3. The total observations were 98, because of some missing values for highway
variable, and lagged explanatory variable.
cr0pland increase by 2.6%, holding other variables constant. Livestock production has
been regarded as one means of improving income and alleviating poverty in some
mountainous region in China. This result calls for attention to the issues of balancing
cropland expansion for feed production and natural vegetation protection elsewhere. It
has been seen that the expanded cattle husbandry was linked to deforestation (Walker et
al. 2000).

It is interesting to note that technological progress has a significant effect on reducing
cropland expansion. Controlling other factors, cropland area is expected to decrease by
0.04% with 1% increase in the technological index. This implies that technological
progress leads to intensive farming and help convert certain marginal cropland to other
uses. Notably, the crop production in the previous period is positively related to the
current cropland area. But, it is not surprising to see such positive sign because of high
correlation between current crop production and one-period lagged crop production (a

correlation coefficient of 0.93) as well as a high correlation between cropland size and

crop production (a correlation coefficient of 0.84). On the other hand, it might imply that

99

.11

f h ‘K
.

higher production in the past encourages farmers to extend the similar cropping practice
to the next year. This also implies that the economy is not completely self-sufficient in
the study region, and farmers apply their cultivation practices extensively to earn more
income. Technological progress helps control cropland expansion and contributes
significantly to crop production, but higher crop production might possibly induce the
increase of cropland area. Does technological progress really control or promote cropland
expansion? Theoretically it is possible for both. But empirically the result indicates that
technological progress contributes to controlling the increase of cropland, holding all
other conditions constant. This finding brings up the question of how to make
technological progress largely exert its function of improving production and conserving
the environment, and of how to guide farmers to make better use of labor and technology,
while not expanding cropland area, and thereby to improve and sustain the crop
production.

The effects of the explanatory variables of technological progress are mostly as
expected. Farmer income, grain price index and highway rate all have significant positive
effects on technological progress, while higher input costs and abundant cropland
resources will reduce farmers’ incentives to apply technology extensively. But the
magnitudes of the variable coefficients are small9. Holding other factors constant, the rate
of technological progress is expected to increase by 0.007 (a 20% increase of 2000’s
level) if farmer’s income increases by 100 RMB. One unit increase of grain price index
can induce 0.14 point increase of technological progress. Increase of 1,000 km highway

per ha results in 0.09 point unit increase in technological progress. On the other hand, a

 

As mentioned earlier, the coefficrent here rs the marginal effect, but not the elastrcrty.

100

one unit increase in the agricultural input cost index reduces the rate of technological
progress by 0.001 point unit. The adoption and extension rates of land-intensive
technology are 0.198 point lower if cropland increases by 100 kmz. This result further
proves our hypothesis of interaction between technology progress and cropland use.
Although most variables we are interested in have significant effects on technological
progress, the overall explanatory capability of this equation is not that strong. Only 17%
of variation of technological progress was explained by these variables.

Implementation of the soil project has effectively reduced the soil erosion area. On
average, the implementation of the soil project leads eroded area to decrease by 0.11%
for a county, controlling other variables. Cropland expansion has significant effect on soil
erosion increase. 1% increase of cropland will induce soil erosion area to increase by
0.38%, controlling other variables. Meanwhile, a large amount of timber production can
also intensify soil erosion. 1% increase in timber production can make the soil erosion
area increase by 0.16%, holding other variables constant.

Overall, results from four equations of the cropland structural model indicate
interactions and feedbacks in the cropland change process. Not only did cropland change
interact with its driving forces, but the environmental consequence of cropland use also
generated feedbacks on land use. Controlling cropland expansion will help mitigate the
severity of the soil erosion, which will subsequently benefit crop production. As
expected, technological progress plays a critical role in the agricultural sector. It not only
contributes to the growth of crop production, both also helps control cropland
encroachment by satisfying food demand from limited croplands and thus helps conserve

natural vegetation. These results provide some implications for how to encourage

101

3 Cl“: {To—33x7: 37H

 

technological progress, such as increasing crop prices, lowering input costs, and increase
farmer’s income. The model has also illustrated the significant effects of policies on land
use decisions and environmental protection, and the result justifies the need for initiating
appropriate conservation projects by the government.
5.3 Structural Model for Forestland Change
5.3.1 Conceptual Framework for Forestland Structural Model

As the largest land use type in the study region, the forest not only plays a critical
ecological and environmental role in the whole Yangtze River Basin, but also provides a
variety of products for local residents such as housing raw materials and fuel wood.
Moreover, the forest also supports the local economy in terms of job opportunities and
fiscal revenues, especially for those counties with less cropland resource. Before the mid
19908 some counties heavily depended on the timber industry in their local economy.
How to use this renewable natural resource in a more sustainable way has become a
challenge in the study region, because degraded forest quality has already affected
potential forest production capacities, and hampered environmental functions. Thus, it is
necessary to examine the forest status change in a systematic way, obtain a understating
of its various causes, and then develop a viable plan for more sustainable management.

The forest status is reflected in not only the forestland area, but also its stocking
volume and potential timber production (Yin and Newman 1997). Alig and others (1998)
also agreed that when building a dynamic agricultural and forestry equilibrium model, the
forestry components need to include forestland area, quantity of stocking volume and

forest products simultaneously and all of them are regarded as endogenous variables.

102

Therefore, this study attempts to analyze the forestland change in consideration of its
connection to stocking volume and timber production.

The conceptual framework for the forestland structural model can be expressed as the
following. In addition to exogenous factors, the forestland area is affected by the timber
production. It is expected that the rising timber production is likely to make the forestland
area decline. As explained in Chapter 4, it is not rare to observe poor regeneration after
harvest and less incentive for forest investment, especially in the mid 1980s when forest
property rights for the collective forest were not clear and secure. Timber production is
expected to increase with the increase of forest stocking volume, and an expanded
forestland will contribute to the growth of the stocking volume. From the biological
perspective, the forest stocking volume depends on various factors including forest age
structure, species structure, growth rate, soil biophysical conditions, management inputs
and events like fire. The timber harvest timing and quantity should be determined by the
stand’s mean annual increment from the biological dimension, and economic factors such
as timber value, costs and the discount rate (T ietenberg 2000). But approaches like those
are unrealistic for this study. First, many biological and production input data are
unavailable. Second, forest species, ages as well as their values and planting/harvesting
costs are so heterogeneous in a county such that it is hard to accommodate them in
modeling the aggregate stocking volume as well as timber production. Thus, this research
only includes some socioeconomic and institutional factors in the stocking volume and
timber production equations.

Sometimes, it is hard to make policies that incorporate both environmental

conservation and timber production. For example, the increase of forestland would

103

increase stocking volume, and indirectly benefit timber production. However, without
active regeneration efforts the growing harvests might put some forestland at the edge of
conversion, which would damage the future timber production as well as other ecosystem
services. Thus, it is meaningful to develop a structural model and understand the

tradeoffs or even counter effects of various factors.

5.3.2 Empirical Model for the Forestland Structural Model

According to the conceptual framework, the forest structural model is composed of
three equations. Forestland area, stocking volume and timber production are the
respective dependent variables for each equation, and meanwhile they are also affected
by each other and act as the independent variables at the corresponding equation.
Generally, the empirical model can be expressed (as the following:

Fir =fl(Tit’Xit)+£it

Tit =f2(Sit’Yit)+6it
Sir = f3(Fit9Zit)+vit

where f1-f3 are functional forms of the three equations; F, T and S denote forestland,
timber production and stocking volume, respectively; X, Y, and Z are explanatory
variables for each equation (see detail below); i and t denote the spatial and temporal
units of observations; and a, 6, v and r are error terms.

Forestland is a function of the amount of timber production, and other exogenous
variables that are used in the fractional logit model in Chapter 4, including industry
development, highway rate, population density and forest tenure variable. Instead of
including both the grain price index and the log price index, here the ratio of grain price

index to the log price index is used. An increasing ratio means that the growth rate of the

104

grain price index is faster than the growth rate of the log price index. It is expected that
forestland will increase when the ratio of price index decreases. This variable directly
reflects how forestland area is associated with the relative change of price indices. The
use of this variable also helps the variable selection for the timber production equation“).

The amount of timber production is a function of stocking volume, the log price
index, highway rate, forest tenure variable, and polices including the NFPP and the
Shelterbelt project. It is expected that a higher log price index will drive more timber
production, although the extent is questionable because of the ‘planned-economy’ system
in the forestry sector that implies a heavy government control. Every year, the provincial
forest bureau formulates the ‘timber production plan’ that specifies the amount of timber
production. Once approved by the national forest agency, the amount of planned
production is assigned to individual counties based on their forestland area, species
structure, and historical production record. In principle, the amount of timber production
for each county should be strictly subject to planning, but in reality it is not. Timber was
the major revenue source for some counties in the study region, so it is expected that
more logging than the planned is observed when the price is becoming favorable.

As before, the highway rate reflects the market access, harvest and transportation
costs. It is recognized that easy access to roads could result in deforestation (Cattaneo
2001; Chomitz and Gray 1996). It is expected that timber production tends to increase
when more forestland is accessible. A clear and secure forest tenure system is beneficial

to forest conservation as well as to timber production because forestland users have

 

10 If we use the grain price and log price separately in the model, there will be no unique exogenous
variable in the timber harvest equation, which causes the serious identification problem when estimating
the structural model.

105

incentives to invest in and manage forests (Yin et al. 2003), but in the short run (for
example, within a year) timber production from the forestland with clearly defined tenure
might be lower than that from other forestland. The unclear and insecure forest property
rights put more uncertainties on the ownership or use rights of forest stands, which might
encourage large amount of timber production in the short run. The forest tenure in this
study is represented by the share of state-owned forest because the state-owned forestland
is usually not allowed to be converted to other land uses. It is anticipated that the amount
of timber production will be less where there is more state-owned forestland. However,
the actual effect is an empirical question. Most of the mature or near-mature forest of
good quality in the study region was classified as the state-owned forest. To fuel the
national economy, since the late 19603 large amounts of logs were cut and delivered to
every comer of the country as commanded. The situation changed from the late 19908
when commercial logging was banned. Thus, estimation results might deviate from
expectations. Both the NFPP and the Shelterbelt project are expected to reduce the
amount of timber production. The NFPP directly requires the reduction of commercial
logging, while the Shelterbelt project mandates a large scale of mountain closure that
constrains the access to certain forestland. These projects will improve the forest quality
as well as the future production that cannot be reflected in this study, however, because
of a comparatively short implementation period in the dataset.

Forest stocking volume is a function of forestland area, forest tenure, forest quality,
and forest projects. Forest quality is reflected by the percentage of close forest. A higher
percentage means a better forest quality. Forest stocking volume is expected to increase

when forestland area is expanded, forest quality is improved, forest tenure is more secure

106

 

(high percentage of the state-owned forestland), and Shelterbelt project is implemented.
The biophysical factor included in all three equations is the county’ elevation, capturing

the external environmental features.

5.3.3 Estimation Method and Data Description

Similar to the cropland structural model, the forestland structural model is also
estimated by a 3SLS method. The error terms assumption for three equations is also the
same as described earlier. Each equation contains at least one exogenous variable that
does not appear in the other two equations, and such a variable selection would make it
possible to meet identification requirements.

The panel dataset covering 31 counties over five time points (mid 1970s, mid 19808,
late 19808, mid 19908, and late 19903) is also used in the model to examine the forest
status change from 1975 to 2000. Table 5.4 lists the variables that are not used in the

fractional logit model and the cropland structural model.

Table 5-5 Additional Variables used in Forestland Structural Model

 

 

 

 

 

Forestland model 1975 1985 1990 1995 2000

Endogenous

Forestland Area Ha 7,487,370 7,530,783 7,498,984 7,551,545 7,466,451

Timber Production m3 259,925 833,007 850,862 854,845 380,868

Stocking Volume 1,000m3 21,126 16,020 18,532 19,736 21,219

Exogenous

firm/Log. Pm" 0.996 1.259 1.048 1.548 1.573
ndex Ratio

Quality Forest Share 0.53] 0.556 0.535 0.554 0.530

NFPP Project Dummy =1 , for the year of 2000; =0 for other years

 

Note: Data of forestland area is derived from remotely sensed images (1985 data is from
digitized land cover map. Data source: IGSNR, CAS). Timber production, stocking
volumes and quality forest share are from local forest bureaus. Grain/Log price
index ratio is calculated All other variables from provincial statistics yearbooks.

107

The forestland area listed in Table 5.4, the total forestland area for all counties in the
study region, came from the remotely sensed images (1985 data is from digitized land
cover map). Chapter 3 gives a detailed description for its change from 1975 to 2000. The
total forestland area for the whole study region fluctuated during the study period. Within
the 31 counties, 17 experienced an increase of forestland, while the other 14 saw a
decrease. The variable ‘share of quality forest’ is the ratio of closed forest area to the total
forestland area in a county, and data source for closed forest is the same as that for the
primary forestland. It is expected that the higher the share of quality forest, the higher the
forest stocking volume.

Data for stocking volume came from local forest bureaus. Forest inventory is
conducted every five years since the mid 1970s at the provincial level, and the sixth
national forest inventory was completed in 2004. The forest inventory records the area
and stocking volume of the forest by their species, ages, functions (such as timber, fuel
wood, Shelterbelt, special purpose), and by ownership (state-owned or collective). The
provincial level forest inventory data were derived from sample plots across the whole
province. At the county level, the inventory is not under taken regularly. Only when the
county is included as the sample plot for the provincial inventory, is the survey
implemented there. A number of counties canied out less than five forest inventory .
surveys during the past three decades (among 31 counties, one had no forest inventory,
six had one forest inventory, three had two, four had three, and three had four.
Altogether, there are 106 observations for 31 counties). Missing stocking volume data are

not interpolated in this research in order to reduce measurement errors, which therefore

108

reduces the total observations for modeling. Stocking volume listed in Table 5.3 is the
average stocking volume of counties whose data are available at that time point”.

The average stocking volume decreased by 24% from 1975 to 1985, when China
witnessed the dramatic change. Also, timber production increased significantly in the late
19708 and the early 19808. In the early 19808, collective forestland was contracted to
farmers for management. Fearing possible policy reversion, farmers increased the harvest
from collective forest sharply at that time. On average, collective forest accounts for
about 70% of the total forestland area in the study region. These two factors explain the
substantial decrease of stocking volume during 1975 - 1985. Since 1985, the stocking
volume increased steadily and in 2000 it recovered to the level of 1975. Such a recovery
may be attributable to the implementation of forest projects and an emphasis on forest
protection by the government. It should be noted that the increase in average stocking
volume in the study region does not mean that every county observed such increase.
Within 31 counties, 13 counties observed an increase of stocking volume, while another
11 observed a decrease during the study period (others had no or only one inventory).

The change in stocking volume over the study period is not consistent with the
change in forestland area”. This results from a couple of reasons. First, the different
sources for forestland area (from remotely sensed images) and stocking volume data
(from forest inventory) can cause inconsistency. The year when the county undertook the
inventory might not be exactly the same year when the remotely sensed images were

taken. For example, the forestland area for the 1990 was interpreted from images of 1989

 

11 If at a time period, for example, 25 counties had stocking volume data, the figure of stocking volume
listed in Table 5.3 is calculated from dividing total stocking volume from the 25 counties by 25. Variable
used in the model is still the county’s stocking volume.

12 For example, from 1975 to 1990 forestland area increased while stocking volume decreased; from 1990
to 2000, forestland area declined while stocking volume increased.

109

and 1999, while the stocking volumes for a county might be investigated in 1987 or 1991.
Second, if the reduced forestland was marginal forestland, the reduction does not affect
the stocking volume. One the other hand, if the increase of forestland was from new
plantations, it also adds little to the total stocking volume.

Data for timber production were also collected from the local forest bureaus. The
information in Table 5.4 is the total production for all the counties in the study region at
each time point. Prior to 1985, timber production statistics only included commercial
timber and fuel woods. To better reflect the resource consumption situation, after 1985
the definition of timber production was expanded to include the amount of timber
harvested for rural resident self-use in construction, agricultural production and
infrastructure development, but it still excluded fuel wood used by farmers themselves”.
So, the sharp increase of timber production from 1975 to 1985 partly resulted from this
definitional change. In case the timber production statistics not available, the ‘timber
production quota’ is used as a proximate, because the actual production is supposed to be
close to the planned amount or quota. Of course, the timber production quota was not
always exactly the same as actual production, but it can approximately reflect the
production trend and variations between counties. From 1985 to 1995, the total timber
production increased steadily, but the production varied among sampled counties. For
example in 1990, the minimum timber production among counties was only 1,000 cubic
meters (m3), while the maximum reached 120,000 m3. The total timber production

declined by more than half in the late 19908, thanks to the commercial logging ban along

 

‘3 The definition change occurred to all counties in 1985, and a year dummy variable is expected to capture
part of such a change and thus coefficient estimates for other variables are still valid.

110

the upper Yangtze River (Yin et al. 2003). Table 5.5 lists expected signs for variables in

the Forestland Structural Model.

Table 5-5 Expected Signs for Variables in the Forestland Structural Model

Timber Stocking
Production Volume
Forestland Area +
Timber Production +
Stocking Volume +
Grain/log price index —
Log Price +
Industry/Agricultural +
Output
Shelterbelt Project + —
NFPP —
Quality Forest
Share of State-owned + —
Forest
Highway Rate — +
Population Density —
Elevation +/— +/—
Note: ‘+’ (‘—’) means positive (negative) of explanatory variable on the dependent
variable.

 

Forestland Area

 

++++

 

5.3.4 Estimation Results for the Forestland Structural Model

Table 5.6 lists the estimation results for the forestland structural model, and numbers
in the table is the elasticity (see Appendix 5.2 for results of coefficient and standard
errors). Because of missing values for stocking volume and for highway variable, only 83
observations are used for estimation. Nonetheless, the sample size is big enough for
employing the 3SLS and statistical inference, and the whole structural model is
identified”. The Chi Square values are significant, which means we can reject the
hypothesis that there is no linear relationships between the dependent variable and all

independent variables for each equation. Year and province dummy variables are used in

 

l4 . . . . . . . . .
At least one exogenous variable for each equation that rs not included in other equations 1s srgnrficantly
different from zero

111

estimation to control heterokedasticity (see Appendix IV. for results of dummies).
Adjusted R2 indicates that explanatory variables in each equation did to some extent

explain the variation of the dependent variable.

Table 5-6 Estimated Results for the Forestland Structural Model

 

 

 

. Forestland Timber Stocking
Explanatory Variables Area Production volume
1.910***
Forestland Area
Timber Production '0'172 f
Stocking volume 1'006 [
Grain/log price index 0‘ 103 .
#31:
Log Price index 0'880
$3101:
Industry/Agricultural output 0'87]
_ =1:
Shelterbelt Project 0'148 005]
l lFP - #3101: _
P 0.272 0.019
**
Quality Forest 0685
** _ ***
Share of State-owned Forest 0'389 0389 0460
- *
Highway Rate 0.115 0.325
- *4:
Population Density 0247
Elevation 0.842** -1 .328***
Adj. R2 062 0.61 0.85
Chi Square 169.42 131.3 462.58

 

Note: 1. “*”, “**” and “***” represent 10, 5 and 1 percent significance level, respectively.

Most significant variables in forestland equation are also significant in the forestland
equation of the fractional logit model, which implies their consistent effects no matter
what model is constructed. However, the structural model can illustrate the complex

interactive relationships embedded in the LUCC process, and the significance of

112

endogenous variables empirically proves such interactions. Moreover, the inclusion of
demographic, market, institutional and policy variables in the model helps us understand
how these factors affecting the forest status.

Forestland area decreased with population expansion, and increased significantly with
the industry development and stable forestland tenure. Specifically, 1% increase in
population density in the study region is expected to induce 0.24% decrease of forestland
area, holding other variables constant. The forestland area increases by 0.87% with 1%
increase in the ration of industry to agricultural output, and increases by 0.39% with 1%
increase of the share for state-owned forest, controlling for other variables.

Forest stocking volume is significantly associated with forestland area, forest
ownership and forest quality. Results show that 1% increase in forestland area would
induce forest stocking volume to increase by 1.9%, holding other variables constant.
Forest stocking volume is greater if forest quality is higher. Forest stocking volume is
0.69% higher if there is one more percent of quality forest. Moreover, stocking volume
will increase by 0.46% if the share of state-owned forestland increased by 1%. In the
study region, natural forest is less disturbed by humans and has high stocking volume.
The Shelterbelt project has no significant effect on stocking volume, possibly because ten
years of project implementation are not long enough to improve the stocking volume.
Biophysical factor has significant effect on forest resource status. Forestland area

increases with elevation, while the stocking volume decreases when elevation increases”.

 

15 The sign of elevation coefficient differs from what is seen in the fractional logit model (where the effect
of elevation on the share of forestland is negative). The dependent variable is different. It is possible that
the forestland increases with elevation but its share decreases, because the area for other land uses (e.g.,
grassland) increases faster.

113

Timber production is significantly affected by stocking volume and socioeconomic
variables including log price index, highway density, and forest policies. Timber
production is higher if forest stocking volume is higher, and this association is
statistically significant. Higher log price index induces more timber production, and 1%
increase in log price index drives timber production to increase by 0.88%. A comparison
of this result with the price effect on forestland suggests that the short-run (e. g., current
year) forest management behavior (timber production) was driven by the market signals,
but price increase did not encourage forestland expansion that would only occur in the
long run. Thus, it becomes necessary to explore under what circumstances the market
signal can guide the long-run forest management. The Shelterbelt project and the NFPP
effectively reduced timber production in the study region”. The implementation of the
Shelterbelt project lowers the annual timber production by 8,900 m3, and the NFPP
reduces the annual production by about 25,000 m3. Road construction facilitates the
access to forestland and results in an increase in timber production. Holding other
variables constant, 1% increase of highway rate would lead its timber production to
increase by 0.3%. Thus, it is a challenge for the local government to balance the
advantage of infrastructural constructions with respect to poverty alleviation and the
possibility of resource exploitation due to transportation convenience.

In summary, the results show that forestland area, forest stocking volume and timber
production are inter-related to each other to determine the forest status under given
socioeconomic and institutional ’ conditions. Population pressure and insecure forest

ownership encourage the reduction of forestland area, while industry development

 

16 . . . . . . .
The Shelterbelt pr0ject had immediate effect on timber harvest, because of its restriction to forestland
(e.g., mountain closure), while the stocking volume does not increase significantly in short term.

114

benefits the forestland conservation. Forest stocking volume is positively linked to
forestland area and timber production. This implies that conserving the forestland is
crucial not only for environmental protection, but also for timber production. The effect
of road construction is positive, calling attention to balancing economic development
with resource protection. The effect of price signal is significant in timber production, but
not in forestland area The forest projects played a noticeable role in controlling timber
production and protecting the forest resources.

Forest resources have both environmental and production functions, but sometimes it
is hard to consider both functions when making land use decisions. An activity good for
production might be an impediment to environmental protection. Thus, it is needed to
integrate governmental support and regulations with market signals and access to manage
forests in a more sustainable matter.

5.4 Discussion of Model Results

This chapter develops and estimates structural models to examine the cropland and
forestland dynamics, respectively. Effects of driving forces from these models are mostly
consistent with findings from the fractional logit models in Chapter 4. Compared to the
fractional logit models presented in Chapter 4, the structural models better illustrate the
interactions in the land use processes, and they also better demonstrate various
perspectives of the driving forces.

Moreover, the structural models include more variables than the fractional logit
model, because of the non-land-use equations involved in structural models. Through the
interactions between the land use variables and other endogenous terms, we can

understand intricate effects of socioeconomic factors on land use decisions, although

115

these factors may not directly drive the LUCC. For example, increase in farmer’s income
encourages the agricultural technological progress that will help control cropland
expansion. Comparing results from cropland and forestland equations provides some
interesting information. Forestland area is negatively connected to the road construction,
but the cropland is positively so; forestland is negatively associated to population
expansion, but cropland is positively 80. All these indicate that the change in forestland at
least is partly related to the change in cropland. The limitations for these two structural
models will be discussed in Chapter 6. Also, Chapter 6 will discuss the details of policy

implications of the empirical results.

116

Chapter 6 Conclusion and Discussion

6.1 Overview

This dissertation research sets out to examine the driving forces of the land use/cover
changes (LUCC) along the upper Yangtze River in China. Using the LUCC data derived
from remotely sensed images and county—level socioeconomic and institutional data, it
not only demonstrates how the land use decision is affected by the population growth,
market signals, economic development, policies and institutional factors along the upper
Yangtze River, but also depicts the interactions in the land use process by structural
models. The cropland structural model illustrates how the cropland use interacts with one
of its driving forces, the agricultural technological progress, and how it is connected to
one of its environmental consequence — soil erosion. The forestland structural model
examines the forestland change in the context of its relationships to forest stocking
volume and timber production. Structural models developed in this research thus help
learn more about the complex interactions in land use process. The insights of the LUCC
driving forces are expected to provide viable suggestions that will help achieve a more
sustainable land use pattern by considering economic development and environmental
protection into land use decisions.

This chapter summarizes the findings from the dissertation research, outlines policy
implications of some LUCC driving forces that have significant effects, and discusses the
future research directions.

6.2 Summary of Research Findings
Both temporal and spatial changes of land uses/covers were observed in the study

region over 25 years. The area of cropland and forestland both decreased from 1975 to

117

_ u
‘1 flu..- ‘,o,l

2000, while the grassland area increased. Although the share for each land use did not
vary evidently over the study period, the land conversion matrix shows that land use
changes did exist, but conversion into and out of a land use simultaneously resulted in the
stable share for each land use over time in the study region. Moreover, the regional
LUCC varied during each time period. Thus, it is necessary to construct models using
local-level information to explain the spatiotemporal land use variations.

The land use pattern differed between the two provinces - Yunnan and Sichuan. For
example, the proportion of both cropland and forestland is higher in Yunnan than that in
Sichuan. The area of forestland in Sichuan increased over the years, while it dr0pped
slightly in Yunnan. Moreover, the land use pattern varied with altitude. Cropland is
mainly located in the low or medium altitude zones and grassland in high altitude zone,
while the forestland is commonly observed in the medium and high altitude zones.

Fractional logit and structural models provide insights on the complex LUCC process.
First, it is shown that socioeconomic and institutional factors perform critical roles in the
LUCC, in addition to the biophysical factor. Non-farming economic development helped
reduce cropland expansion and conserve forest resources, while population growth
contributed to deforestation and cropland expansion. Lowering grain quota levied on
farmers could mitigate cropland expansion, and stable forest tenure arrangement could
stabilize the forestland base. Road construction facilitated the market access and thus was
positively connected to the increase of cropland area. Second, it is already seen that
driving forces for each land use differ. Thus, multiple driving forces and their different
impacts on the LUCC make it necessary to understand the specific role of each potential

driver. Doing so will help prioritize or balance various policies in accomplishing a more.

118

satisfactory goal of land use. For example, industry development can attract labor out of
farming, reduce cropping area, and even promote forestland use, but such a labor transfer
may also cause concerns with the growth of idled cropland or lowered crop productivity.
The natural factor, such as elevation, has significant impact on land allocation. Therefore,
the constraint of natural conditions needs to be fully considered when implementing land
use policies.

Third, interactions among various facets of cropland and forestland use processes are
well illustrated by structural models, and importance of agricultural technological
progress and environmental policies are also highlighted. Cropland structural model
examines the cropland change in light of its interaction with agricultural technology and
environmental condition. Technological change, as reflected in fertilizer application,
irrigation and multiple cropping indices, contributes to increase of crop production and
decease of cropland use. Meanwhile, agricultural technology improvement is also
affected by the extent of cropland scarcity and other factors like market price and access.
Cropland expansion in the study region resulted in more erbded area; in turn, soil erosion
hampered cropland production, leading to more cropland use to maintain crop
production. The forestland change structural model was built in considering its
interrelationships with stocking volume and timber production. The increase of forestland
area contributes to the growth of forest stocks, and an increase in stocking volume is
positively connected to a higher level of timber production. However, timber production
can result in the reduction of forestland area, without adequate regeneration and

management.

119

In both structural models the effects of relevant policies and institutions are
noteworthy. The Shelterbelt Development Project and the Natural Forest Protection
Project effectively reduced timber production, and the decrease in timber production
helped control soil erosion. The Soil Conservation Project significantly reduced soil
erosion area, which benefited crop production. Finally, stable and clearly defined
forestland tenure resulted in the growth in both forestland area and forest stocks.

It should be noted that there are limitations in this research due to the data availability
and data quality”. First, land use data for the mid 1980s came from a digitized land cover
map, while the same data for other time points were from remotely sensed images. So,
the data of the mid 1980s may not be consistent with those for other time points. This
calls caution in interpreting spatial and temporal LUCC. It also may add measurement
errors to the dependent variable, influencing the significance of certain explanatory
factors. Moreover, classification errors exist in the land use data derived from remotely
sensed images. Thus, care should be taken when stating the magnitude and investigating
the drivers of LUCC.

Secondly, some explanatory variables are of concern. For one thing, the price and
cost data are provincial indices, which may have obscured the price and cost variations
and made it challenging to reflect their actual effects on the LUCC. Further, only the
average elevation of a county is used as a proximate biophysical factor, which hardly
captures the full variability of the natural conditions. More variables, such as slope, soil
property, temperature and landscape feature, should be considered in the future. The

equations of timber production and stocking volume incorporate only socioeconomic and

 

l7 . . . . . .
Other improvements on tlus research Will be discussed in the final sectron.

120

 

institutional variables. It will be interesting to create aggregate biophysical variables,
such as stand origin, species, and growth rate, and include them in the analysis.

Third, potential drivers for uses, such as grassland and shrub as a secondary forest
category, were not well specified. While this had to do with the current status of data
availability, more efforts should be made to address the issue and improve the modeling
of driving forces. Related, some other variables have missing values, which reduces the
number of observations for estimation. While the sample size is enough to explore the
LUCC determinants and related dimensions in this study, the multi-year and uneven data
intervals between neighboring time points make it hard to employ potential time lags to
deal with serial and spatial correlations effectively. If possible, the dataset should be
improved by covering a longer period of time with more frequent observations.

Fourth, also due to the limited number of observations, model validation was not
conducted. It is ideal that the whole dataset is divided into two parts, one for model
building and the other for validation, to ensure that the right model is employed to
represent the accurate results. For example of investigating changes on land uses/covers,
the land use conversion matrix within a period from some counties is applied to the
remaining counties to derive conversions among different land use, and then be compared
to the actual changes. In terms of modeling the driving forces, models are developed
using data of some counties, and then be applied to other counties to check the
significance of drivers or the overall model goodness of fit. Limited sample size in this
study makes model validation unrealistic. However, driving forces and their effects
identified from this research are mostly consistent with findings from other researchers,

which indirectly validated the model performance.

121

Nonetheless, this dissertation research makes a valuable contribution to the literature.
The study is successful in integrating the interactions of relevant variables and the
environmental feedbacks embedded in the land use process into the analysis. Not only
has such a study never been done for the study region, but also this type of structural
modeling of the LUCC determinants is rarely developed in the international literature.
The rich results have clearly demonstrated the benefit of studying the LUCC as coupled
human and natural processes.

6.3 Policy Implications

The research findings carry significant policy implications. It is revealed that the
LUCC is a result of the multiple driving forces playing out simultaneously and that the
same drivers can have different impacts on various land use decisions. These and other
results have generated numerous policy insights regarding how to allocate the scarce land
resources to various uses and how to incorporate effects of food production, economic
development, and environmental conservation into land use decisions.

First, agricultural technological progress is important to continued grain production
growth on limited cropland. As such, technological innovation and adoption should be
encouraged and supported, and actions should be taken persistently such as promoting
extension services, enhancing the distribution network of technology-embedded
agricultural inputs, or strengthening investment in agricultural research and development.

Second, the effects of industrial development on the LUCC should be well
recognized. Off-farming business and employment opportunities not only increase
farmers’ income, but also attract them out of farming and rural economy. These effects

help control cropland expansion and conserve forest resources. Also, an increased income

122

from off-farming opportunities will in turn promote the adoption and diffusion of
agricultural technologies. Thus, it is crucial to understand the complex interactions of
industrial development, labor transfer, technological change, and their potential of
collectively promoting more sustainable land uses.

Third, the role of market should be appreciated if it can exert its function in efficient
resource allocation. It is found that agricultural technological progress is directly driven
by the changes in input costs and output prices. Moreover, it is uncovered that grain and
livestock price changes impacted changes in forestland and cropland, although other price
variables did not have the expected effects on the LUCC. These cross effects of
commodity prices should be recognized. The fact that prices did not broadly affect land
use decisions in this past is partially because those prices were not real market signals
and real prices were heavily controlled by the government. Unfortunately, such distorted
and depressed prices could not adequately guide long-term land use decisions. Therefore,
it is desirable to introduce the market mechanism and enhance the incentive structure to
advance more efficient commodity production.

Fourth, govemment—sponsored projects and policies can be important and effective
instruments in conserving the environment. The Shelterbelt Development Project and the
NFPP have significantly reduced timber harvesting and protected forest resources in the
study region; similarly, the Soil Conservation Project have contributed to controlling soil
erosions and improving vegetation cover and cropland productivity. So, the government
should continue its support for ecosystem restoration and environmental protection.

Meanwhile, it must be noted that without a strong and lasting complement of other

123

drivers like market prices, technological advancement, land tenure, and land use
regulation, these policies alone may not be able to make a large and sustainable impact.

Fifth, it is well known that clearly defined tenure arrangement encourages long-term
planning and protection of forest resources (Chomitz 2006). State ownership represents a
stable forest tenure, which has reduced the possibility of forest conversion. However,
unclear beneficiaries of collective forests and distorted market prices discouraged farmers
from forest investment and management, and consequently the collective forests were
more likely to be degraded and even converted to other uses. These results imply that it is
essential to implement tenure reforms for the collective forests in order to slow down or
even reverse forest conversion to other uses and increase investment on the collective
forests. Reform might include the clarification of use and benefit rights, the creation of a
well-functioning monitoring and enforcement system, and the dissemination of
transparent and fair market information to the local forest managers (Yin et al. 2003). It
needs to be noted that this does not mean that the current forest management on the state
forest is a good example for the collective forest.

Finally, it is worth noting that it is uncommon to have win-win solutions. Some
policies formulated for economic development or food safety may lead to adverse
environmental consequences. Studying the LUCC driving forces and their cross effects
underscores the need for making land use decisions or policies after carefully considering
the trade-offs of alternative scenarios. For example, the food self-sufficiency policy as
part of the old planned economy was not conducive to efficient and sustainable land use
because the grain procurement quota disrupted the trade flows of agricultural products

across the nation and caused more land and other inputs to be used in crop production

124

(Lin 1992). It is unnecessary to meet food demand with local production for a region, like
our study site, that possesses poor farming conditions and limited cropland. With
abundant grassland and forestland, farmers should be facilitated to specialize in livestock
and forest industries and establish their comparative advantage in the marketplace.
6.4 Future Research Discussion

A significant advancement of our knowledge of the LUCC driving forces
notwithstanding, this dissertation research can be improved along several directions in the
future. First, while the LUCC from related human activities is regarded as a driver of
environmental and climate changes, a change in environment can in turn impact the land
use and cover changes through its effects on hydrological and terrestrial biological
systems (USGCRP 2003). Thus, how to integrate more biophysical components into the
land use change model is an interesting research topic. Such efforts will also help predict
the LUCC under various natural changes scenarios. Of course, integrating temporal
biophysical dynamics into the land use models requires not only interdisciplinary
cooperation, but also system modeling tools to incorporate interaction mechanism.

Second, due to the real-world complications, this research separately developed a
cropland structural model and a forestland structural model. The two structural models
can be regarded as ‘partial equilibrium model’ for the respective land use process, and
such an approach can actually be applied to other land uses, including grassland or urban
land uses. Moreover, once we have constructed the separate ‘partial equilibrium’
structural model for each land use, the next interesting research is to create the
connections among various structural models and generate a ‘general equilibrium’ system

for all the land uses. This also means to investigate the potential effect of a driver in one

125

land structural model on other land .uses. For example, the cropland structural model in
this research demonstrated the relationships among agricultural production, cropland use,
agricultural technology and environmental consequence. However, the cropland
structural model did not explicitly show the source or destination of the cropland change.
By building the linkage of different land use structural models, it is expected to explicate
the effects of driving forces on land conversions. Looking from another perspective, the
linkage between technological change and deforestation has been investigated (Angelsen
and Kaimowitz 2001). But in my forestland structural model, agricultural technology was
not included as one of potential driving forces, in light of the endogeneity of agricultural
technology in the cropland structural model. A system of several land use structural
models could help elucidate how agricultural technology affects forestland changes, or
how the cost changes of technological inputs influences forestland use. Therefore,
connecting structural models for different land uses will provide a more comprehensive
understanding of the complex land use processes.

Third, the LUCC possesses the feature of spatial complexity. The pattern and the
intensity of the LUCC may be different if examined at different scales (e.g., county scale
or household scale). On the one hand, a large scale LUCC analysis sometimes obscures
the variation at the finer scale; on the other hand, the change pattern that emerges at large
scales actually results from the environmental, economic or land use changes at a finer
scale that, however, is not explained or exhibited with finer scale dynamics (Parks et al.
2002). Thus, it would be beneficial for the future research to integrate the LUCC studies
at different scales and thus capture the spatial cross-scale, emergence and complexity in

the land use change process. Moreover, the effects of policy and institutional factors may

126

 

trifl] :
rIIItlll {j

vary at different scales. Therefore, examining the LUCC driving forces at various scales
will help policies target at relevant population, location, and activity to increase their
effectiveness.

Finally, this research focused on explaining the driving forces of the LUCC; less
attention was devoted to exploring how these driving forces impact the future land uses
and land covers. Thus, it would be interesting to make projections of the future LUCC
under various biophysical, socioeconomic, and institutional scenarios. Such projections
can be integrated into relevant land planning projects, or help evaluate the long term

effects of policies on the land uses.

127

Appendices

Appendix 3.1: Technical Notes on Classification of Remotely Sensed Images

To the extent possible, cloud-free images were selected consistently to match the
growing season to avoid misclassification of changes due to phonological differences.
Fifteen scenes were needed at each time point to cover the whole study area. Remotely
sensed images were geometrically corrected and geo-referenced, ahead of interpretation
and classification. The land use classification was conducted with the supervised
approach of computer-human interactive interpretation. Under this approach,
representative spots of known cover types, such as ground truth points, were delineated in
a satellite image and the statistical properties of these "training sites" were used to
classify the entire scene. In addition to ground truth points, landscape photos and
vegetation maps were also used to validate the classification. The test procedure shows
that the classification accuracy rate for 1975 data is 88 percent; the accuracy rate for
1990, 1995, and 2000 is 92.9 percent, 98.4 percent, and 97.5 percent, respectively (C AS
2006). Accuracy rate for primary land use category is available for 1990 and 1999 data,

and listed below:

 

 

 

 

 

 

 

 

 

 

 

 

Primary category 1990 ' 1999
Cropland 94.94 97.33
Forestland 90.13 97.42
Grassland 88.16 97.43
Build up 96.32 99.15
Water 96.67
Other 9572 98.36
Overall 92.9 97.5

 

Source: IGSNR, CAS.

128

 

Appendix 4.1: Testing Estimation of the Fractional Logit Model for Primary Land
Uses

This test is to examine how the high correlation between price indices may affect
their coefficient estimates as well as those of explanatory variables in the fractional logit
model. In the test, only one price index variable is included for each equation and that
variable is selected based on the significance level in the model and its economic
rationale (e. g., including livestock price index in the grassland equation).

When keeping one price index, the signs and significance levels of other explanatory
variables are almost the same as when including all three price indices. Thus, the effects
of other significant variables will not deviate. However, the high correlation between
price indices does affect the coefficient significance of the price index variables,
especially in the cropland equation. Thus, caution is needed to interpret the effects of

price signals on the LUCC.

129

Table A-4.1 TestinLEstimated Results of the Fractional Logit Model

 

 

 

Explanatory Variables Cropland Forestland Grassland
Grain Price Index -0.235
(0.092)**
Log Price Index
Livestock Price Index 0.362 0.489
(0.345) (0.428)
Industry/Agricultural Output -0.052 0.067 -0.11 1
(0.01 l)** * (0.026)** * (0.059)*
Highway Rate 0.204 0.017 0.113
(0.077)** * (0.052) (0.135)
Population Density 0.486 -0.334 0.154
(0.028)*** (0.081)*** (0.136)
Per Capita Grain Quota 0.030 —0.007 -0.011
(0.017)* (0.010) (0.032)
Share of State-Owned Forest -0.135 0.116 -0.133
(0.085) (0.072)* (0.107)
Elevation -1.239 -0.636 1.230
(0.221)*** (0.139)*** (0.331)***
Province Dummy -0.157 0.126 -0. 158
(0.071 )** (0.038)*** (0.064)**
1975 Dummy 0.163 -0.048 0.128
(0.097)* (0.022)“ (0.1 19)
1985 Dummy 0.128 -0.042 0.1 1 1
(0.091) (0.021) (0.109)
1990 Dummy 0.109 -0.038 0.096
(0.072) (0.018)** (0.091)
1995 Dummy 0.054 -0.016 0.046
(0.035) (0.010) (0.043)
Observations 122 122 122
d.f. 109 109 109
Log-likelihood -34.217 -54.91 1 -47.502
Pearson Chi Square 1.059 5.238 6.232

 

Note: 1. “*”, “**” and “***” represent 10, 5 and 1 percent significance level, respectively.
2. Numbers in parentheses are standard error of the coefficient.

130

Appendix 5.1: Year and Province Dummy Estimate for the Cropland Structural
Model

 

Grain Production Cropland Technology Soil Erosion

 

Province Dummy -0.275 0.046 -0.077 -1610
(0.063)** (0.169) (0.033)** (184.02)***
1985 Dummy 0.002 1.957 0.766 -556
(0.053) (1.151)* (0.356)” (284.23)** ‘
1990 Dummy -0.043 1.548 0.705 -102
(0.046) (0.823)* (0.316)** (235.79)
1995 Dummy 0.002 0.858 0.353 16.43
(0.040) (0.398)** (0.179)** (235.05)

 

Note: 1. Numbers (above parentheses) listed here are variable coefficients, not elasticity.
2. Numbers in parentheses are standard error of the coefficient.
3. Because of one lagged explanatory variable, all observations of year 1975 is not
used for estimation. Thus, only 3 yearly dummies (year of 2000 is the base year).

131

Appendix 5.2: Coefficients and SD. for the Forestland Structural Model

 

 

. Forestland Timber Stocking
Explanatory Varrables Area Production volume
149.54
Forestland Area (1 1 .00)“ at:
-2.737
Timber Production (2.633)
0.808
Stocking volume (0099):”:4:
17,475
Grain/log price index (29,709)
18,172
Log Price index (7,971)“:
Indus /A 'cultural 625,747
outputtry grr (159,723)***
. 12,370 -8,933 3,806,651
Shelterbelt Preject (47,310) (1025):: (3,475,549)
-1 13,718 -24,589 -2,166,357
NFPP (79,900) (6,827)*** (3,876,605)
1 8,200,000
Quality Forest (8,963,319)**
Share of State-owned 231,823 -14,528 21,400,000
Forest (98,180)" (10,062) (705,1929)***
_ —20,900,000 3,707,700
Highway Rate (20,800,000) (1,905,725)*
-121,230
Population Density (62,091 )4":
' 161,726 -15,430 7,377,691
Provrnce Dummy (43,794)*** (4,940)*** (3,572,312)“
(87,828) (13,360)** (5,207,404)
9,067 13,283 -3,586,568
1985 Dummy (42,968) (5,170)*** (3,864,897)
-14,302 8,174 -474,293
1990 Dummy (57,479) (4,920)* (3,580,153)

 

Notezl. “*”, “**” and “***” represent 10, 5 and 1 percent significance level, respectively.

2. Year of 1995 is taken as the base year, because the year 2000 represents the

NFPP implementation.

132

BIBLIOGRAPHY

Agarwal, C., G. M. Green, J. M. Grove, T. Evans, and C. Schweik. 2002. A Review and
Assessment of Land-Use Change Models: Dynamics of Space, Time, and Human
Choice. Gen. Tech. Rep. NE-297. Newton Square, PA: U.S. Department of
Agriculture, Forest Service, Northeastern Research Station.

Agricultural and Rural Development Taskforce. 2004. China’s Agricultural and Rural
Development in the New Era: Challenges, Opportunities and Policy Options.
Beijing: China Council for International Cooperation on Environment and
Development.

Ahn, S., Plantinga, A. 1., Alig, R.J.. 2000. Predicting future forestland area: a comparison
of econometric approaches. Forest Sciences 46(3), 363-376.

Angelsen, A. 1999. Agricultural expansion and deforestation: modeling the impact of
population, market forces and property rights. Journal of Development Economics
58, 185-218.

Angelsen, A and Kaimowitz, D. 2001. Agricultural technologies and tropical
deforestation. Oxon, UK, CABI Publishing

Alig, J.R., Adams M.D., and McCarl, AB. 1998. Impacts of incorporating land
exchanges between forestry and agriculture in sector models. Journal of
Agricultural and Applied Economics 30(2), 389-401.

Barlowe, R. 1958. Land Resource Economics: The Pelitical Economy of Rural and
Urban land Resource Use. Englewood Cliffs, N.J., Prentice-Hall, Inc.

Besley, T., 1995. Property rights and investment incentives. Journal of Political Economy
103, 913—37.

Caldas, M., Walker, R. and Perz, S. 2002. Small Producer Deforestation in the Brazilian
Amazon: Integrating Household Structure and Economic Circumstance in
Behavioral Explanation. Working Paper No.96, Center for International
Development, Harvard University.

Chinese Academy of Sciences. 2006. Methods and procedures of classifying and
interpreting the remotely sensed images for the J insha River region.

Cattaneo, A. 2001. Deforestation in the Brazilian Amazon: comparing the impacts of

macroeconomic shocks, land tenure, and technological change. Land Economics
77 (2), 219-240.

133

Chomitz, K.M., D.A. Gray., 1996. Roads, land use, and deforestation: a spatial model
applied to Belize. World Bank Economic Review 10(3), 487-512.

Chomitz, K. 2006. At Loggerheads? Agricultural Expansion, Poverty Reduction, and
Environment in the Tropical Forests. Washington, DC: World Bank.

Conservation International, 2002. Biodiversity Hotspots.
(http://wwwbiodiversityhotspots.org/xp/hotspots/China)

Cropper, M. and Griffiths, C. 1994. The Interaction of population growth and
environmental quality. American Economic Review 84(2), 250-254.

Davis, LE. and North, DC. 1971. Institutional Change and Economic Growth.
Cambridge, UK: Cambridge University Press.

Deininger, K. and Minten, B. 1997. Determinants of Forest Cover and the Economics of
Protection: an Application to Mexico. Washington, DC: World Bank.

Du, SF, 2001. Environmental Economics. Encyclopedia Press, Beijing.

Crook, F.W., 1991. Underreporting China’s grain area: a short story. China: Agriculture
and Trade Report. RS-91-3, Economic Research Service, United States
Department of Agriculture, July.

Evans, P.T., Manire, A., Castro, D.F., Brondizio, E. and McCracken, S. 2001. A
Dynamic model of household decision-making and parcel level landcover change
in the eastern Amazon. Ecological Modeling 143, 95-113.

Fischer, G., Sun, L.X., 2001. Model based analysis of future land-use development in
China. Agriculture, Ecosystems and Environment 85, 163-176.

Fitz,H.C., Debellevue, EB. and Costanza, R. 1996. Deve10pment of a general ecosystem
model for a range of scales and ecosystems. Ecological Modelling 88, 263-295.

Geoghegan, J., Villar, S.C., Klepeis, P., Mendoza, P.M., Himmelberger, Y.O.,
Chowdhury, R.R., Turner H, B.L., Vance, C., 2001. Modeling tropical
deforestation in the southern Yucatan Peninsular region: comparing survey and
satellite data. Agriculture, Ecosystems and Environment 85, 25-46.

Higgitt, D.L., Lu, X.X., 1999. Challenges in relating land use to sediment yield in the Upper
Yangtze. Hydrobiologia 410 (0), 269-277.

Irwin, GE. and Geoghegan, J. 2001. Theory, data, methods: developing spatially

explicitly economic models of land use change. Agriculture, Ecosystems and
Environment 85, 7-23.

134

 

Jacoby, H.G., Li, G., Rozelle, S., 2002. Hazards of expropriation: tenure insecurity and
investment in rural China. American Economic Review 92 (5), 1420-1447.

Ji, C.Y., Liu, Q., Sun, D., Wang, S., Lin, R, Li, X., 2001. Monitoring urban expansion
with remote sensing in China. International Journal of Geographical Information
system 22 (8), 1441-1455.

IPCC (Intergovernmental Panel on Climate Change). 2000. Special Report on Land Use,
Land Use Change and Forestry. New York: UNEP. '

Kaimowitz, D. and Angelsen, A., 1998. Economic Models of Tropical Deforestation: A
Review. Center for International Forestry Research, Bogor, Indonesia.

Lambin, E.F., J .H. Geist and E. Lepers. 2003. Dynamics of land-use and land-cover

change in Tropical Regions. Annual Review of Environment and Resources 28,
205-241.

Li, X., Peterson, J .A., Liu, G., Qian, L., 2001. Assessing regional sustainability: the case
of land use and land cover change in the Middle Yiluo Catchment of the Yellow
River Basin, China. Applied Geography 21, 87-106.

Lin, J.Y. 1992. Rural Reforms and Agricultural Growth in China. American Economic
Review 82 (1), 34-51.

Liu, J .Y., Liu, M.L., Zhuang, D.F., Zhang, Z.X., Deng, X.Z., 2003. Study on spatial
pattern of land-use change in China during 1995-2000. Science in China (Series
D) 46(4), 373-384.

Loucks, C.J., L11, 2., Dinerstein, E., Wang, H., Olson, D.M., Zhu, C.Q., Wang, D.J.,
2001. Giant pandas in a changing landscape. Science 294, 1465.

Lu, W.M., Landell-Mills, N., Liu J.L., Xu J .T. and Liu C. 2002. Getting the private sector
to work for the public good: instruments for sustainable private sector forestry in
China. Instruments for sustainable private sector forestry series. International
Institute for Environment and Development, London.

Lu, X.X. 2005. Spatial variability and temporal change of water discharge and sediment
flux in the lower Jinsha tributary: impact of environmental changes. River
Research and Applications 21 (2-3), 229-243.

Ma, X.H., 1998, Forest functions on preventing floods in Upper Yangtze River Basin.
Forest Technology Bulletin, 10, 44-45 (in Chinese).

Mamingi, N., Chomitz, K.M., Gray, D.A., and Lambin, E. 1996. Spatial Patterns of

Deforestation in Cameroon and Zaire. Poverty, Environment, and Growth
(Working Paper 8). Washington, DC: World Bank, Policy Research Department.

135

 

 

Manson, MS. and Evans, T. 2007. Agent-based modeling of deforestation in southern
Yucatan, Mexico, and reforestation in the Midwest United State. PNAS 104 (52),
20678-20683.

McCracken, S.D., Brondizio, E.S., Nelson, D., Moran, E.F., Siqueira, A.D., Rodriguez-
Pedraza, C., 1999. Remote sensing and GIS at farm property level: demography
an deforestation in the Brazilian Amazon. Photograrnmetric Engineering and
Remote Sensing 65 (l 1), 1311-1320.

Mertens, B., Sunderlin, W.D., Ndoye, O., Lambin, E.F., 2000. Impact of macroeconomic
change on deforestation in South Cameroon: Integration of household survey and
remotely sensed data. World Development 28(6), 983-999.

Mendelsohn, R. 1994. Property rights and tropical deforestation. Oxford Economic
Papers 46 (1994), 750-756.

Miller, J.D., Plantinga, J. A., 1999. Modeling land use decision with aggregated data.
American Journal of Agricultural Economics 81, 180-194.

Mi'rller, D. and Zeller, M., 2002. Land use dynamics in the central highlands of Vietnam:
a spatial model combining village survey data with satellite imagery
interpretation. Agricultural Economics 27, 333-354.

Munroe, D.K., York, A.M., 2003. Jobs, houses and trees: changing regional structure,
local land-use patterns, and forest cover in Southern Indiana. Growth and Change
34 (3), 299-320.

National Research Council. 2001. The Science of Regional and Global Change: Putting
Knowledge to Work. Report of the Committee on Global Change Research.
Washington, DC: National Academy Press.

Nelson, C. G., Hellerstein, D., 1997. Do roads cause deforestation? Using satellite image
in econometric analysis of land use. American Journal of Agricultural Economics

79, 80-88.

 

Nelson, C.G., V. Harris and S. W. Stone. 1999. Spatial econometric analysis and project

evaluation: modeling land use change in the Darien. Working paper.

Pan, J .G., 1999, The Characteristics of water runoff and suspended sediment along the
Jinsha River. Journal of Sediment Research 2.

Parks, J .P., Barbier, BE, and Burgess, C]. 1998, The Economics of Forest Land Use in

Temperate and Tropical Areas, Environment and Resource Economics 11(3-4),
473-487.

136

Parks, J .P., Hardie, W.l., Tedder, AC. and Wear, ND. 2000. Environmental Monitoring
and Assessment 63, 175-185.

Parker C.D., Berger, T. and Manson M. S. 2002. Agent-Based Models of Land-Use and
Land-Cover Change. Report and Review of an International Workshop, October
4-7, 2001, Irvine, California, USA.

Pfaff, A. 1999. What drives deforestation in the Brazilian Amazon? Evidence from
satellite and socioeconomic data. Journal of Environmental Economics and
Management 37, 26-43.

Plantinga, A.J., 1996. The effects of agricultural policies on land use and environmental
quality. American Journal of Agricultural Economics 78, 1082-1091.

Pontius Jr, G.R., Cornell, DJ. and Hall, A.S.C. 2001, Modeling the spatial pattern of
land-use change with GEOMOD2: application and validation for Costa Rica,
Agriculture, Ecosystems and Environment 85, 191-203.

Ruttan, V. W. and Y. Hayami. 1984. Toward a theory of induced technical change.
The Journal of Development Studies 20(4): 203-22.

Ruttan, V.W., 2001. Technology, Growth, and Development: An Induced Innovation
Perspective. Oxford University Press, Oxford, UK.

Sichuan Statistics Yearbook. 1975-2003. Sichuan Statistics Press, Chengdu, Sichuan.

Shively, G. and Fisher, M. 2004. Smallholder labor and deforestation: A systems
approach. American Journal of Agricultural Economics 86(5), 1361-1366.

Shively, G. 2001. Poverty, Consumption Risk and Soil Conservation. Journal of
Development Economics 65, 267-290.

Sunding, D., and Zilberman, D. 2000. Research and technology adoption in a changing
agricultural sector. Draft for the Handbook of Agricultural Economics.

Tietenberg, T. 2000. Environmental and Natural Resources Economics (5"n ed.) Prentice
Hall.

Turner, B.L., Moss, RH, and Skole, D.L., 1993 Relating land use and global land cover
change: A proposal for an ICBP-HDP core project. Report from the IGBP-HDP
Working Group on Land Use/Land Cover Change. Stockholm: Royal Swedish
Academy of Science.

Turner H, B.L., Meyer, B. W., Skole, L. D., 1994. Global land use/land cover change:
towards an integrated study. AMBIO 23 (1), 91-94.

137

 

UNESCO. 2003. World Heritage List (http://whc.unesco.org/sites/1083-loc.htm).

USGCRP. 2004. Land use / land cover change: USGCRP program element
(http://www.usgcrp.gov/usgcrp/ProgramElements/land.htm).

Veldkamp, A and LO. Fresco. 1996. CLUE: a Conceptual Model to Study the
Conversion of Land Use and its Effects. Ecological Modeling 85: 253—270

Veldkamp, A., and E. F. Lambin. 2001. Predicting land-use change. Agriculture,
Ecosystems and Environment 85: 1-6.

Verburg, RH. and Chen, Y.Q. 2000. Multiscale Characterization of Land Use Patterns in
China. Ecosystems 3, 369-385

Verburg, RH. and Veldkamp, T. 2001, The Role of Spatially Explicit Models in Land-
use Change Research: a Case Study for Cropping Patterns in China. Agriculture,
Ecosystems and Environment 85, 177-190.

Verburg, P.H., Veldkamp,W.S.A., Espaldon, R.L.V., Mastura, S.S.A., 2002. Modeling
the spatial dynamics of regional land use: the CLUE-S model. Environmental
Management 30(3), 391-405.

Walsh, J. S., Evans, P. T., Welsh, F. W., Entwiste, B., and Rindfuss, R. 1999. Scale-
Dependent Relationships between Population and Environment in Northeastern
Thailand. Photogrammetric Engineering and Remote Sensing 65 (1), 97—105.

Walker, R., Moran, E. and Anselin, L. 2000. Deforestation and cattle ranching in the
Brazilian Amazon: external capital and household processes. World
Development 28(4), 683-699.

Wang, X.T., 2003. Building an Ecological Shield along the Upper Yangtze River:
Priorities and Measures. China Agriculture Publishing House, Beijing, P.R.China.

Wang, Q. and Otsubo, K. 2001. Relationship between Agricultural Land-use Change and
Socioeconomic Factors in Northern China. Working Paper, National Institute for
Environmental Studies, Tsukuba, Japan.

Weng Q., 2001. Modeling urban grth effects on surface runoff with the integration of
remote sensing and GIS. Environmental Management 28 (6), 737 —748.

Wooldridge, J .M., 2002. Econometric Analysis of Cross Section and Panel Data. The
MIT press, Cambridge, Massschusetts.

Xu, J .T., Katsigris, E., White, T.A., 2002. (ed) Implementing the Natural Forest
Protection Program and the Sloping Land Conversion Program: Lessons and
Policy Recommendations. China Forestry Press, Beijing.

138

 

Xu, J., Yin, R., Li, Z., Liu, C., 2007. China's ecological rehabilitation: progress and
challenges. Ecological Economics 57(4), 595-607.

Yang, Z. and Liang, L., 2004. Soil Erosion under Different Land Use Types and Zones of
Jinsha River Basin in Yunnan Province,China. Journal of Mountain Science 1, 46-
56.

Yeh, AG. and Li, X., 1998. Suitable land development model for rapid growth areas
Using GIS. International Journal of Geographical Information system 12 (2), 169-
189.

Yin, H. and Li, C., 2001. Human impact on floods and flood disasters on the Yangtze
River, Geomorphology 41 (2-3), 105-109.

Yin, RS. 1998. Forestry and the environment in China: the current situation and strategic
choices. World Development 26 (12), 2153—2167.

Yin, RS. and Hyde, W.F.. 2000. Trees as an agriculture sustaining activity: the case of
northern China. Agroforestry Systems 50: 179-194.

Yin, RS. and Newman, DH. 1997. Long run timber supply and the economics of timber
production. Forest Science 43:113-120.

Yin, R.S., Xu, J.T. and Li, Z. 2003. Building Institutions for Markets: Experience and
Lessons from China's Rural Forest Sector. Environment, Development, and
Sustainability 5, 333-351.

Yu, X., Jiang, J., Dou, H., Yang G., 1999. The warning of the 98’s Yangze deluge &
flood control strategy. CAS Bulletin 13 (1).

Yunnan Statistics Yearbook. 1975-2003. Yunnan Statistics Press. Kunming, Yunnan.

Zhang, P.C., G.F. Shao, D.D.Le Master, G.R. Parker, J .B. Dunning Jr. and Q.L, Li. 2000.
China’s forest policy for the 21St century. Science 288, 2135-2136.

Zhang,Y.Q., Uusivuori, J. and Kuuluvainen, J. 2000. Econometric analysis of the causes
of forest land use changes in Hainan, China. Canadian Journal of Forest Research
30,1913-1921.

Zhang, X., Mount, T. D., Boisvert, RN, 2003. Industrialization, urbanization and land

use in China. IFPRI, Environment and Production Technology Division Discussion Paper
No. 58.

139

 

   

"71111111111111filljlllllllll‘“

489