Hohenheim University Institute of Plant Production and Agroecology in the Tropics and Subtropics

(380)

Nguyen, Thi Thanh

M.Sc. Thesis Assessment of land cover change in Chieng Khoi Commune, Northern Vietnam, by combining remote sensing tools and historical local knowledge Supervisors Prof. Dr. Georg Cadisch (Institute of Plant Production and Agroecology in the Tropics and Subtropics) Prof. Dr. Karl Stahr (Institute of Soil Science and Land Evaluation) Dr. Carsten Marohn (Institute of Plant Production and Agroecology in the Tropics and Subtropics)

Garbenstraße 13, D-70599 , Stuttgart – Germany

September 2009

ii

Abstract

In mountainous areas of Northern Vietnam, increasing population density has

forced agricultural production to expand into upland areas. This trend resulted in

decreasing forest resources with associated soil erosion and resource degradation.

The reduction of natural forests with a conjoint increase of forestry plantations and

a replacement of upland rice-based swiddening farming with continuous maize

cropping systems are threatening the sustainability of local land use systems.

Although annual land uses provide increased short-term cash income, food

security is facing long-term problems due to severe soil degradation.

Reconstructing past land use dynamics is crucial to understanding the present

situation in the research area and to find ways towards sustainable land use.

However, access to classical sources of land use data such as maps and high

resolution satellite imagery is difficult in a border region that in the recent past has

been affected by military operations.

The aim of this study was to classify land cover changes based on available

moderate resolution remote sensing imagery and historical local knowledge as a

basis for understanding and future modelling of land use dynamics.

This master thesis was conducted within the framework of the Uplands Program

SFB 564, Subproject C4. To analyse the change of land cover in Chieng Khoi

commune, Northwest Vietnam, from 1993 to 2007, LANDSAT and LISS III images

were used, ENVI 4.3 and ArcGIS 9.3 software were employed to digitize aerial and

satellite images and overlay field data collected with a handheld GPS. To ensure a

consistent dataset, image-based land cover classification was combined and

cross-checked with several data sources: Inventories of land cover, a decision-tree

based on land suitability and cropping season for different crops, participatory soil

maps and local stakeholder interviews regarding land cover history of plots

representing different slope positions.

This study provides an overview of land cover change in Chieng Khoi commune

iii

based on historical knowledge and generating land cover maps in 1993, 1999 and

2007. Hybrid classification was used, with overall accuracies achieved in 1993 of

81.1%, 1999 of 98.5% and 2007 of 82.5%. This method was also applied to

distinguish upland crops in 2007, overall accuracy obtained was 66.7%. These

results may provide input data for the biophysical model LUCIA (Land use Change

Impact Assessment tool) to enable reverse modelling and an ex-ante assessment

of land cover change consequences at the landscape level.

Keywords: Northwest Vietnam, land cover, satellite imagery, historical knowledge

iv

Author’s declaration

I, Nguyen, Thi Thanh (matriculation number: 439794, University of Hohenheim), hereby affirm that I have written this Master Thesis titled: “Assessment of land cover change in Chieng Khoi Commune, Northern Vietnam, by combining remote sensing tools and historical local knowledge”, independently and by

myself.

All authors mentioned in this work have been cited and no work has been included

in the Thesis without the authors being listed. This Master Thesis was not

submitted in the framework of any other examination process.

Stuttgart, _____________________ Signature______________________

v

Approval This thesis has been accredited for final submission with our approval as University Supervisors. Prof. Dr. Georg Cadisch (Institute for Plant Production and Agroecology in the Tropics and Subtropics) Signed: ________________________ Date: __________________________ Prof. Dr. Karl Stahr (Institute of Soil Science and Land Evaluation) Signed: ________________________ Date: __________________________

vi

Acknowledgements

Firstly, I would like to give many thanks to my first supervisor, Prof. Dr. Georg

Cadisch for the opportunity to work on interesting topic, fruitful ideas and

encouragement during the study. And I also would like to thank Prof. Dr. Karl

Stahr for accepting to be my second supervisor. To express heartfelt thanks to Dr.

Carsten Marohn for a lot of his time for great knowledge and skills on field works,

so helpful guide and revising report, also for running around of technical troubles

and the patience during my study.

I sincerely thank Eiselen Foundation, Ulm, Germany for financial supporting for my

studying in Germany in the first year and thesis field work expenses, thank to

subproject C4, Upland Program of Hohenheim University, funded by the Deutche

Forschungsgemeinschaft, Germany (DFG) for addition funds for field data

collection. I am grateful to Ford Foundation and Center for Educational Exchange

with Viet Nam (CEEVN), USA for financial support and the coordination my study

in Germany. Thank you very much.

I am also grateful to Dr. Thomas Hilger who provided me much support on advices

and comments. To members of Institute 380a, Melvin Lippe, Petra Schmitter,

Yohannes Zergaw Ayanu and Irene Unoma Chukwumah, I would like to thank

sincerely for very helpful data, and knowledge, special the encouragement and

cooperation.

The big thanks to Dr. Tran Duc Vien, Director and Nguyen Thanh Lam, Executive

Director of CARES for the acceptance and equipments support in Vietnam, to

specialists of RS and GIS in CARES: Pham Tien Dat, Tran Trung Kien, Vo Huu

Cong who were willing to exchange their knowledge and efficient online support.

vii

Also to friendly colleagues of Upland Program in Vietnam and Germany, Dr

Gerhard Clemens, Nguyen Huy Thuy, Moritz Koch, Nadja Reinhardt, Nguyen Dinh

Cong and all students have worked in Yen Chau, thanks for great knowledge of

beautiful soil profiles, great management of transportation and fruitful time in Yen

Chau. And I could never forget excellent experiment, information, assistance from

Lo Van Chung and farmers in Ban Me and Ban Tum, Chieng Khoi commune.

I express my sincere gratitude to my family, Mr Nguyen Trung Dung and Mrs Tran

Thi Tam my parents, my sisters, brother and my grandmother who are always

beside me anytime, and anywhere.

Nguyen Thi Thanh, Hohenheim University, September 2009

viii

Table of Contents Abstract .................................................................................................................... ii Author’s declaration ................................................................................................ iv Approval ...................................................................................................................v Acknowledgements .................................................................................................vi Table of Contents.................................................................................................. viii List of Abbreviations................................................................................................ ix List of Figures...........................................................................................................x List of Tables...........................................................................................................xi List of Appendices.................................................................................................. xii 1 Introduction ...................................................................................................... 1

1.1 Background............................................................................................... 3 1.2 Hypotheses............................................................................................... 3 1.3 Objectives of the study.............................................................................. 4

2 Literature Review ............................................................................................. 5 2.1 Land use and land cover change in upland Northern Vietnam ................. 5 2.2 Remote sensing - GIS and land cover mapping ....................................... 6

3 Materials and Methods..................................................................................... 8 3.1 Study area ................................................................................................ 8 3.2 Field survey and interviews..................................................................... 10 3.3 Image processing and classification ....................................................... 14 3.4 Detection of land cover change .............................................................. 24

4 Results ........................................................................................................... 25 4.1 Land cover change as witnessed by local farmers ................................. 25

4.1.1 History of land cover change ........................................................... 25 4.1.2 Changes in local soil type and soil fertility ....................................... 27

4.2 Land cover change analysis based on classification of satellite images. 28 4.2.1 Separability analysis ........................................................................ 29 4.2.2 Combination of remote sensing and interview data ......................... 31

4.3 Land cover map, statistics and land cover change ................................. 33 4.4 Classification accuracy ........................................................................... 41

5 Discussion...................................................................................................... 41 5.1 Validity of data ........................................................................................ 41

5.1.1 Expansion of upland agriculture ...................................................... 41 5.1.2 Forest cover..................................................................................... 42

5.2 Limitations and potentials to use satellite imagery combined with data deriving from local farmers ................................................................................ 45

5.2.1 Limitations of classification satellite images..................................... 45 5.2.2 Limitations of using interview data ................................................... 46 5.2.3 Potentials of combining satellite images and farmers’ knowledge ... 47

5.3 Hypotheses posed .................................................................................. 48 6 Conclusions ................................................................................................... 48 7 References..................................................................................................... 51 Appendices ........................................................................................................... 55

ix

List of Abbreviations BD Bhattacharrya Distance

CARES Center for Agricultural Research and Ecological Studies, Hanoi

University of Agriculture

COSAMD Centre of Mapping, Dept. of Survey and Mapping, Hanoi, Vietnam

DEM Digital Elevation Model

ETM Enhanced Thematic Mapper

GIS Geographic Information System

GPS Global Positioning System

GTPs Ground Truthing Points

IRS Indian Remote Sensing

LISS Linear Imaging and Self Scanning sensor

LUCIA Land Use Change Impact Assessment tool

NDVI Normalized Difference Vegetation Index

NRSA National Remote Sensing Agency, India

PAN Panchromatic

RS Remote Sensing

TD Transformed Divergence

TM Thematic Mapper

USGS United States Geological Survey

UTM Universal Transverse Mercator

WGS World Geodetic System

Yr Year

x

List of Figures Figure 1 Location of study site adopted and modified from Häring (2009)................9

Figure 2 Group discussion...............................................................................................13

Figure 3 Process of classification ...................................................................................21

Figure 4 Local soil map, adopted and modified from Clemens et al. (2008)...........23

Figure 5 Decision – tree for land use classification .....................................................32

Figure 7 Land cover map 2007 based on LISS and classification by maximum

likelihood procedure in ENVI 4.3 & ArcGIS 9.3 .............................................................35

Figure 8 Land cover change from 1993 – 2007 in Chieng Khoi ................................37

Figure 9 Distribution of upland crops in Chieng Khoi commune, March 2007,

based on LISS imagery, local soil map and decision tree procedure carried out in

ENVI 4.3 & ArcGIS 9.3 ......................................................................................................40

Figure 10 Comparison of the annual land use report 2005 by the Chieng Khoi

People’s Committee and the classified satellite image 2007 ......................................44

xi

List of Tables Table 1 Images used for land cover classification, technical properties and

sources ................................................................................................................................15

Table 2 Criteria used for discrimination of land cover classes used for classified

satellite images ...................................................................................................................16

Table 3 Soil types and history of land cover based on farmers’ discussion ............26

Table 4 History of soil erosion problems and change of soil types ...........................28

Table 5 Separability of land cover classes 1993 (minimum of TD and BD) ............30

Table 6 Separability of land cover classes 1999 (minimum of TD and BD) ............30

Table 7 Separability of land cover class 2007 (minimum of TD and BD) ................30

Table 8 Land cover statistics...........................................................................................36

Table 9 Detection of land cover change 1993 - 1999, bold figures refer to class

areas without change ........................................................................................................38

Table 10 Areas of upland crops in Chieng Khoi, March 2007 classified LISS III

image....................................................................................................................................40

Table 11 Overall accuracies and Kappa coefficient of land cover classification

produced from basic supervised classification with five classes................................41

xii

List of Appendices Appendix 1 Discussion sheet ..........................................................................................55

Appendix 2 Field site description....................................................................................57

Appendix 3 Transect sheet..............................................................................................59

Appendix 4 Summary information for final group discussion ......................................60

Appendix 5 Original local soil map .................................................................................61

Appendix 6 Error matrix ...................................................................................................62

Appendix 7 Seasonal calendar .......................................................................................63

Appendix 8 Land cover map 2007 without application of the majority procedure ...64

Appendix 9 Workshop summary after each day ...........................................................65

1

1 Introduction

In Vietnam, lowland areas for agricultural land use are becoming scarcer and

scarcer because of the fast growing population since the American War 1975. For

this reason, agricultural production is increasingly practised on steep upland slopes

(Wezel et al., 2002). In addition, agricultural policies affected land use and

landscape at commune level after the “Doi moi” - Renovation in 1986 (Sikor and

Truong, 2002), which promoted small scale farm systems with diversified cropping

patterns on small plots. At the same time, government programmes in context with

decollectivisation of agricultural means of production and development of markets

allocated forestry land to private households. This was supposed to lead to an

increase in forest area (including natural forest, secondary forest, planted forest

and production forest) once the policy was implemented (Meyfroidt and Lambin,

2008).

In Chieng Khoi commune, Yen Chau district in Northern Vietnam, maize extended

since 1990 due to availability of new varieties, improved maize yields and better

market access. This led to significant changes in farming systems towards maize-

based, mainly monocultural systems (Keil, 2008). The expansion of agricultural

area led to drastic changes in local vegetation at the cost of forest cover. Periods

of deforestation during the last two decades were closely related to increasing

problems of erosion and soil depletion in natural reserve areas. In addition, fallow

periods were drastically reduced or even completely omitted. From initially 5-7

years of fallow in a crop rotation, a maximum of 3 years remained. This caused a

continuous decline of soil fertility; organic matter and nitrogen contents decreased

in cultivated fields (Wezel et al., 2002). Until the late 1980s, the Black Thai ethnic

group in Chieng Khoi had practised composite swiddening farming. The system

combined cultivation of wet rice fields in lowlands with rotational swiddening on the

upper slope positions with distinct fallow periods. Nowadays, this system has

2

almost vanished, and been replaced almost completely by monocropping without

crop rotation or intercrop and agroforestry production systems (Lippe et al., 2009).

The time that marked the change of the system can be assumed to coincide with

strong promotion of hybrid seed technologies by the government in the 1990s. The

higher- yielding hybrid varieties have been widely adopted since, thus allowing to

reduce rural poverty through high maize incomes, which improved Black Thai

economy (Keil, 2008; Wezel et al., 2002)

Remote sensing and GIS (Geographic Information Systems) provided secure and

established foundations for measurement, mapping and analysis of natural

resources in the world. There are various ways to classify land cover based on

remote sensing and GIS such as supervised classification, unsupervised

classification or combination of supervised and unsupervised procedures with other

sources like economic, social and historical information as hybrid classification

methods which are well known and established today (Leisz et al., 2005; Goetzke

et al., 2008; Geissen et al., 2009; Berberoglu et al., 2009; Dewan and Yamaguchi,

2009; Dong et al., 2009). Especially for small catchments such as Chieng Khoi

commune, assessing land cover changes by combining satellite imagery and

historical data appears appropriate to better understand development in an area

where access to remote sensing information is strongly limited.

The term land cover describes the natural and anthropogenic features on the

Earth’s surface (such as forests, water bodies, agricultural areas, grassland etc).

For some time land cover was confused and interchangeable with land use, which

describes activities associated with humans, taking place on the land and

describing the current use of a property (production forest, secondary forest,

maize, cassava, etc). In this study, land cover is used in a broad sense, including

uses by humans.

3

1.1 Background

This Master Thesis was conducted within the framework of the Subproject C4.1

(phase 3) of the Special Research Program SFB 564 - The Uplands Program - in

Vietnam and connected to the framework of the subproject C4.2 (phase 4).

Continuing from phase 3, the subproject aims at better understanding of the

biophysical linkages and drivers of land use change in upland areas (C4.2 report

and proposal, personal contact). To this end, the landscape level model LUCIA

(land use change impact assessment tool) as a dynamic spatially explicit

biophysical model is applied (C4.2 report and proposal, personal contact). The

scope of the model is the simulation of material flows in the landscape. Required

input parameters include land use and soil maps, texture and organic matter

content. One method of model validation is inverse modeling like retrospective

simulation of soil fertility and the decline caused by land cover change from known

land cover history. This study aims at providing the historical basis of land cover

change within a time frame of 20 years.

1.2 Hypotheses Land cover in Chieng Khoi changed drastically during the last two decades. A

record of land cover history would be of much use to understand spatial patterns of

processes like soil degradation, erosion or deposition, which are influenced by land

cover. Due to difficulties in obtaining time series of high resolution imagery, a

hybrid classification approach was chosen to determine land cover change over

time. In this context the following hypotheses are formulated:

Based on the change of the policy of reforestation, forest cover increased from

1993 to 1999. It is expected that population pressure and markets for commercial

crops make the upland areas increase until 2007, thus reverting this tendency.

4

Farmer interviews on historical land cover will significantly improve land cover

classification from limited resolution satellite images with respect to accuracy.

Land cover change can be classified using a hybrid classification method and

maize can be distinguished from cassava (and from maize & cassava intercrop).

1.3 Objectives of the study Land cover change in Chieng Khoi commune in the past decades threatens

agricultural production as soil fertility declines and water erosion from upland to

lowland areas occurs. Models have proven to be an important tool to assess such

impacts but also to understand drivers of land use and land cover change (Verburg

et al., 2008). In that sense, they may provide policy makers with recommendations

to improve land use decision planning for sustainable agriculture in tropical and

subtropical upland areas. However, applying spatially explicit models requires a

consistent database of land cover and soil maps. A reverse modeling approach

could be particularly appropriate to reconstruct the development that led to the

current situation on site and identify the main drivers responsible for this

development. In addition to the abovementioned inputs, a reverse modeling

approach requires detailed spatially explicit information on land cover history. Due

to the limitation of high resolution satellite images and historical information in the

study area the objective of this study was to develop a hybrid classification scheme

and classify the change of land cover from selected points in time – 1993, 1999

and 2007 – by using satellite images and locally derived land use history

information with a special emphasis on maize crop cover expansion. The

generated results will provide part of the required input parameters to run the

LUCIA model.

5

Specific objectives of this study were to

• obtain satellite images and classify different land covers using remote

sensing and GIS software

• collect field data on soil properties and land cover

• collect information on current land cover and land cover history from

farmers and local authorities

• elaborate a decision-tree for land suitability based on soil properties and

participatory soil mapping and subsequently derive decision rules for

image classification based on farmers’ practise

• cross-check the image-based land cover classification on the basis of

the soil and interview data

2 Literature Review

2.1 Land use and land cover change in upland Northern Vietnam

Land use change, the physical change in land cover caused by human activities

such as agriculture and silviculture, is a common phenomenon associated with

population growth, market development, technical and institutional innovation, and

related rural development policy. Changes in land cover in turn can have various

consequences on economic growth, the level and distribution of income, and on

natural resources such as biodiversity, ecosystems, water, and soils (Müller and

Zeller, 2002). Degradation of natural resources is a global problem. In Vietnam, the

mountainous upland area is the ecological zone where resource degradation is

most serious. The uplands consist of hills, highlands and plateaus, occupying 24.4

million hectares (74%) of the country’s total area (Leisz et al., 2005). Agriculture is

more and more practiced on steep hillsides as the population of Vietnam’s northern

mountainous regions increases. Subsequent consequences are deceasing land

per person ratios. As an example, in Yen Chau district agricultural land per person

6

decreased from 0.5 ha per person in 1980 to 0.2 ha per person in 1998 (Wezel et

al., 2002). Farmers’ changes of farm practices and cropping types results in

change of land cover and vegetation type. A non-permanent vegetation cover will

lead to land degradation through increased soil erosion. In that sense, it will affect

upland farming activities by declining soil fertility, and further negatively influence

lowland areas by sedimentation, deterioration of water quality and direct effects on

the water balance.

Land cover changes are linked to the change of weather and climate in many

ways, as well as to human livelihoods and environmental components such as air

quality and water supply. They modify weather and climate through changed

albedo, transpiration rates and filtering of aerosols, thus directly influencing the

Earth’s radiation budget. Land cover change and biomass burning have impacts on

emissions of CO2, CH4, aerosols and dust. Land cover change itself can also

modify the surface energy and moisture budgets through changes in evaporation

and the fluxes of latent and sensible heat, directly affecting precipitation and

atmospheric circulation as well as temperature (Forster et al., 2007).

2.2 Remote sensing - GIS and land cover mapping Geographic Information Systems (GIS) and remote sensing (RS) are known to be

not only powerful, but also cost-effective tools for assessing the spatial distribution

and dynamics of land cover (Giridhar, 2008; Wilson and Fotheringham, 2008;

Zhiliang et al., 2008; Tottrup, 2002; Dewan and Yamahuchi, 2009). Lentes (2006)

also indicated the necessity of using RS and GIS to improve data bases on land

cover and slopes and to link spatial models to better understand, explain and

assess strategies of rural development.

There are many classification methods to distinguish and quantify land cover based

7

on satellite images. Generally, continuous spectral features in imagery are clumped

into discrete classes representing land cover types. Unsupervised methods are used

to produce RS-based maps of distinct zones and subsequently assign real land

cover classes to each zone. In contrast, supervised classification first calculates

spectral characteristics of spots with land covers known by maps and fieldwork on

the image. Subsequently, pixels on the image are assigned to the respective pre-

defined categories of land cover type (Wilkie and Finn, 1996). To classify images,

supervised methods always need homogeneous regions to be identified within the

image. It is not possible to combine present ground truthing points and old images

because of changes in spectral response from day to day and of ground patterns,

weather, sensors and many other factors from year to year.

RS can be a good tool for getting a more detailed impression on land cover change

(Zhou et al., 2008). The term remote sensing includes a wide range of applications

from digital scanning to optochemical photography and is widely used to produce

land cover information in GIS. Accuracy can reach up to more than 90% for land

cover classification (Tottrup, 2004). With the purpose of better understanding the

effect of land cover change on tropical forests by mapping, Tottrup and

Rasmussen (2004) introduced the basis and intuitive methods as one kind of

hybrid classification to classify temporal land cover change. By using ground

truthing data derived from historical interviews and multi-date LANDSAT imagery

which contains information on phenological properties such as canopy structure

and pre-classification image smoothing, it was expected to achieve higher

accuracy of classification. This procedure would allow using low resolution images

to determine vegetation cover at commune level.

Satellites images are secondary data with varying spatial resolution. In terms of

cost, high resolution images such as ALOS, SPOT or aerial photographs are more

expensive. Some sources with low to average resolution such as LANDSAT are

freely available. However, for some areas in the world, availability of satellite

8

images is limited due to missing coverage, restricted access to existing images or

high costs. Several researchers developed and tested basis methods with

supervised and unsupervised classification. Tottrup (2002) combined RS data with

social and historical data to distinguish different tropical forest types; Leisz et al.

(2005) combined RS with laboratory analysis to create an enhanced soil map

before applying a basis classification method. Thematic information improved

classification of mixed pixels in a study by Dewan and Yamaguchi (2009).

Hybrid classification methods are not restricted to combining supervised and

unsupervised methods, but can also be linked up with socioeconomic databases

(Fox et al., 2005). Together with several combination tools to understand the

interactions between human and natural systems like ABM (Agent Based spatial

computation Model), GIS, participatory methods can also be appropriate tools to

disentangle cause and effect relationships between factors and changes in land

use observations at different scales (Castella et al., 2005). At a landscape level in

small areas, hybrid classification can be a good option to classify and detect land

cover change and thus improve stand-alone supervised classification methods.

3 Materials and Methods

3.1 Study area This study was conducted in Yen Chau District, Son La province, Vietnam. The

research areas were located in the villages of Ban Me and Ban Tum, Chieng Khoi

Commune. The areas were characterized by tropical monsoon climates with very

hot, wet summers and dry, cool winters, with annual maximum temperature of

38°C and minimum temperature of 7°C, and annual rainfall of around

1,500 mm per year. The total population encompasses 2,717 people living in 567

households with a majority belonging to the Black Thai ethnic minority group. In

2005, the total communal natural area was 3,189 ha including agricultural areas of

1,111 ha (35%); these were used for paddy rice (79 ha – 2.5%), maize, cassava

and other crops. Forestry covered 1,741 ha (55%) mainly of protected forest and

some production forest. 75 ha (2.3%) were used for aquaculture. 162 ha (5.1%)

were non-agricultural and 101 ha (3.2%) unused land. The average household

farm size was 1.65 ha resulting in a land per person ratio of 0.35 ha (Dang, 2008;

Annual land use report 2005).

The study area experienced significant expansion of forest to agricultural production

areas in the last decades. Government policies implemented deforestation bans,

thus, teak (Tectona grandis) and bamboo plantations were planted widely by

extension programs and protected by commune committees. Nevertheless, the

increasing demand of food and cash due to the nationwide economic development

are increasingly pressuring the local natural resource entities.

Yen Chaudistrict

Chieng Khoicommune

Figure 1 Location of study site adopted and modified from Häring (2009)

9

10

Chieng Khoi has the possibility for low land agricultural production since

construction of the Chieng Khoi dam was started in 1968. The retained water fills a

40 ha large artificial lake during rainy season. It took 3 years to finish and 3 years

more for fully filling the lake with water. Paddy rice production and home water

consumption are supplied mostly from this lake. Within 5km from the main road

and Yen Chau town, farmers in Chieng Khoi access markets more easily

compared to surrounding communes. Construction of a new road in 2007 has

additionally improved market access for farmers in Chieng Khoi.

3.2 Field survey and interviews

Fieldwork and image processing were carried out from March to September 2009.

According to previous studies by Lippe (2009), the dominating vegetation classes

in Chieng Khoi in 2007 were: Forest (primary and secondary), shrubs and

successional fallow vegetation, teak, bamboo, fruit tree plantations and agricultural

fields (paddy and upland). Therefore, during the first field survey, for each major

land cover type three plots were selected as training dataset for the image

classification step. The training stages were needed to classify the different remote

sensing sources in a later stage of the study (Wilkie et al., 1996). Local land cover

history information was generated by participative group discussion with selected

commune stakeholders (Wezel et al., 2002).

a. Field data collection and individual interviews

In Ban Me and Ban Tum, 10 farmers (aged 40+) were selected, who were

members of the farmers’ social organization in the village.

During the field survey, local names for each area were recorded as reference for

mapping during the group discussion. Ground Truthing Points (GTPs) were

11

collected by using a handheld GPS (Global Positioning System) Garmin CSX60

(Garmin Asian Corporation, Taiwan). GTPs of all upland plots belonging to the 10

farmers were collected, land cover of the upland plots was described and

information on land cover history was gathered in semi-structured interviews.

Information on crop yields was collected at the same time.

Ground truthing data were collected during the field survey by stratified random

sampling, the training sets were gathered in a relatively homogeneous area

minimum consisting of three plots, subclasses were recorded for each class and

then combined to five classes (Wilkie and Finn, 1996). Extension of water bodies

and primary forests were homogeneous and remained stable, with exception to

lake size, which varied considerably between seasons. Primary forests were

located on rugged terrain and difficult to access, but, based on aerial photographs

in 1999, assumed to be relatively homogeneous. The training sets were created by

personal observation and from an aerial photograph. The quality of training areas

was evaluated and identified by histogram plots (Saha et al., 2005). Photos and

sketches were taken and drawn by the participants to delineate the location, land

cover and soil properties of their areas (Tottrup and Rasmussen, 2004). GTPs

were generated from interviews and discussions. Three transect walks were done

during the survey and GTPs taken (Appendix 3).

In total, 262 GTPs were collected during the survey, after combining with individual

interviews and group discussions:

• 150 points were generated and used for 2007 (out of which 50 points were

used to create the training set and 100 points to validate accuracy).

• 102 points were used for the year 1999 (23 points were used to create

training sets and 73 points to assess accuracy)

• 81 points were used for the year 1993 (25 points were used to create

training sets and 56 points to assess accuracy).

12

b. Group discussion

Land cover history was explained by the participants during the interview sessions

(Doppler et al., 2006). Another 15 experienced farmers who were also members of

the farmers’ social organisation were invited to take part in the discussions. The

discussion part was divided to 3 sessions.

Session 1 (day 1): 10 farmers who joined both individual interviews and field

survey and 5 additional farmers were invited, out of which 12 farmers attended,

divided into two groups. Each group was provided with one map which marked the

local names for their plots. The main task was to determine land cover per area, for

the complete questionnaire see Appendix 1. After discussion, farmers groups drew

land cover maps and listed land use history. Soil type classification into local

categories was also carried out during the discussion.

Session 2 (day 2): Ten farmers who had joined interviews and survey before plus

an additional 10 farmers were invited. 17 farmers came and were divided into two

groups. Questions were distributed concerning land cover changes, soil fertility

changes, erosion problems and soil conservation in context with land cover

change. Two groups worked on two subtopics:

(i) Group 1: The group of elder farmers (more than 50 years old) responded

to questions on land cover change;

(ii) Group 2: Younger (less than 50) farmers worked on soil fertility change.

Both groups also discussed about erosion problems and soil conservation methods

that had been used from the past to recent times.

After session 1 and session 2, farmers presented their results and discussed

between groups to get to the same level of information.

13

(a) (b)

(c) (d) Figure 2 Group discussion

(a) Session 1: Mapping and land cover history discussion (b) Session 1: Presenting results after discussion (c) Session 2: Separate group discussions on two subtopics (d) Session 3: Summary and conclusions

Session 3 (day 3): All farmers were invited to summarize their results after 2 days

of discussion. Additional information was collected during the last discussion about

institutional programs and projects conducted in Chieng Khoi, which was assumed

to be one of the major factors to change land cover (Fagerstroem, 2004).

14

To supplement field data and locally derived historical land cover information,

governmental secondary datasets and local reports were collected from communal,

district and provincial sources. The annual land use report 2005, and land use

planning to 2010 in Chieng Khoi commune were obtained from the Land

Management Department of the Chieng Khoi Committee, crop production reports

and a land use map for Chieng Khoi were collected from the Natural Resource

Management Department of the People’s Committee in Yen Chau. Agricultural

development reports were collected from the Agricultural Department, and the

provincial government gave information about the Chieng Khoi Lake.

c. Analysis

Information on land cover history was obtained for generated GTPs in the past.

Then GTPs were used for image processing and classification as described below.

Information from group discussions and secondary data of land cover history were

analysed qualitatively to get an overview of land cover change during the period

1993 to 2007, interpretive techniques and recursive abstraction analysis were

applied (Thai, 2009 and personal communication).

3.3 Image processing and classification LANDSAT TM (Thematic Mapper) images 1993 and LANDSAT ETM+ (Enhanced

Thematic Mapper) of 1999 from USGS (US Geological Survey) with spatial

resolution of 30 m were used for image land cover classification. Training sets of

primary forest and water bodies were created using an aerial photograph of 1999

from COSAMD (Centre of Mapping, Dept. of Survey and Mapping, Hanoi, Vietnam)

with a resolution of 5m. IRS LISS III 1C 2007 (Indian Remote sensing - Linear

15

Imaging and Self Scanning sensor) and a PAN (Panchromatic) image shaped with

LISS III were obtained from the National Remote Sensing Agency (NRSA), India.

(Table 1)

Table 1 Images used for land cover classification, technical properties and sources

No

Data

Path/Row

Sensor

Date of acquisition

Spatial resolution

Source

1 LANDSAT 128/45 TM 1/2/1993 30 x 30 m USGS

2 LANDSAT 128/45 ETM+ 27/12/1999 30 x 30 m USGS

3 IRS LISS III and PAN 123/57 Self

Scanning 4/3/2007 23.5 x 23.5 m 5.8 x 5.8 m NRSA

4 Aerial photograph 11/1999 5 m COSAMD

By combining the training stage information, a supervised maximum likelihood

classification (Wilkie and Finn, 1996) and a principal component analysis (Lillesand

and Kiefer, 2004) were carried out. Both classification methods differ in their

principal processing steps. Therefore, the outputs of both methods were compared

separately with the farmer-derived maps. The orthorectified aerial photographs and

satellite images were classified independently.

The image processing and classification work was done with ENVI 4.3

(Environment for Visualizing Images, ITTVIS, USA) and ArcGIS 9.3 (Environment

Systems Research Institute - ESRI, USA) software. This step was conducted in

close cooperation with the Center for Agricultural Research and Ecological Studies

16

(CARES), the local partner organisation of subproject C4. The CARES office at

Hanoi University of Agriculture employs experienced GIS staffs and a computer lab

with the required computer and software sources. The organisation has a long

experience in using GIS and remote sensing applications, and therefore was

closely related to all steps of this study. A soil map based on local criteria such as

soil colour and top soil thickness was used from subproject B4 (Clemens et al.

2008), and combined with results from the survey and farmers discussion; overlap

areas between Clemens et al. 2008 and own study areas were considered.

Table 2 Criteria used for discrimination of land cover classes used for classified satellite images

General cover class Criteria

Water Deep water level

Paddy rice Shallow water level, flat land

Upland Slope >5°, elevation >300m asl, annual or perennial crops

Maize Upland position and bare soil during dry season, with maize

Cassava Upland position, ground cover during dry season

Maize and cassava Upland, intercrop maize and cassava

Tree Mixed small tree, bush, fruit tree, bamboo, grazed secondary forest, planted forest as less dense vegetation

Forest/dense tree Dense primary forest, tree plantation, secondary forest, planted forest as dense vegetation

Pre- classification

The images were corrected by radian and geometric corrections and then

registered by standard projection WGS84 (World Geodetic System), datum UTM

17

(Universal Transverse Mercator), 48 North. GTPs from previous research were

also registered according to the same projection and coordinate system. The

image was enhanced for clear identification before classification: LISS III with

23.5x23.5m was combined with PAN data to improve solution to 5.8x5.8m (Saraf,

1999)

Classification

Supervised classification was carried out by maximum likelihood classification

using ENVI software; training sets were used to create samples. One of the

problems in mountain areas is shadow because of the sun angle. Shaded and non-

shaded slopes reflect waves in a different way. In Chieng Khoi commune, forest

was located in high elevation and rugged terrain, and thus affected by this problem.

Consequently, forest was classified as subclass forest 1 and forest 2 to avoid

shadow problems. Separability analysis used for this study (Saha et al., 2005) was

based on Transformed Divergence and Bhattacharrya Distance measures.

The Transformed Divergence is given by the following equations: TD(i,j) = 2*[1-exp(-D(i,j)/8)] where TD(i,j) = Transformed Divergence between classes i and j D(i,j) = divergence between classes i and j D(i,j) = 0.5*T [M(i)-M(j)]*[InvS(i)+InvS(j)]*[M(i)-M(j)] + 0.5*Trace[InvS(i)*S(j) +InvS(j)*S(i) -2*I ] where M(i) = mean vector of class i, where the vector has Nchannel elements (Nchannel is the number of channels used) S(i) = covariance matrix for class i, which has Nchannel by Nchannel elements InvS(i) = inverse of matrix S(i) Trace[] = trace of matrix (sum of diagonal elements) T[] = transpose of matrix I = identity matrix

18

The Bhattacharrya (or Jeffries-Mastusuta) Distance is calculated using the following formula: BD(i,j) = 2*[1-exp(-a(i,j))] where BD(i,j) = Bhattacharrya Distance between class i and j a(i,j) = 0.125*T[M(i)-M(j)]*Inv[A(i,j)]*[M(i)-M(j)] + 0.5 *ln{det(A(i,j))/SQRT[det(S(i))*det(S(j))]} where M(i) = mean vector of class i, where the vector has Nchannel elements (Nchannel is the number of channels used) S(i) = covariance matrix for class i, which has Nchannel by Nchannel elements Inv[] = inverse of matrix T[] = transpose of matrix A(i,j) = 0.5*[S(i)+S(j)] det() = determinant of a matrix ln{} = natural logarithm of scalar value SQRT[] = square root of scalar value

TD and BD values between 1.8 and 2.0 indicated complete separation between

two classes and such from 0 to 1.4 indicated complete overlap between two

classes. Values from 1.4 to 1.8 indicated the signatures generated from training

sets were not separable from other classes, so that resampling was required.

(Tottrup, 2004; Saha et al., 2005; Wilkie and Finn, 1996)

A Maximum Likelihood procedure was applied to determine the true shape of the

spatial distribution of each class; training samples were defined, refined, renamed,

deleted or resampled based on evaluation of the class histogram and statistical

parameters (Dewan and Yamaguchi, 2009; Wilkie and Finn, 1996). GTPs for

images 1993 and 1999 were generated from interviews and group discussions and

manual interpretation and classification of aerial photographs (1999). GTPs for

2007 were obtained combining own field surveys and data by Lippe (2009)

subproject C4.1, Upland Program (personal communication).

The modified local soil map and results from the first classification were used for

building a decision –tree to distinguish upland crops; maize, cassava and intercrop

maize & cassava. Results from interviews and cropping calendar obtained from the

Upland Program (Appendix 7- Author unknown, 2007) were used to determine

harvesting time which allowed crop cover in cultivated areas to be distinguished.

Post – classification

The classified maps were smoothened by Majority function (Tottrup and

Rasmussen, 2004; Tottrup, 2002; Saha, et al., 2005). Accuracy assessment was

presented in a confusion matrix table by Kappa coefficient. The confusion matrix

table was created by comparing error values for each class that was classified with

its respective value in the ground truthing data. The table has the same number of

columns and rows which equal the number of classes. The land cover classes in

the ground-truth image head the rows, while the same classes for the classified

image head the columns (Wilkie and Finn, 1996). The error matrix with the off-

diagonal elements represents the errors of omission (pixels that from ground truth

data are class i but were labelled as class j in the classified image) and

commission (pixels that were labelled as class i in the classified image but were

actually members of another class according to the ground truth data) (Wilkie and

Finn, 1996).

The Kappa coefficient is then calculated as follows:

∑

∑ ∑

=++

= =++

−

−= r

iii

r

i

r

iiiii

XXN

XXXNk

1

2

1 1ˆ

Where : r: Number of row in confusion matrix

Xii: Number of row i and column i

Xi+ and X+i: Total of row i and column i

N: Number of observations

19

20

After image processing, the farmer -derived land cover maps were used to verify

the processed satellite images. Kappa statistics (Tottrup and Rasmussen, 2004)

were used to assess overall image accuracy.

Geom etric Correction

Registry projection Enhance

-m ent

Satellite Landsat

TM ETM +

Land cover m apping

Data conversion / Extraction of

inform ation for land cover

types

Detection of land cover

changeAccuracy assessm ent

Sam ples for Land cover classes Supervised classif ication

Pre processing

Post processing

Processing

Satellite im age Landsat T M/ET M+

LISS III+PAN

Boundary studyarea data

GTPs and discussion inform ation

Topographic data GTPs NDVI

In terview

Figure 3 Process of classification (Adopted and modified from CARES unpublished report in Vietnam Upland Forum 2009)

21

22

After post classification, a modification to the supervised method was used to

classify different types of forest (Tottrup et al., 2007). Upland areas in 2007 were

converted to vector files and only upland areas were masked and extracted from the

LISS III images as regions of interest. A decision-tree was built and maximum

overlay with the local soil map by ENVI 4.3 package to obtain. Firstly, a supervised

Maximum Likelihood procedure was applied again with 3 classes – maize, cassava

and maize & cassava. Training sets were created based on the repeated evaluation

of class histograms and statistical parameters. Secondly, overlay with the local soil

map in ENVI 4.3 and ArcGIS 9.3 and combination with the decision-tree were

implemented to arrive at an enhanced classification that took land suitability into

account.

Figure 4 Local soil map, adopted and modified from Clemens et al. (2008)

23

24

3.4 Detection of land cover change

The independently classified and orthorectified aerial photographs and satellite

images were used to generate the respective land cover maps of the selected years

1993, 1999 and 2007. The land cover change for the respective time period was

finally analysed by applying different methods described by Leisz et al. (2005) and

Dewan et al (2009): Cross tabulation analysis on pixel - by – pixel basis allowed to

determine the quantity of conversions from a particular land cover class to others in

the categories. The areas of each class were derived from initial year to final year of

the period. In this study, initial year was 1993 and final year was 1999. In this study,

we didn’t examine the period from 1999 to 2007 because of different spatial

resolution of the respective images (LANDSAT with 30x30m and LISS III with

5.8x5.8m).

25

4 Results

4.1 Land cover change as witnessed by local farmers Information given in this section stems from individual interviews with older farmers

conducted during the workshops in 2009 with farmers of Ban Tum and Ban Me

villages.

4.1.1 History of land cover change

Major changes in land cover began after 1954, the year of the victory over the

French. Before 1954, almost the entire area of Yen Chau was forest including

primary forest and dense secondary forest. Local varieties of upland rice and local

maize were the main crops in this area grown under shifting cultivation. Only few

houses existed and fruit was harvested mainly from wild fruit trees or home

gardens, not plantations. Few rain fed paddy rice plots existed, but paddy rice

cultivation was not common. After 1954, the first settlement program started after

the war, and the view of landscape totally changed. In order to fight poverty and

hunger after the war, the government encouraged farmers to enlarge agricultural

areas. From 1960 onwards, agricultural area was significantly expanded and at the

same time natural forest area decreased rapidly.

From 1954 to 1990, upland rice areas became abundant and took the main position

in food consumption. Fruit trees were increasingly planted from 1980 onwards, after

several housing and settlement programs had been initiated. Chieng Khoi dam was

constructed in 1971 and the artificial lake was fully established in 1974.

Subsequently, paddy rice areas increased that were irrigated by water from the lake

(Hager, 2006).

From 1990 on, local bi- or triennial cassava varieties became more common and

26

cultivation areas increased. Farmers described the following typical crop rotation

after forest clearing:

Forest – upland rice – upland rice – upland rice – cassava – cassava – upland rice

– cassava – cassava – cassava – upland rice – longer-lasting cassava or fallow.

Table 3 Soil types and history of land cover based on farmers’ discussion Local site

Soil type 1954 1980 1990 1993 1996 2002 2007 2009

Co

Hin

h &

K

hau

The

Yellow soil with stones

Forest regrowth

Upland riceFruit trees

Upland rice and fruit trees

Cassava Cassava+ maize+ fruit trees

Khu

m N

am

Clayey black soil

Upland rice Upland rice Upland rice Cassava

Upland rice Cassava Teak

Maize and cassava

Me

Nha

ng

Ta H

in

Sandy black and yellow soil

Upland rice (with fallow) Upland rice Upland

rice Upland rice and cassava (slopes) White mulberry (flat land) Maize and cassava

Sur

roun

ding

s La

ke

Khu

m P

hung

Sandy black or yellow soil

Upland rice (with fallow) Upland rice Upland

rice

Upland rice+cassava+ teak (slopes) White mulberry until 1998

Teak, pine, timber trees

Cassava 2003-5 Maize

Upland rice almost disappeared from Chieng Khoi after 1990 because of land

degradation; the soil was not fertile enough for upland rice anymore. During 1992 –

1994, white mulberry (Morus alba) was introduced and covered almost all fields in

low elevations and flat upland areas.

27

Maize and commercial 1-year cassava were introduced and swiftly became popular

from 1996 on. Further, high yielding hybrid maize varieties spread as market

demand for maize was high and prices were good. Consequently, both were

planted widely, initially intercropped with fruit and teak trees, but expanding also into

production forest. Gradually, perennial trees in production forest and fruit tree areas

lost importance and were replaced by monoculture of maize.

4.1.2 Changes in local soil type and soil fertility

Black Thai farmers recognize changes in soil fertility indicated by changes of soil

color and texture. According to farmer interviews conducted in Ban Tum and Ban

Me, topsoil color was predominantly black, indicating high fertility, as long as the

major part of land area was covered by forest and only few upland fields with

shifting cultivation existed. On slopes, the cultivated soils eroded by almost 3 cm

per year at that time. Soil loss, however, was compensated by a top soil gain of 1-2

cm during the first year of fallow. With only small cultivated areas in long forest

slopes and long fallow periods, erosion was not yet a problem. Hoe, plough and

fertilizer were not used. From the 1980s, hoe and plough were used to prepare the

soil for cultivation on sloping land. At this time farmers began to recognize that the

color of the soil in most areas turned more brownish. Browning strongly occurred in

‘non-sticky’ soils (of low clay contents). Soils regenerated to a minor degree as

compared to previous times with rates around 1cm per year. As fallow periods were

shortened, erosion problems became more visible. From 1992 on, cultivation areas

expanded, most accessible areas became fields or were intercropped with upland

crops. On sloping lands, farmers observed increasing patches of red-yellowish and

stony soil outcrops. Only small newly cleared forest areas with ‘sticky’ (i.e. clayey)

soil showed black or brown colors. Black Thai farmers were conscious of soil

erosion problems. On their fields, around 5-6 cm of soil were eroded per year while

topsoil regeneration was minimal. Sandy and stony soils were quite common after

28

1994. Chemical fertilizer was widely used from 2000 on. Farmers believed that

chemical fertilizer caused soil crusting, and as a consequence hoeing and deep

ploughing on steep slopes became more common.

Table 4 History of soil erosion problems and change of soil types 1954 1980 1992 2009

Erosion

Low erosion rates Erosion about 3cm/year Topsoil formation about 1-2 cm/year

Increased erosion rates Topsoil formation 1cm/year

Very high erosion rates, Sandy texture stone outcrops Soil eroded 5-6cm/year Almost no soil compensated

Soil color Back and fertile soil Brown Change of soil color: Yellow and reddish

Farm type Individual (until 1959), group (until 1961) co-operative (until 1986) Contract 10 Contract 10

Crop management

Burning, raking (until 1979) No fertilizer

Hoeing, raking No fertilizer

(until 1994) intensive cultivation hoe, raking, no fertilizer until 2000 hoe, plough until 2004 hoe plough and fertilizer

Soil conservation measures were applied Wood trees as barrier to stop water flow Fallow, or planting 2 - 3 years cassava Grass and pineapple strips on contour lines

Reasons of change in crops Soil color change - less fertile Market orientation Diseases and drought

4.2 Land cover change analysis based on classification of satellite images

In this section results of image analysis as obtained from the hybrid procedure of

supervised classification and stepwise quality enhancement using complementary

information are presented. First, results of the separability analysis are presented,

which were acquired using the supervised classification based on training datasets

29

before processing maximum livelihood classifiers. This method did not allow to

distinguish different upland crops, which was considered crucial for later evaluation

of land use change. Thus, in a second step, results of the initial supervised

classification were combined with information derived from participatory discussion.

Finally land cover change after processing as described in the previous sections

was assessed and accuracy of the procedure was validated.

4.2.1 Separability analysis Signatures for each land cover type were created and optimized until the spectra

of signatures for one region of interest did not vary, i.e. until good agreement with

the training sets was achieved. Then, ENVI 4.3 was used to carry out the

separability analysis of classes after evaluating statistical parameters and

histograms. In order to allow for better detection, forest areas were subdivided into

two classes: Forest1 referred to dense forests on steep slopes, shaded as a

consequence of topography or low sun angle, while dense forests without shadow

were labeled as Forest2. In general, five classes could be clearly distinguished

from each other. Both Transformed Divergence (TD) and Bhattacharrya Distance

(BD) measures were used to quantify separability between land covers. TD and

BD values differed only minimally, so only the lower of both values is presented

here. TD/BD of 1.8 to 2 indicated full separability of two land cover classes, while

lower values showed that distinction was not possible on the basis of the given

image. Values of TD and BD of upland crops (maize, cassava and maize &

cassava intercrop) ranged from 0.32 to 1.28. This meant the upland crops were

combined in one class – labeled as Upland. Results of the separability analysis for

the three years under investigation are shown in Tables 5, 6 and 7.

Table 5 Separability of land cover classes 1993 (minimum of TD and BD)

Tree Paddy Water Forest1 Forest2

Upland 1.710 1.999 2.000 2.000 2.000

Tree 1.989 1.999 1.999 1.999

Paddy 1.999 2.000 2.000

Water 2.000 2.000

Forest1 1.999 Table 6 Separability of land cover classes 1999 (minimum of TD and BD)

Tree Paddy Water Forest1 Forest2

Upland 1.913 1.999 1.999 1.999 1.999

Tree 1.999 1.999 1.999 1.881

Paddy 1.999 2.000 1.999

Water 2.000 2.000

Forest1 1.999 Table 7 Separability of land cover class 2007 (minimum of TD and BD)

Tree Paddy Water Forest1 Forest2Upland 1.995 1.999 1.999 2.000 1.999

Tree 1.998 1.999 1.892 1.826

Paddy 1.999 1.999 1.999

Water 1.999 1.999

Forest1 1.886

30

31

Results of separability showed that forest, water and paddy classes got high TD/BD

values, even more than 1.9 in all three years in cases of water and paddy. In 1993,

the results showed the complete separation between four classes in pair, except for

tree and upland classes (TD/TB=1.701). Results in 1999 and 2007 indicated the

same; five classes were separated and completely distinguishable from each other;

TD/BD values ranged from 1.826 to 2.000.

4.2.2 Combination of remote sensing and interview data

In order to come to a more detailed classification including discrimination of the

most important annual upland crops maize and cassava, remote sensing data were

complemented with information obtained from farmers. Combining results from the

first classification with five land cover classes, a decision tree was built based on

information from group discussions, interviews and the local soil map. The results

showed the relationship between soil quality, cropping calendar, soil characteristics

and crop cover. According to the decision-tree developed with the farmers, good

soil could be used for all crops while poor soil was considered suitable only for

cassava. In case of soils with intermediate fertility level, maize could only be

cultivated with fertiliser inputs, which were usually not affordable for farmers. Early

harvesting maize varieties led to longer periods of bare soil exposed before being

covered by cassava or intercrop. Both cassava and intercrop could be distinguished

in the LISS III image taken in March 2007. Stone contents of the soil helped farmers

to decide whether to plant cassava or maize in pure stand or intercropped.

Observed changes in soil color gave further indications in deciding which crop to be

planted (See Figure 5).

LANDSAT, LISS III

Forest TreeWater Upland Paddy rice

Supervised classification

Cassava MaizeMaize &Cassava Cassava 2yr

Poor soil

Cassava orMaize & Cassava

Soil with stones Soil w/out stones

Maize or cassava 2yror Maize & Cassava

Bare soil indry season

Cassava 2yr orMaize & Cassava

Soil cover indry season

Black soil Brownish soil

Intermediate soil

Fertile soil

Figure 5 Decision – tree for land use classification based on farmers’ criteria for crop rotation

32

33

4.3 Land cover map, statistics and land cover change a. Land cover map Land cover maps indicated that the upland and forest classes strongly changed

from 1993 to 2007. Forest areas firstly moved close to the lake from 1993 to 1999

indicated by an increase of dense tree cover and extending forest from forest

boundaries into other land cover types. But from 1999 onwards, forest margins were

farther from the lake again, even when compared to 1993.

Upland areas (red color in maps) clearly increased from 1993 to 2007. In 1993,

upland crops were concentrated in few areas, spread out more in 1999 and

expanded significantly in 2007, when upland systems were the dominant land cover

in the lower part of Chieng Khoi. In the southwestern forested areas, in 1993, the

non-forest tree class was detected in small areas while in 1999 upland class

appeared. In the land cover map of 2007, not only larger areas of upland and tree

classes were located in the southwestern primary forest, tree class was also found

in the center of the primary forest and upland class appeared in the southern part of

the primary forest.

In context with the DEM (Digital Elevation Model) shown before, land cover maps

indicated that forest areas firstly increased in the lower part (from 1993 to 1999) and

then withdrew to the upper parts of the catchment (from 1999 to 2007). At the same

time upland areas continuously expanded from 1993 to 2007 from the lower to the

upper parts, especially in the primary forest areas.

Fi

gure

6 L

and

cove

r map

s 19

93 (a

) and

199

9 (b

) bas

ed o

n LA

ND

SA

T an

d cl

assi

ficat

ion

by m

axim

um li

kelih

ood

proc

edur

e in

EN

VI 4

.3 &

Arc

GIS

9.3

34

Figure 7 Land cover map 2007 based on LISS and classification by maximum likelihood procedure in ENVI 4.3 & ArcGIS 9.3

Figure 6 and Figure 7 present land cover maps with 5 land cover classes

distinguished and optimized by a majority function. The unprocessed maps,

classified without majority function of maximum likelihood, are presented in

Appendix 8.

35

36

b. Statistics of land cover and land cover change

Areawise, water bodies occupied the smallest, while forest and other trees covered

the largest portions of the catchment. Forest area was larger than tree area in 1993

and 2007, but in 1999 it was lower. Upland fields occupied intermediate shares in all

three observation years, with a strongly increasing tendency (Table 8). Table 8 Land cover statistics

Land cover area (ha) 1993 1999 2007

Upland 308 423 708Water 18 26 17Tree 1,572 1,283 1,361Paddy 83 92 110Forest 1,146 1,303 929

Water body area showed some oscillation between years due to lake water level.

This was the case because the LANDSAT image 1993 was taken in February – late

dry season – while the 1999 image was taken in December, during early dry

season. In contrast, 2007 was a drought year with very limited rainfall. Paddy rice

areas steadily increased at an overall rate of +33%, but on a small scale, from 1993

to 2007. Upland areas increased by 100 ha in 6 years (37% from 1993 to 1999) but

almost doubled in the following 8 years (from 1999 to 2007). Forest areas and tree

areas were converted into each other: From 1993 to 1999 forest areas increased

while tree areas decreased. From 1999 to 2007, forest areas decreased, while tree

areas increased.

Time (a)1993 1999 2007

Are

a (h

a)

0

200

400

600

800

1000

1200

1400

1600

1800Upland Forest Tree Paddy rice Water

Figure 8 Land cover change from 1993 – 2007 in Chieng Khoi

Due to the differences of resolution between LANDSAT and LISS III, the exact

change detection could be carried out only for the period of 1993 to 1999, which

were available as LANDSAT images from identical sources and with identical

resolution. The change detection table indicated that tree areas were not only

converted to forest areas. 217.4 ha (out of a total area of 1,572 ha) were converted

into upland areas while 177.2 ha (out of 1,572 ha) were converted into forest areas

(Table 9).

37

Table 9 Detection of land cover change 1993 - 1999, bold figures refer to class areas without change

Area (ha) 1993

Upland Tree Paddy Forest Water Sum 1999

Upland 189.7 217.4 9.3 6.7 0.2 423.3

Tree 116.6 1,148.5 4.1 13.6 0.3 1,283.1

Paddy 0.7 21.4 69.9 0.0 0.0 92

Forest 0.2 177.2 0.0 1,125.9 0.0 1,303.3

1999

Water 0.7 7.5 0.0 0.0 17.5 25.7

Sum 1993 307.9 1,572 83.3 1,146.2 18.0

Class Changes1 118.2 423.3 13.3 20.2 0.6

Image Difference2 115.4 -288.9 8.7 157.1 7.7 1 Area in one class converted to other classes 2 Area 1999 subtracted from area 1993

38

39

d. Distribution of upland crops In case of 1993 and 1999, LANDSAT with resolution 30 by 30m had been used to

classify land uses. Upland plots were small, ranged from 200 to 1500 m2 , and plots

of more than 1000 m2 size used to be divided into several subplots for different

purposes such as cultivated cassava intercrop with upland rice or maize, one year

cassava, 2 years cassava for home consumption or planted fruit trees. So the

spectra of those crops were mixed in one pixel in LANDSAT images. The statistic

parameters analysis of separability for upland crops gave TD/BD value ranging

from 0.32 – 1.03 in both years. Additionally, no soil map of Chieng Khoi in 1993

and 1999 existed and few GTPs were found during the interviews. This lack of

information made it impossible to distinguish the different annual upland crops in

1993 and 1999.

During the time of observation, maize and cassava were the most common annual

crops in Chieng Khoi. Both were cultivated either as mono crops or as maize-

cassava intercrop. As results of this study, monocropped maize and cassava areas

were similar in size, namely 283 ha cassava and 282 ha maize. Table 10 shows

that intercropped maize and cassava area was less, 163 ha (23 % of total upland

cultivated areas). Figure 9 shows only upland areas extracted in the entire

commune of Chieng Khoi.

Figure 9 Distribution of upland crops in Chieng Khoi commune, March 2007, based on LISS

imagery, local soil map and decision tree procedure carried out in ENVI 4.3 & ArcGIS 9.3 Table 10 Areas of upland crops in Chieng Khoi, March 2007 classified LISS III image Upland crops Area (ha) %Cassava 262.8 37.1Maize 281.9 39.8Maize and cassava 163.3 23.1Total 708.0 100.0

40

4.4 Classification accuracy

Overall accuracy and Kappa Coefficient in all cases were acceptable, overall

accuracy values were presented as a percentage of test-pixels successfully

assigned to the correct class in the classified images. The highest overall accuracy

value was reached for 1999 (98.5%) and the lowest one for 1993 (81.1%). The

highest Kappa coefficient was obtained in 1999 (0.977) with the highest agreement

in the error matrix, and lowest one in 1993 (0.684) (Table 11). The error matrix

which presented omission and commission values in 1993, 1999 and 2007 is

shown in Appendix 6. The overall accuracy assessment achieved was 66.7% when

classifying different upland crops. That means to distinguish upland crops, the test

pixels assigned 66.7% correctly to classified images.

Table 11 Overall accuracies and Kappa coefficient of land cover classification produced from basic supervised classification with five classes.

Year 1993 1999 2007

Overall accuracy 81.1% 98.5% 82.5%

Kappa Coefficient 0.684 0.977 0.740

5 Discussion

5.1 Validity of data

5.1.1 Expansion of upland agriculture

In the Northern Mountain Region of Vietnam, expansion of cash crops grown in

monoculture replacing swidden agricultural system has been reported by several

41

42

authors (Castella et al., 2005; Fox and Vogler, 2005; Wezel et al., 2002; Nguyen et

al., 2004). As one result of this study, exact data on the area informally converted

to upland area in Chieng Khoi commune could be obtained. This is in line with

findings by Hager (2006), who stated that the land allocation process after

decollectivisation in 1993 led to a gradual reduction of common land used for

grazing. These lands were cultivated for cash crops which generated more income

than livestock keeping. In Chieng Khoi, 119 ha of other lands were converted into

cultivated land in slope areas in the uplands from 1993 to 1999 - at the same time

when farmers indicated the decline of soil fertility.

Land use allocation in 2007 as obtained from our classification came close to areas

published by the Yen Chau People’s Committee in their annual report of 2005.

Field observations showed that upland crops are also intercropped with fruit trees

in plantations or planted under production forest as agroforestry systems in sloping

land. Thus, it can be stated that expansion of upland crops tends to be

underestimated in both surveys. Population pressure, food demand and a

transition from subsistence and planned economy towards market orientation all

contributed to increase areas under cultivation. From 2000 to 2005, Yen Chau

committee reported an increase in cultivated areas of 121 ha, converted from

unused land and production forest in Chieng Khoi commune.

5.1.2 Forest cover

After the Socialist Reform and Economic decline during the American war and after

1975 with several famines, improving food security was necessary. Additionally,

after reunification of the country, population increased and extension programs

were launched to expand agricultural areas for food security. One part of the forest

areas was converted to agricultural land. During collectivization, forest lands

43

belonged to state–owned enterprises. Timber exploitation and lack of regulation to

protect and sustainably use forest resources led to reduction of natural forest cover

(Hager, 2006; Meyfroidt and Lambin, 2008; Clement and Amezage, 2008). After

decollectivization in the 1980s, forest land increased again. In addition,

government policies to protect forests were implemented between 1991 and 1999.

Hager (2006) listed relevant laws and regulations affecting forest cover:

- 1991: Law on protection and development of forest enacted.

- 1993: Decree 327 – “Regreening of the barren hills”, a program with

purpose to cover barren hills with planted forest. In the same year, a law of

allocation of forest land to individual households or user groups on local

level was implemented, which gave farmers the interest and responsibility of

protection and use of forest sources.

- 1998: “5 million ha program” with targets to restore forests on barren hills

and to protect existing forests.

Outcomes of these regulations were reflected in the results of the land cover

classification conducted within this study, which showed that forest cover in 1999

was higher than in 1993.

Upland Forest

Are

a (h

a)

0

200

400

600

800

1000

1200

1400

1600

1800

Committe report 2005 Cassification2007

Figure 10 Comparison of the annual land use report 2005

by the Chieng Khoi People’s Committee and the classified satellite image 2007

We assumed that land use had not changed from 2005 to 2007. Comparing the

land use map published by the Yen Chau committee with the classified land cover

map, some protected forest areas were determined as tree areas. In the annual

land use statistics report 2005 by the Chieng Khoi committee, protected forest

defined as dense forests occupied 1,692 ha in total land, while results from our

research showed only 912 ha (Figure 10). In the annual land use report, the

definition of forest was used from a legal perspective in the sense of protected area

of forest while in our study it was used from a biomass perspective as dense forest

or dense tree plantations with the signatures created from primary forest in Chieng

Khoi. Part of the lower classified results can be explained by illegal extraction of

44

45

wood in protected forest that reduced the density of trees with lower canopy. So

the spectral reflection was determined as tree class with lower tree density as fruit

tree plantation or grazing and secondary forest, which according to Tottrup (2004)

were defined as non-forest.

5.2 Limitations and potentials to use satellite imagery combined with data deriving from local farmers

5.2.1 Limitations of classification satellite images

Remote sensing data are widely used for land cover mapping. However, a

classification approach built exclusively on the basis of spectral data from a remote

sensing sensor alone may not be sufficient to gather effective land cover

information (Saha et al., 2005). In a case study in the Himalayas, Saha et al.

(2005) introduced a multisource classification approach by combining remote

sensing data (IRS LISS III image) with NDVI (Normalized Difference Vegetation

Index) and DEM data layers using a maximum likelihood classifier. In our study,

primary forest was restricted to high altitudes and rugged terrain, whereas other

land cover types could be distributed anywhere in the catchment. Additional data

were necessary to complement information to classify old satellite images. The aim

of this study was to classify land cover change based on satellite imagery

complemented by information gathered through participatory methods from

farmers’ historical knowledge as hybrid classification. Generally problems with

remote sensing data may arise from shadows, sun angle, clouds, haze, smoke,

dust and multi solution sources and sensors. Additionally, in 2007 GTPs collected

by two different observers were used, which may have reduced accuracy.

However, distinguishing crop types is still a challenging issue and accuracy

achieved reached only 66.7%. During this study, we tried to include an NDVI layer

46

to separate maize, cassava and intercropped maize and cassava. Cross checking

with GTPs of cassava and maize, there was a wide range in NDVI values of

cassava crop and overlap with maize and maize intercrop with cassava classes.

Comparison with the local soil map by subproject B4 showed that some maize

areas were found in yellow soil. Those were the cultivation areas where our survey

results and the local soil map did not agree. Within this study it was not possible to

determine whether these deviations were owed to errors of the local soil map or to

the overlaying process required for classification. However, disagreement between

soil layers explains the relatively low accuracy

5.2.2 Limitations of using interview data

The combination of local and scientific knowledge was used in a study by

Fagerstroem (2004). Participatory Landscape Analysis (PaLA) survey proved

useful both for spatial and temporal aspects where no data on land cover change

in the past were available. Combination of remote sensing data with participatory

data allowed to include more observation points for classification. Making use of

farmers’ knowledge to assess plausibility and accuracy of the image-based

classification depended on the understanding of the study site and special skills to

transfer social data to geographical data might have further improved our results.

Using interview data, there are some challenges to tackle. In the short run, farmers

are usually able to remember a story correctly, but with longer time, memorized

data may become more and more vague and incorrect. Additionally, information on

past times can only be retrieved from experiments or older farmers. Individual

interviews showed that information derived from different persons could be

contradictory. Thus, triangulation using experimental or survey data to cross check

between individual interviews and the participatory approach was necessary.

Further, less data, compared to recent surveys, existed regarding past land cover.

47

In 2007, 150 GTPs were used while in 1993, 81 GTPs were used. This may have

led to limitations of results (lower accuracy in 1993) while using the same method.

5.2.3 Potentials of combining satellite images and farmers’ knowledge

Unsupervised classification is undertaken most often as a guide to collecting

ground data which only generates a set of spectral clusters as classes. The

supervised method converts the spectral data contained within remote sensing

directly into thematic land cover information, which is why unsupervised

classification is normally carried out as a guideline to select a training set and for

ground data collection before a supervised classification is started. However,

supervised classification always requires homogeneous patches of a known

landscape type to statistically generate spectral signatures characteristic of each

landscape (Wilkie and Fine, 1996). So the combination with farmers’ knowledge

may turn out to be important to create spectral signatures for old imagery.

However, accuracy needs to be assessed critically and additional cross-checking

as well as in-depth interviews on site are necessary for further research. In

addition, field work combined with laboratory work may also complement datasets

based on farmers’ knowledge.

NDVI datasets have the potential to facilitate the separation of different crops.

However, in this study reflection of chlorophyll due to certain weather conditions at

the time when the images were taken may have given confusing results. Further

studies removing external effects may help to solve problems in combining NDVI

datasets with farmers’ knowledge and may allow distinguishing upland crops on a

satellite image.

48

5.3 Hypotheses posed

Based on the results shown and discussion above, the hypotheses presented at

beginning of this study could be answered as follows:

As stated in hypotheses 3, the hybrid classification method which combined

supervised classification and farmer knowledge allowed to distinguish upland

crops. This would not have been possible without the hybrid approach. Accuracy

improved from 50% to 66.7% after manual correction of wrongly classified pixels,

which was carried out based on interview data. Thus integration of supplementary

information improved accuracy as assumed in hypothesis 2. However, this

approach worked out in case of 2007 when a local soil map and sufficient GTPs of

upland crops were available, but accuracy was low (66.7%). Further studies are

necessary to link results to laboratory results and to supplement the current

datasets with more field data. In 1999 and 1993, GTPs of upland crops were

limited, the farmers weren’t sure which crops they planted in their plots 10 years

ago.

Farmers’ information from interviews and discussion complemented the land cover

history information. So, the land cover changes were derived from classified land

cover maps. From these maps, they also indicated the consequences of the

conflict between reforestation policy, population pressure and commercial crops.

The forest cover was not increased continuously after policy established as stated

earlier in hypothesis 1.

6 Conclusions

Individual interviews and participatory methods making use of farmers’ knowledge

were used to build a decision-tree, which was then applied for hybrid classification.

Combining RS data, GIS gave an overview of land cover change in Chieng Khoi

49

commune especially of forest and cultivated upland areas. This method also firstly

gave potential to distinguish upland crops.

Changes of forest cover asserted the conflict between management of natural

resources and population pressure, food security and economic growth.

Sustainable development is an important challenge for local committees in

Northern Vietnam. Policies were a key factor leading to increase in forest cover

found on the classified satellite images from 1993 to 1999. But after that, the forest

covered areas decreased while cultivated upland areas increased within the period

from 1999 to 2007.

From the results of this study, hybrid classification methods can be used to

determine land cover changes in small catchments like Chieng Khoi commune,

complementing data from farmers and cooperative projects (like the Uplands

Program). In the sense of local benefit, the Yen Chau committee should consider

about decreasing of forest cover in relation with definition of primary forest, the

area of forest cover may not decrease, but the density of tree and canopy of

natural primary forest may decrease.

Time frame and methods applied in this study did not allow for separation of upland

crops from LANDSAT imagery, so land cover change could not be detected from

1993 to 2007. In the first place, this was due to the low resolution of the LANDSAT

images (30 by 30m) and different technical properties of LANDSAT and LISS,

which made a direct comparison impossible. Secondly, additional processing

steps, data layers and tools in the ENVI and ArcGIS packages might have been

needed.

The impacts of land cover change on soil degradation, erosion or deposition were

not investigated. These are being investigated by Ayanu (2009, unpublished) for

50

Ban Tat catchment in Hoa Binh province, and will be part of future research within

the Uplands Program. Moreover, land cover map and historical information of

some areas in the map also will provide the basis for implementation of the LUCIA

model in Chieng Khoi watershed. Results generated from LUCIA can be applied for

land use planning authorities in Yen Chau people committee, and may thus

improve the management of natural resources for sustainable development in

Mountain Regional in Vietnam.

51

7 References

Ayanu, Y. Z. (2009). Simulating consequences of landuse change and spatial distribution of landuses on sustainability of crop production in Northern Vietnam. Master thesis. Institut of Plant Production and Agroecology in the Tropics and Subtropics, University of Hohenheim. In preparation.

Berberoglu, S., & A. Akin (2009). Assessing different remote sensing techniques to detect land use/cover changes in the Eastern Mediterranean. International Journal of Applied Earth Observation and Geoinformation, 11(1), 46-53.

Castella, J. N.R. Tronche & N. Vu (2005). Impact of land use changes on mountain landscapes in Northern Vietnam during the 1990s. [Impact des dynamiques agraires sur les paysages de montagne au nord du Vietnam au cours de la décennie 1990] CyberGeo, 2005.

Castella, J., T.N. Trung & S. Boissau (2005). Participatory simulation of land-use changes in the Northern Mountains of Vietnam: The combined use of an agent-based model, a role-playing game, and a geographic information system. Ecology and Society, 10(1).

Clement, F., & J.M. Amezaga (2008). Linking reforestation policies with land use change in northern Vietnam: Why local factors matter. Geoforum, 39(1), 265-277.

Clemens, G., S. Fiedler, N.D. Cong, N.V. Dung, U. Schuler, & K. Stahr (2008). Soil fertility affected by land use history, relief position and parent material under tropical climate in NW Vietnam. Submitted.

Dang, V.Q. (2008). Agricultural statistics of two subcatchments in Yen Chau district, Son La province, Vietnam 2007. Unpublished manuscript.

Dewan, A. M., & Y. Yamaguchi (2009). Land use and land cover change in greater Dhaka, Bangladesh: Using remote sensing to promote sustainable urbanization. Applied Geography, 29(3):390-401.

Dong, L., W. Wang, M. Ma, J. Kong & F. Veroustraete (2009). The change of land cover and land use and its impact factors in upriver key regions of the Yellow River. International Journal of Remote Sensing, 30(5):1251-1265.

Doppler, W., S. Praneetwatakul, N. Mung Kung A. Satttarasart, J. Kitchaicharoen & C. Thongthap (2006). Resources and livelihood in mountain areas of South East Asia: Farming and rural systems in a changing environment. Margraf, Weikersheim, Germany.

Fagerstroem, H. M, D.T. Tran, H. Sodarak, M. v. Noordwijk and L. Joshi (2004). How to combine scientific and local knowledge to develop sustainable land use practices in the Upland - A case study from Vietnam and Laos. Proceedings of the workshop: Poverty reduction and shifting cultivation stabilization in the uplands of

52

Lao PDR: Technologies, approaches and methods for improving upland livelihoods. Luang Prabang, Lao, 27-30/1/2004 www.nafri.org.la/document/uplandproceedings/TOC.PDF

Forster, P., V. Ramaswamy, P. Artaxo, T. Berntsen, R. Betts, D.W. Fahey & J. Haywood (2007). Changes in atmospheric constituents and in radiative forcing. In S. Solomon (ed.): Climate change 2007: The physical science basis (pp. 180-186) Cambridge University Press, Cambridge, United Kingdom.

Fox, J., & J.B. Vogler (2005). Land-use and land-cover change in montane mainland Southeast Asia. Environmental Management, 36(3):394-403.

Giridhar, M. V. S. S., & G.K. Viswanadh (2008). Evaluation of watershed parameters using RS and GIS. , Earth and Space Conference 2008: Proceedings of the 11th Aerospace Division International Conference on Engineering, Science, Construction, and Operations in Challenging Environments, 323.

Goetzke, R., M. Braun, H. Thamm and G. Menz (2008). Monitoring and modeling urban land-use change with multitemporal satellite data.. (2008) International Geoscience and Remote Sensing Symposium (IGARSS), 4 (1), art. no. 4779770.

Hager, J. (2006). Effects of the land allocation process and market- oriented economy on common grazing land of ethnic minorities in Northwest Vietnam. Retrieved 26/08/2009 www.indiana.edu/~iascp/bali/papers/Hager_joerg_effects.pdf

Keil, A., Saint-Macary, C., Zeller, M. Maize boom in the uplands of Northern Vietnam: economic importance and environmental implications. Department of Agricultural Economics and Social Sciences in the Tropics and Subtropics (Ed.), Forschung zur Entwicklungsökonomie und -politik – Research in Development Economics and Policy, Discussion Paper No. 4/2008

Leisz, S. J., N.T.T. Ha, N.T.B. Yen, N.T. Lam and T.D. Vien (2005). Developing a methodology for identifying, mapping and potentially monitoring the distribution of general farming system types in Vietnam's Northern Mountain Region. Agricultural Systems, 85(3) Spec. Issue):340-363.

Lillesand, T.M and R.W. Kiefer (2000). Remote sensing and image interpretation Wiley&Sons. West Sussex, England.

Lippe, M., T. Thai Minh, A. Neef, V. Hoffmann, T. Hilger and G. Cadisch (2009). Integration of interdisciplinary participative research tools: New vistas for land use modelling? Accepted Presentation, AgSAP Conference, March, 2009, Egmond aan Zee, The Netherlands.

Meyfroidt, P. (2007). Integration of heterogenous land cover maps for forest cover change detection at national scale: The case of Vietnam. Analysis of Multi-temporal Remote Sensing Images, 2007. International Workshop, Publication date: 18-20 July 2007

Meyfroidt, P., & E.F. Lambin (2008). The causes of the reforestation in Vietnam. Land use Policy, 25(2):182-197.

53

Müller, D. and M. Zeller (2002). Land use dynamics in the central highlands of Vietnam: A spatial model combining village survey data with satellite imagery interpretation. Agricultural Economics, 27(3):333-354.

Neef, A. P. Elstner & J. Hager. (2006). Dynamics of water tenure and management among Thai groups in highland Southeast Asia: A comparative study of muang-fai systems in Thailand and Vietnam. Retrieved 26/08/2009, www.indiana.edu/~iascp/bali/papers/Neef_andreas_dynamics.pdf

Nguyen, T. L., A. Patanothai and A.T. Rambo (2004). Recent changes in the composite swidden farming system of a Da Bac Tay ethnic minority community in Vietnam's Northern Mountain Region. Southeast Asian Studies 42(3):273-293.

Longley, P.A., M.F. Goodchild, D.J. Maguire and D.W. Rind (2005). Geographical information systems and science. John Wiley &Sons, Ltd, West Sussex, England.

Saha, A.K., M.K. Arora, E. Csaplovics and R.P. Gupta (2005). Land cover classification using IRS LISS III image and DEM in a rugged terrain: A case study in Himalayas. Geocarto International 20:33-40. Geocarto International Center, Hong Kong.

Saraf, A.K. (1999). IRS-1C-LISS-III and PAN data fusion: An approach to improve remote sensing based mapping techniques. International Journal of Remote Sensing 20(10):1929-1934.

Sikor, T. and D.M. Truong (2002). Agricultural policy and land use changes in a Black Thai commune of Northern Vietnam, 1952-1997. Mountain Research and Development 22(3):248-255.

Thai, T.M. (2009): Agricultural Innovation Systems in Vietnam’s Northern Mountainous Region – six decades shift from a supply-driven to a diversification-oriented system. PhD thesis, University of Hohenheim, Germany. Unpublished.

Tottrup, C. (2004). Improving tropical forest mapping using multi-date landsat TM data and pre-classification image smoothing. International Journal of Remote Sensing 25(4):717-730.

Tottrup, C. (2007). Forest and land cover mapping in a tropical highland region. Photogrammetric Engineering and Remote Sensing 73(9):1057-1065.

Tottrup, C. (2002). Deforestation in the upper Ca river basin in North Central vietnam: A remote sensing and GIS perspective. Discussion paper retrieved in library of CARES 20.04.2009 http://www.cares.org.vn/webplus/Library/Forestry.html

Tottrup, C., & M.S. Rasmussen (2004). Mapping long-term changes in savannah crop productivity in Senegal through trend analysis of time series of remote sensing data. Agriculture, Ecosystems and Environment 103(3):545-560.

Verburg, P. H., J. van de Steeg, A. Veldkamp and L. Willemen (2008). From land cover change to land function dynamics: A major challenge to improve land characterization. Journal of Environmental Management.2009 Mar; 90(3): 1327-35. Epub 21/09/2008. Retrieved 21/07/2009. http://www.ncbi.nlm.nih.gov/pubmed/18809242.

54

Wezel, A., A. Luibrand and L.Q. Thanh (2002). Temporal changes of resource use, soil fertility and economic situation in upland Northwest Vietnam. Land Degradation and Development 13(1):33-44.

Wilkie, D.S. & J.T. Finn (1996). In Pearl, M.C. (ed.) (1996). Remote sensing imagery for natural resources monitoring. Columbia University Press, USA.

Wilson, J.P., and S.A. Fotheringham (eds.) (2008). The handbook of geographic information science. Blackwell publishing, Victoria, Australia.

Yen Chau People committee (2005). Annual land use report 2005, 7/2005. (Vietnamese version).

Zhiliang, C., L. Xulong, P. Xiaochun, X. Zhencheng and W. Zhifeng (2008). Land use/cover changes between 1990 and 2000 based on remote sensing and GIS in Pearl River Delta, China. (2008). Proceedings of SPIE - The International Society for Optical Engineering, 7144, art. no. 71440N.

Zhou. Q, B. Li and B. Sun (2008). Modelling spatio-temporal patterns of land use change using multitemporal remotely sensed imagery. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008. Retrieved 25/06/2009

http://www.isprs.org/congresses/beijing2008/proceedings/7_pdf/5_WG-VII-5/01.pdf.

55

Appendices Appendix 1 Discussion sheet Ban Me and Ban Tum Group discussion

- Part 1: Land cover history by mapping

Use topography map, let the farmer draw land cover history in 3 times step 2009 – 2007; 1999; 1990 by small areas.

- Part 2: Group discussion (10 farmers from 2 villages divided to 2 small group) use A0 papers for 2 groups answer main questions.

Main questions:

1. When the did farmers in Ban Me and Ban Tum start to use upland fields? Which main crop were grown in these areas before?

- Bản Tủm – Mé bắt đầu sử dựng nương khi nào? Cây trồng chủ yếu trên các nương là gì?

- Hình thức sử dụng và canh tác trên nương trước kia khác gì so với hiện nay

Vẽ sẵn sơ đồ thời gian, yêu cầu nông dân điền vào, cả hình thức canh tác khác cũng vẽ trên sơ đồ thời gian.

2. When were the upland cropped areas expanded? Main time events?

- Có thể ko cần câu hỏi này, vì thông tin sẽ lấy được từ sơ đồ

3. How did land use change at each time events (how did cropping, plantation, project and other areas change)? Reason?

- Các lý do thay đổi các cây trồng? (điền vào sơ đồ thời gian)

- Giá cả các sản phẩm có ảnh hưởng như thế nào đến các quyết định thay đổi cây trồng?

- Những mốc sự kiện quan trọng đánh dấu sự thay đổi? (Khuyến nông, dự án hoặc chương trình)

4. When did farmers start to use chemical fertilizer? Evaluate benefits and threats?

- Phân bón được sử dụng từ khi nào? Hiệu quả và ảnh hưởng?

56

5. How did soil fertility change (how do farmers classify acc. To their knowledge: Color etc.) ?

- Cách nhận ra đất mất màu? Màu sắc đất thay đổi như thế nào?

- Nguyên nhân chính làm thay đổi màu đất?

- Cầy và cuốc đất ảnh hưởng đến màu đất như thế nào?

6. What do/did farmers do/will do to manage erosion, improve soil fertility? In the past, were measures successful or not? Why?

- Khi đất mất màu, các biện pháp xử lý đã làm gì trước kia?

- Hiện tại, các bác xử lý như thế nào?

- Tương lai, sau này khi đất mất màu, không thể cho năng suất cao, các bác định xử lý như thế nào?

- Cày đất hoặc cuốc đất ảnh hưởng đến xói mòn như thế nào?

7. Situation of erosion, sedimentation and effects on productivity and low lands? Soil erosion problem in 5 levels (in time periods).

- Hiện trạng xói mòn diễn ra như thế nào? Ảnh hưởng của xói mòn đất và phù sa đến năng suất của nương và ruộng.

- Đánh giá ảnh hưởng của xói mòn đất theo từng thời gian (đánh giá 5 mức- hiện tại để mức 5)

- Part 3: Conclusions (third day)

Present the results from part 1 and 2, correct and discuss the final result

57

Appendix 2 Field site description

Field site description

Date : ……………………….. …………Name of the Owner…….……………………..

Waypoint:………………………………Current cover crop/land cover :……………….

GPS Location: UTM 48N X…………………………….

Y.............................................

E..............................................

Slope (%)…………………………….……………………

Position:………………………………………………Weather:…………………………

Name of the area (local name):……………………………………………………………

Management:………………………………………………………………………………

Stagnic properties (yes/no):

Inherent fertility (low, middle, high): ……………………………………………………..

Sample: yes no

Top soil thickness:………………………..

Farmer’s soil classification:……………………………………………………………….

Yield for each crop

Year 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 Crop Yield (kg/ha)

Pesticides

Fertilizer

Manure

The possible and best crop for that soil:………………………… Note:

58

59

Appendix 3 Transect sheet

Transect number:………………………. Date:……………………………………………

Start: ……………………………………… End:…………..……………………………..

Photo number:

GPS Location: UTM 48N X…………………………….

Y.............................................

E..............................................

Land covers history:

Note:

Appendix 4 Summary information for final group discussion

60

Appendix 5 Original local soil map

61

62

Appendix 6 Error matrix Error matrix 1993

Class (pixel) Upland Tree Paddy Total

Class commission

(%)

Class omission

(%)

Prod Acc. (%)

User Acc. (%)

Upland 5 1 0 6 16 58.3 41.7 83.3

Tree 7 24 0 31 22 11.1 88.9 77.4

Paddy 0 2 14 16 12.5 0.0 100.0 87.5

Total 12 27 14 53 Error matrix 1999

Class (pixel) Upland Tree Paddy Total

Class commission

(%)

Class omission

(%)

Prod Acc. (%)

User Acc. (%)

Upland 27 2 0 29 6.9 0.0 100.0 93.1

Tree 0 55 0 55 0.0 3.5 96.5 100.0

Paddy 0 0 47 47 0.0 0.0 100.0 100.0

Total 27 57 47 131 Error matrix 2007

Class (pixel) Upland Tree Paddy Total

Class commission

(%)

Class omission

(%)

Prod Acc. (%)

User Acc. (%)

Upland 29 7 0 36 19.4 0.0 100.0 80.6

Tree 0 8 0 8 0.0 55.6 44.4 100.0

Paddy 0 3 10 13 23.1 0.0 100.0 76.9

Total 29 18 10 57

Appendix 7 Seasonal calendar

Feb Mar Apr May June July Aug Sep Oct Nov Dec JanAnnual seasonsRainy seasonTET festivalWeeding season

Transplanting Harvesting Transplanting Harvesting

Mango Flowering

Morning gloryLeaf Mustard

Buffalo

Labour

Income

WinterSpring Summer Autumn

Selling fingerling

Seasonal calendar of Chieng Khoi CommuneNov-06

Pest seasonRiceCaring Caring

Pest season Disease season

Fingerling

Cassava

Releasing fry

Feeding

Harvesting

Autumn crop

Maize

Land preparation

Spring-summer crop

Maize

Planting

Caring: Weeding, fertilizing… Harvesting

Land preparation

Planting

Caring: Weeding, fertilizing… Harvesting

Disease season

Disease season Disease season

Market-size fishHarvesting and selling

Feeding

Releasing fingerling

63

Appendix 8 Land cover map 2007 without application of the majority procedure

64

65

Appendix 9 Workshop summary after each day Thảo luận nhóm bản Mé và bản Tủm Phần 1- Ngày thứ 4: Làm bản đồ lịch sử sử dụng đất, sử dụng bản đồ, hướng dẫn điền thông tin

(15 hộ, 10 cũ + 5 mới chia 2 nhóm, mỗi nhóm 1 câu hỏi)

1. Đất sử dụng làm gì trong các giai đoạn 1954, 1993, 1999, 2007 và 2009. Land use in 1954, 1993, 1999, 2007 and 2009.

- Có những loại cây trồng sử dụng ở khu đất đã được khoanh trên bản đồ ở những năm 1993, 1999, 2007 và 2009

- Có những mốc thời gian nào khác liên quan đến sự thay đổi cây trồng?

2. Có những loại đất nào hiện đang có tại Bản Mé và Bản Tủm? Vùng phân bố của các loại đất đó?

- Gợi ý: Phân loại theo màu sắc hoặc theo chất đất

Ghi chú: Điền thông tin điều tra về hộ gia đình, cuối buổi.

Tổng kết: 12 hộ tham gia (danh sách kèm theo) hỏi bác Chung - Ko Hình và khẩu Thê:

1954 – 1980 là rừng tái sinh, đã từng canh tác rồi, được tái sinh sau quá trình bỏ hoang

1981-1990 làm nương lúa kiểu, còn riêng Khẩu thê bắt đầu phát năm 1990, bắt đầu trồng lúa và cả cây ăn quả.

1990-1996 Trồng lúa nương và sắn, một số hộ trồng lúa nương, một số trồng sắn. Bắt đầu trồng Tếch từ năm 1996, tre nứa?

1997 đến 2009 trồng sắn và ngô.

- Khúm nặm: Rừng tếch trồng năm bao nhiêu?

1954 – 1996 nương trồng lúa, không bao giờ bỏ hóa.

1997-2009 trồng ngô và sắn

- Mẻ Nháng: Tô Múi, Tà Thín cây ăn quả và tếch trồng khi nào?

Riêng khu đồi có đá chồng (Pom xì phí) chưa bao giờ làm nương.

1954 – 1995 là trồng lúa

1993 – 2002 Vũng bãi bằng thấp trồng dâu nuôi tằm, sườn đồi trồng lúa và sắn

66

2003- 2009 Trồng ngô và sắn

- Lòng hồ, Cụ thạo, Khúm Phúng (Cụ Thạo liền mép núi đá)

1954 – 1992 Trồng lúa nương+ Bông (ít không đáng kể), tuy nhiên bỏ hóa theo thời gian, cứ 3-4 năm bỏ hóa thì phát lại canh tác khoảng 2 năm.

1993-2002 trồng dâu, ở vùng trũng và thấp, sườn đồi thì trồng lúa nương.

1996-1997 Trồng tếch và trồng Lát, đồi thông chương 327 năm 1996-1997

2003-2005 chủ yếu là trồng sắn

2006-2009 trồng ngô

Phân hạng đất

Loại 1. Loại đất đen pha sét gômg:

- Khúm nặm

- Phu Soi

- Tà Thín (Đất bằng)

- Lọng Én

Loại 2: Loại đất đỏ vàng pha cát

- Khúm Phúng

- Lóng Nộc

- Lóng Ke

- Tát Thín (Đất cao)

Loại 3: Loại đất vàng pha sỏi

- Bó Cốp

- Khẩu Thê

- Nóng Trang

- Pom Si Men

Đất bỏ hoang: Một số năm gần đây không còn bỏ hoang nữa, một số nơi vẫn còn bỏ hoang nhưng số lượng ít không đáng kể, vì dụ như Lòng Trang còn bỏ hoang từ năm 1990 đến 1992 diện tích khoảng 2 ha còn Mẻ Nháng và Lóng Nộc thì chỉ có khoảng 2ha bỏ hoang.

Phần 2 – Ngày thứ 5:

Nhóm 1: Thay đổi tình hình sử dụng đất 8. Quá trình sử dụng đất nương diễn ra như thế nào kể từ khi bắt đầu canh tác?

Câu hỏi gợi ý

67

- Bản Tủm – Mé bắt đầu sử dựng nương khi nào? Cây trồng chủ yếu trên các nương là gì?

- Hình thức sử dụng và canh tác trên nương trước kia khác gì so với hiện nay.

- Diện tích nương thay đổi như thế nào? Thời điểm diện tích nương tăng nhanh khi nào?

(Vẽ sẵn sơ đồ thời gian, yêu cầu nông dân điền vào, cả hình thức canh tác khác cũng vẽ trên sơ đồ thời gian, chỉ sử dụng cho phần tổng kết).

9. Các nguyên nhân chính dẫn đến những thay đổi về hiện trạng sử dựng đất trên?

Câu hỏi gợi ý

- Các lý do thay đổi các cây trồng, hoặc chuyển đổi mục đích sử dụng đất (ví dụ chuyển từ đất rừng sang đất nương, hoặc từ đất nương thành vườn cây ăn quả)?

- Giá cả các sản phẩm có ảnh hưởng như thế nào đến các quyết định thay đổi cây trồng?

- Những mốc sự kiện quan trọng đánh dấu sự thay đổi? (Khuyến nông, dự án hoặc chương trình)

Tổng kết Thay đổi các hình thức quản lý sản xuất:

- Từ năm 1954 đến năm 1959 đất được sử dụng riêng lẻ

- 1960 đến năm 1961 làm tổ đổi công (có nghĩa là một nhóm họ đổi công giúp nhau một số buổi)

- 1962 đến 1986 đổi thành hợp tác xã chấm công

- 1986 đến 2009 Canh tác theo khoán 10 (là canh tác giao đất cho các hộ gia đình và nộp thuế theo sản phẩm, khoản hẳn ruộng cho các hộ nông dân).

- 1995-2009 Chuyển từ lúa sang Sắn và Ngô

Hình thức canh tác đất:

- Từ năm 1954 – 1979 chỉ đốt dọn cào và trồng không bón phân

- 1980-2000 Cào và cuốc không bón phân

- 2000-2009 Cầy cuốc đến năm 2004 sử dụng phân bón gồm Lân, đạm, NPK kết hợp phát triển mạnh trồng ngô

- Bắt đầu từ năm 1994 diện tích đất được khai thác triệt để

Các nguyên nhân thay đổi cây trồng:

- Do hạn hán, đất bạc màu, hiệu quả kinh tế thấp nên chuyển sang trồng sắn xen ngô

68

- Đến năm 1992, thực hiện chủ trương của huyện, đất bãi bằng chuyển sang trồng dâu nuôi tằm, nhưng do chất lượng kém, giá cả bấp bênh nên đến năm 1997 chuyển sang trồng ngô nhưng sắn xen ngô.

- Cũng vảo năm 1992, thực hiện chủ trương của huyện và dự án 327 đất đồi núi trọc được trồng cây ăn quả và trồng rừng.

Bổ sung: Chủ yếu là 2 nguyên nhân chính:

- Đất mất màu và chuyển màu, nghèo dinh dưỡng

- Kinh thế thị trường thay đổi

Sử dụng phân hóa học đẩy nhanh quá trình xói mòn đất, ví dụ bón phân đạm làm đất vón thành cục, xói mòn thành cục nên cũng nhanh hơn

Rừng – Lúa –Lúa-Lúa – Sắn – sắn – sắn- lúa – sắn –sắn – sắn – lúa .

Nhóm 2: Dinh dưỡng đất

10. Màu đất và độ màu của đất thay đổi như thế nào từ khi bắt đầu canh tác đến nay?

Câu hỏi gợi ý

- Cách nhận ra đất mất màu? Màu sắc đất thay đổi như thế nào?

- Nguyên nhân chính làm thay đổi màu đất?

- Cầy, cuốc đất và các biện pháp làm đất khác ảnh hưởng đến màu đất như thế nào?

4. Với các loại đất khác nhau, mục đích sử dụng khác nhau như thế nào? Trình bày các loại đất phù hợp với mỗi loại cây trồng?

Câu hỏi gợi ý

- Với từng loại đất, trồng cây gì hoặc sử dụng làm gì?

- Lý do sử dụng với mục đích đó?

Tổng kết

- 1954 – 1980 hầu hết màu đen màu mỡ

- 1980 – 1992 Chuyển sang màu nâu

- 1993 – 2009 đất chuyển sang màu vàng, có nơi màu đỏ và có sỏi

Cây trồng phù hợp :

Màu đen, các loại cây

Màu nâu, hầu hết trồng sắn và ngô hoặc sắn xen ngô

Nếu có đá và sỏi có thể trồng cây họ đậu, bông hoặc dứa, chuối

Tăng dinh dưỡng đất bằng phân hóa học, chưa sử dụng phân chuồng.

69

Nhóm 3: Tình hình xói mòn

11. Tình hình xói mòn đất diễn ra như thế nào trên đất nương?

Câu hỏi gợi ý

- Đất bắt đầu có hiện tượng xói mòn khi nào?

- Khi đất mất màu, các biện pháp xử lý đã làm gì trước kia? Hiện tại, các bác xử lý như thế nào?

- Cày đất hoặc cuốc đất ảnh hưởng đến xói mòn như thế nào?

12. Đánh giá mức độ xói mòn đất theo từng thời gian

- Các loại xói mòn, xói mòn rãnh nhỏ, xói mòn rãnh lớn

- Đánh giá mức độ (Đánh giá 5 mức)

Tình hình xói mòn

1954 – 1980 Xói mòn ít, chậm, do luân phiên cây trồng, có thời kỳ bỏ hóa, cho nên độ mòn bình quân khoảng 3cm/năm

Tuy nhiên được bồi bổ xung 1-2 trong 2 hoặc 3 năm bỏ hoang hoặc do lá sắn bồi thêm, đặc biệt là vùng đất thấp được bồi hàng năm 1-2 cm

1981-1992 Xói mòn nhanh hơn, nhiều hơn, bồi bổ xung ít hơn không đáng kể khoảng 1cm sau thời gian trồng sắn 2 năm. Ở khu vực trồng lúa nương, do có thời kỳ đất trống chờ mưa. Độ che phủ ít cho nên xói mòn diễn ra nhanh hơn. Độ bồi hàng năm không đáng kể

1993-2009 Xói mòn diễn ra nhanh hơn, nhiều hơn, nghiêm trọng hơn, đất bề mặt gần như biến mất, đổi màu đất, trơ sỏi đá, không còn được bồi đất

Các biện pháp chống xói mòn:

- Lấy cây gỗ chắn ngang nương, chặn dòng nước

- Lấy cây sắn, ngô sau thu hoạch bó lại và chắn ngang nương

- Trồng sắn: lâu năm giữ màu và chồng xói mòn và bổ sung màu đất.

- Băng cây xanh (cây cốt khí, keo dậu)

- Trồng dứa theo hàng

Phần 3: Ngày thứ 5 ,Tổng kết

Trình bày những kết quả thu được theo chủ đề

1. Lịch sử sử dụng đất

2. Thay đổi tình hình sử dụng đất

70

3. Dinh dưỡng đất

4. Tình hình xói mòn đất

Thảo luận kết quả cuối cùng Những thông tin bổ sung: Các dự án đã từng có:

- 1992 có dự án 327 cho đến năm 1995

- Xẻ phí bắt đầu trồng Teak và Lát vào năm 1996-1997 khu nhà bác Ân. Tuy nhiên Teak không phát triển tốt một số cây được trồng bổ sung vào năm 1998 và 1999.

- Đến năm 1999 kết thúc trồng rừng và bắt đầu dự án Việt Đức.

- Tại nương nhà chú Thái: Những cây lớn: được trồng vào năm 1995, những cây nhỏ được trồng theo dự án Việt Đức năm 1999

- Dự án Việt Đức tại Xẻ Phí chủ yếu có 6 hộ làm.

- Dự án 327 làm tại Khẩu Thê, Kò HÍnh, Mẻ Nháng

Đa số chưa khép kín.

- 661 là dự án quản lý và chăm sóc của dự án đào nợ Sông Đà.

- Khu vực rừng nhà chú Xuân khu Khúm Nặm

- Năm 661 tại Mai Sơn Yên Châu.

- Rừng phòng hộ nhà nước

- Sau đó Nương trở thành rừng trồng bảo vệ, rừng tái sinh bảo vệ với diện tích khoảng 14.7ha.

- Năm 1999, giao sổ đỏ trên nương ruộng.

- Mức xói mòn các khu vực: Khúm nặm -> Khúm Phúng->Tà Thín, Mè Nháng->Xé Phí->Khẩu Thê-> Lòng Hồ

- Một số biện pháp chống xói mòn: Phụ thuộc vào độ dầy của mùn, do chất đất. Một số biện pháp được nông dân được khuyến nông hướng dẫn như: Cầy theo đường đồng mức, cầy theo băng hoặc trồng cây băng xanh.

Các kết quả tổng kết cơ bản không thay đổi trong các ngày, không có ý kiến trái ngược, kết quả trên 2 bảng