land cover dynamics (1990-2002) in binh thuan province, southern central vietnam

7/31/2019 Land Cover Dynamics (1990-2002) in Binh Thuan Province, Southern Central Vietnam

1/15

Online Publication Date: 15 March 2012

Publisher: Asian Economic and Social Society

Land Cover Dynamics (1990-2002) in Binh Thuan Province,Southern Central Vietnam

HOUNTONDJI Yvon-Carmen (Department of NaturalResources Management, Faculty of Agronomy, University ofParakou, BP : 123 Parakou, Benin)

De LONGUEVILLE Florence (Department of Geography,FUNDP-University of Namur, Rue de Bruxelles 61, 5000 Namur,Belgium)

OZER Pierre (Department of Environmental Sciences andManagement, University of Liege Avenue de Longwy 185, B-6700 Arlon, Belgium)

Citation:Hountondji YC, De Longueville F, Ozer P (2012): Land Cover Dynamics (1990-2002)in Binh Thuan Province, Southern Central Vietnam International Journal of Asian SocialScience Vol.2, No.3, pp. 336-349.


2/15

International Journal of Asian Social Science, 2(3), pp.336-349

336

Author(s)

HOUNTONDJI Yvon-CarmenDepartment of Natural ResourcesManagement, Faculty of Agronomy,University of Parakou, BP : 123 Parakou,Benin.

E-mail:[email protected]

De LONGUEVILLE FlorenceDepartment of Geography, FUNDP-University of Namur, Rue de Bruxelles 61,5000 Namur, Belgium

OZER PierreDepartment of Environmental Sciences andManagement, University of Liege Avenuede Longwy 185, B-6700 Arlon, Belgium

Land Cover Dynamics (1990-2002) in Binh Thuan

Province, Southern Central Vietnam

Abstract

This paper describes the use of satellite imageries and GIS datafor identifying key environmental characteristics of BinhThuan Province in south central Vietnam and for detecting themajor changes patterns within this region. Landsat TM (1990)and Landsat ETM+ (2002) imageries were used to classify thestudy area into seven land use and land cover (LULC) classes.A post-classification comparison analysis was used to quantifyand illustrate the various LULC conversions that took placeover the 12-year span of time. Results showed that a steady

growth in population has caused extensive changes of landcover throughout the area. The maps also indicate that the lossof woody land (forest) and the extension of wetlands (irrigatedarea), combined with built-up encroachment, remains one ofthe most serious environmental problems facing the BinhThuan Province today. The post-classification change detectionanalysis showed that critical habitats accounted for nearly38.5% of the intensive study area between 1990 and 2002while 61.5% remained stable. Results also showed over the 12-year span, approximately 1151.2 km (115.120 ha) forest wereconverted respectively to brush, irrigated area (wetlands),

cropland and built-up. This is an overall average decrease of9594 hectares of forested area per year. Throughout the studyarea, districts most affected by forest conversion to anotherland cover are: Bac Bihn (2798 ha/year), Than Linh (2717ha/year), Ham Thuan Nam (1601 ha/year) and Ham Thuan Bac(1524 ha/year). Based on the identified causes of thesechanges, we made policy recommendations for bettermanagement of land use and land cover.

Key words: Remote sensing, Land cover, Change detection, Binh Thuan, Vietnam

Introduction

The research on the land use/cover change(LULC) is one of the frontiers and the hot spotsin the global change research. The urbangrowths, deforestation, extension of croplandsare the main talks in the current scenario ofrapid climate changes. The land cover of aspecific area over a time period can provide usinformation on how sustainable the land hasbeen used. While standard ground surveymethods for undertaking such measurements are

imperfect or expensive it has been demonstratedthat satellite-based and airborne remote sensing(RS) systems offer a considerable potential.

This capacity for quantitative land-surface

monitoring over large areas makes RS well-suited for a very wide range of disciplines,including land-use planning and providingspatial information needed for local or regional-scale analyses of the relationships betweenclimate change, land degradation anddesertification processes (Singh, 1989).Studying the change detection by remotelysensed data of different dates remains achallenge because of factors such as thecomplexity of the landscape in a study area.

Change detection is the process of identifyingdifferences in the state of a feature orphenomenon by observing it at different times
mailto:[email protected]:[email protected]:[email protected]:[email protected]


3/15

337

(Imbernon, 1999). This technical issue is usefulin many applications related to land use andland cover (LULC) changes, such as shiftingcultivation and landscape changes (Serra et al.,2008), land degradation and desertification(Adamo and Crews-Meyer, 2006; Gao and Liu,2010), urban sprawl (Yuan et al., 2005),deforestation (Schulz et al., 2010; Wyman andStein, 2010), landscape and habitatfragmentation and other cumulative changes(Nagendra et al., 2006). Continual, historical,and precise information about the LULCchanges of the Earths surface is extremely

important for any kind of sustainabledevelopment program, in which LULC serves asone of the major input criteria. Satellite remote

sensing is the most common data source fordetection, quantification, and mapping of LULCpatterns and changes because of its repetitivedata acquisition, digital format suitable forcomputer processing, and accurategeoreferencing procedures (Lu et al., 2004;Chen et al., 2005). The successful use ofsatellite remote sensing for LULC changedetection depends upon an adequateunderstanding of landscape features, imagingsystems, and methodology employed in relation

to the aim of analysis (Yang and Lo, 2002;Sivrikaya et al., 2007). Many change detectiontechniques have been developed and used formonitoring changes in LULC from remotelysensed data, such as post-classificationcomparison (PCC), image differencing,principle components analysis, and vegetationindex differencing (Lu et al., 2004). The PCCmethod, which is recognized as the mostaccurate change detection technique, detectsland cover changes by comparing independentlyproduced classifications of images fromdifferent dates. Using the PCC method thusminimizes the problems associated with multi-temporal images recorded under differentatmospheric and environmental conditions. Dataacquired from different dates are separatelyclassified, and hence, reflectance data frommulti-dates do not require adjustment for directcomparability (Coppin et al., 2004; Zhou et al.,2008). There are currently numerous satelliteprograms in operation. For change detectionstudies, the Landsat program is unique because

it provides an historical and continuous recordof imagery. Landsat images can be processed torepresent land cover over large areas and over

long time spans, which is unique and absolutelyindispensable for monitoring, mapping, andmanagement of LULC (Wulder et al., 2008).

This paper aims to investigate spatial andtemporal land cover changes in Binh Thuanprovince in the southern central Vietnam and tounderstand the possible causes of the changes.To do this, we applied land classificationschemes to classify the land cover types usingLandsat data focus on six districts included inBinh Thuan province. Based on the previouslydeveloped methodology such as maximum-likelihood classification and change detectiontechniques, we assessed the accuracy of the landclassification techniques by comparison with

supervised classification based on numerousground truth data. Land cover changes between1990 and 2002 were quantitatively presentedwith the results of accuracy assessments. Wealso suggested recommendations regardingtowards better management of LULC.

Data and Methods

Study area

Locate between 10724 10850E and10331133N (Fig. 1), Binh Thuan is a seacoast province of south of central Vietnam,having north-east border with Ninh Thuanprovince, north-west border with Lam Dongprovince; west border with Dong Nai province;south-west border with Ba Ria - Vung Tauprovince; east and south-east border with theSouth China Sea. Total land area is 7830.4 km(General Statistical Office, 1994) and theprovince has 8 districts, 1 city and 125 towns,

communes.This province is the driest and hottest region ofVietnam. The region faces the Pacific Ocean tothe east with a coastline of about 192 km and ischaracterized by a combination of tropicalmonsoon and dry and windy weather. TheBinh Thuan Province can be divided into 4 mainlandscapes: Sand dunes along the coast -alluvial plains - hilly areasand the Truong Sonmountain range. The mean annual temperatureis 27C, with average minimum 20.8C in thecoldest months (December, January), and anaverage maximum of 32.3C in the hottestmonths (May and June). Rainfall in this area is


4/15


338

limited and irregular. Annual averageprecipitation is 1024 mm, while evaporation insome years is equivalent to precipitation. Atsome locations annual rainfall can be as low as550 mm. The dry season is from November toApril, with 60 days of January and Februaryhaving almost no rain. The rainy season is from

May to October with heavy rains concentratedin a short periods with up to 200 mm/day. Thetotal population of the province in 2004 isaround 1.140.429 so urban conditions haveimproved in recent years through the extensionof water supply networks and road upgrading.

Fig-1 Location of Binh Thuan province and area covered by Landsat ETM+ image subset(05.01.2002) False color (RGB: TM3, TM 4, TM 2)

Data acquisition and processing

A pair of cloud-free Landsat images wasselected to classify the study area: December 30,1990 (Landsat-5 TM) and January 05, 2002(Landsat-7 ETM+). These time series of

Landsat images are freely available from theLandsat archive from the United StatesGeological Survey (USGS). All visible and

infrared bands (except the thermal infraredband) were included in the analysis. Only theintersected area (province boundary and imagesubset) has been taken into account for spatialstatistics. Six of the nine districts of Binh Thuanprovince were contained within Landsat path

124, rows 52. All images were rectified to UTMzone 48N, GRS 1984, using at least 35 welldistributed ground control points. The root mean


5/15

339

square errors were less than 0.25 pixel (7.125m) for each of the two images.

We used the Environment for Visual Imagingsoftware: ENVI version 4.4 (ENVI, 2006)developed by Research System Incorporated(RSI) for digital image processing of the subsetsof the chosen Landsat images described above.Image processing included images calibration toreflectance, enhancement, rectification, and sub-setting. The accuracy of the classification wasdetermined based upon ground truth region ofinterest (ROI). Each class included a differentnumber of ROIs, depending upon the albedo

and contrast of class types. Training sampleswere selected for each of the predetermined

LULC types by delimiting polygons aroundrepresentative sites. Using the pixels enclosed

by these polygons, we derived spectralsignatures for the respective land cover typesrecorded by the satellite images. A spectralsignature is considered to be satisfactory whenconfusion among the land covers to be mappedis minimal (Gao and Liu, 2010). Once thespectral signature was deemed satisfactory, weentered it into the classification process. Groundtruth ROIs were divided into two groups. The

first group is for classification procedure and thesecond group is for accuracy assessment. TwoLULC maps were produced (for 1990 and2002). Our classification scheme, with sevenclasses (Table 1), was based on the land coverand land use classification system developed(Bauer et al., 1996) for interpretation of remote

sensing data at various scales and resolutions.

Table-1 Land cover classification schemeLand cover class Acronym DescriptionBuilt-up BUP Urban or rural build-up - industrialtransportationcities

towns -villagesForest-Plantation FPO Natural forest - reforested land - mixed forestorchards -

grovesDunes and sandbank DSB Dune roving - sandbankCrop land CLD All arable land (not limited to land under crops) rain-fed

crop fields and bare fields

Highland and Bush HIB High lands or hills with vegetation wild lands with sparevegetationunused land

Water bodies WAT Permanent open waterlakesreservoirsstreams - baysand estuaries

Lowland WET Irrigated cropland - paddy field - water ponds - floodedlands.

A supervised training approach was used withmaximum likelihood classification (Bauer etal., 1996). Since several researchs have

produce different land classes (Le et al., 2002;Yaun et al., 1998) than those of our Landsatclassification, data from both sources wereaggregated into seven categories. Classhistograms were checked for normality andsmall classes were deleted. Post-classificationrefinements were applied to reduceclassification errors caused by the similaritiesin spectral responses of certain classes such asbare fields and urban and some crop fields andwetlands. An independent sample of an

average of 40 polygons, with up to 100 pixelsfor each selected polygon, was randomlyselected from each classification to assess

classification accuracies. Error matrices ascross-tabulations of the mapped class vs. thereference class were used to assess

classification accuracy (Foody, 2002). Overallaccuracy, users and producers accuracies, and

the Kappa statistic were then derived from theerror matrices. The Kappa statistic incorporatesthe off diagonal elements of the error matrices(i.e., classification errors) and representsagreement obtained after removing theproportion of agreement that could be expectedto occur by chance. Finally, a 3x3 majorityfilter was applied to each classification torecode isolated pixels classified differently than

the majority class of the window.


6/15


340

Change DetectionFollowing the classification of imagery fromthe individual years, a post-classificationcomparison change detection algorithm wasused to determine changes in land cover in19902002 intervals. This is the most commonapproach to change detection (Jensen, 2004)and has been successfully used to monitor landuse changes in previous studies (Coppin, et al.,2004; Kamusoko and Aniya, 2007; Zhou et al.,2008). Post-classification comparison (PCC)was employed to detect the differencesbetween each pair of LULC maps (1990 and2002). The PCC approach provides fromto

change information and the kind of landscapetransformations that have occurred can be

easily calculated and mapped. A changedetection map with 49 combinations of from

to change information was derived for each of

the two seven-class maps and then, a changemap was compiled to display the specificnature of the changes between the classifiedimages.

Results

Results presented in the following sections arerestricted to the area of Binh Thuan provincefor which the chosen Landsat scenesintersected with provincial boundaries. Thisoverlapped area covers 6070 km representing 78.9% of the total area of theprovince.

Classification maps accuracies and statistics

Before using the classification results forchange detection, we assessed their validity bytesting the results against ground truth data.

Error matrices were used to assess

classification accuracy and results aresummarized for the two years in Table 2.

The overall accuracies for 1990 and 2002 were,respectively, 98.05% and 91.41%, with Kappastatistics of 96.61% and 87.98%. Users andproducers accuracies of individual classeswere consistently high, ranging from 99% to76%. FPO, WAT and HIB classes were allcharacterized by the highest classificationaccuracy. This is because the integration of theresults of visual interpretation and supervisedclassification, which allowed us to correct themisclassified pixels.

The data presented in Fig. 2 and Table 3

represent respectively the rate and the total areaof each LULC class for 1990 and 2002. In1990, HIB was the largest class, representing2407 km of the total LULC categoriesassigned, occupying 39.7% of the study area.In 2002, this class amount had increasedsignificantly, resulting chiefly in the reductionof forest and plantation class (from 30% in1990 to 20.7% in 2002). With the exception ofthe lowland class (WET: 4.9%) and the duneroving (DSB: 2.3%) in 1990, built-up (BUP)

and free-water bodies classes (WAT)accounted for the lowest percentages of thetotal LULC categories in all studied years.

The major changes during 1990-2002 periodswere found in FPO class (natural forestsincluding planted forest), WET class (irrigatedland, paddy fields) and HIB class (vegetationre-growth, shrubs, hills vegetation). The areasof water bodies and built-up (urban or ruralsettlements) also increased significantly. Thus,

from 1990 to 2002, BUP, HIB, WAT and WETarea increased respectively 0.7% (44.3 km)

Table-2 Summary of Landsat subset classification accuracies (%) for 1990 and 2002

Land cover class Acronym1990 2002Producers Users Producers Users

Built-up BUP 97.59 87.22 80.04 76.66Forest-Plantation FPO 99.47 77.58 98.48 85.75Dunes - sandbank DSB 98.13 99.58 88.66 99.63Crop land CLD 94.82 98.05 95.85 94.39Highland and Bush HIB 98.80 99.95 91.25 99.96Water bodies WAT 99.18 99.98 99.96 98.17

Lowland WET 97.89 78.58 95.26 86.51Overall accuracy (%) 98.05 91.41Kappa statistic(%) 96.61 87.98


7/15

341

Fig-2 Area percentage of each land cover class resulting from the classified images (1990 and2002)

3.8% (233.6 km), 1.1% (67.1 km) and 5.6%(341.2 km) while FPO, DSB and CLDdecreased 9.3% (562 km), 0.5% (31.5 km) and1.5% (92.2 km). Although the extent ofwetlands may change from year to year due tovarying precipitation and temperature, thevariation in wetland area is also likely due to

classification errors. However, the fluctuationsin lowlands classification errors are believed tobe related to varying sol humidity levels giventhe high classification (producers accuracy

about to 97.89% in 1990). The spatialdistribution of the assigned categories for eachyear is shown in Fig. 3

Table-3 Area of each land cover class resulting from the classified images

Land cover class Acronym

1990 2002

Area(km) Rate(%) Area(km) Rate(%)Built-up BUP 60.8 1.0 105.1 1.7Forest-Plantation FPO 1818.1 30.0 1255.4 20.7Dunes - sandbank DSB 141.9 2.3 110.4 1.8Crop land CLD 1332.2 21.9 1240.0 20.4Highland and Bush HIB 2407.0 39.7 2640.6 43.5Water bodies WAT 13.4 0.2 80.5 1.3Lowland WET 296.6 4.9 637.8 10.5Total (km) 6070 100 6076 100


8/15


342

Fig-3 Land use and land cover maps (1990 and 2002) resulting from the integration of supervisedclassification and visual interpretation

Analysis of change patterns

By constructing a change detection map, theadvantages of satellite remote sensing inspatially disaggregating the change statistics

can be more fully appreciated. Our resultsrevealed that over the 12-year span, 3735.5 km(61.5% of the studied area) were unchangedwhile changes occurred elsewhere (38.5%)(Table 4).

These unchanged areas (not shown) weremainly composed by HIB (34%), CLD (12.7%)FPO (11%) and WET class (2.2%). In order toexamine more specifically areas where changesaffected more than 5 km, a table showing

combinations of from

to change has beengathered (Table 5). The selected resultsrepresented 99% of the major changes that

occurred during the study period and the spatialdistribution of the resulting conversion classesis shown in Fig. 4. Overall, 82.8 km of built-upwas converted from forests and agriculturalland (rainfed: CLD and irrigated: WET), while

in the same time, 39 km of urban or ruralsettlement land was converted to forest andcropland. These changes may seem to beclassification errors, but forested areas areamong some of the most sought after areas fordeveloping new settlements. Streets, roads andrural tracks were generally classified as urban,but when urban tree canopies along the streetsgrow and expand, the associated pixels may beclassified as forest. Also, the results of imageinterpretation and classification together with

an access to archives on agricultural policiesindicated that an intensive developmentprograms have been implemented in the study


9/15

343

area during the last two decades (World Bank, 1996).

In 21.1% of the cases the change was Forest(FPO) to wild-land with spare vegetation(HIB) and approximately 14% was forest to

irrigated agriculture change. Table 5 alsoreveals that residential uses comprise over halfthe cases that changed to urban. Relatively rareand unlikely types of conversions, such asagriculture to forest, and then to urban uses and

forest to agriculture, and then to urban, totaling5%, are assumed to largely be classificationerrors. The spatial interpretation of these results

indicates that urban (BUP) increases mainlycame from conversion of FPO, CLD and WETclasses (representing 80.2% of the change rate).

Table -4 Area and percentage of change in each LULC class and annual rate of change between thedifferent dates

LULC class

Districts includedTotalchange

(km)

Rate(1990-

2002)

Bac

Binh

Duc

Linh

Ham

ThuanBac

Ham

ThuanNam

Phan

Thiet

Tanh

LinhBUP 17.7 29.1 13.5 6.3 6.2 17.0 89.9 1.5FPO 149.2 78.1 162.8 77.3 37.6 83.6 588.6 9.7DSB 14.2 0.1 9.4 0.5 10.5 0.3 35.1 0.6CLD 187.7 36.6 92.7 68.9 18.4 65.4 469.6 7.7HIB 108.8 54.6 64.0 94.8 1.2 252.5 575.9 9.5WAT 10.9 7.9 27.0 2.2 3.6 18.0 69.6 1.1WET 204.2 16.2 130.3 118.5 3.4 33.1 505.7 8.3Changed area(km)

692.7 222.8 499.7 368.5 80.8 469.9 2334.4 38.5

Unchanged area(km) 1169,0 323,3 845,2 536,1 130,9 731,0 3735,5 61,5District area (km) 1861.7 546.1 1344.9 904.6 211.7 1200.9 100

Fig-4 Change detection map for the studied area (spatial extends of converted classes arehighlighted).


10/15


344

These programs included land reclamation andthe expansion of irrigated culture areas, as wellas the establishment of rural villagecommunities into foothills of mountainspreviously considered as wild-lands or unusedlands. Further GIS analysis revealed a strongrelationship between new development andproximity to roads. Almost half (49%) of theBuilt-up development detected in ourclassifications occurred within 2 km of mainroads and rural tracks, and 35% was between 2and 4 km. Duc Linh, Tanh Linh, Bac Binh andHam Thuan Bac are the districts where thisurban sprawl is the most remarkable during thestudied (19902002). All of these (bio) physicalchanges within the setting of the regions

ecosystems reflect the dynamics of humanimpacts on the study area.

Cropland (CLD) area declined from 1332.2 kmin 1990 to 1240 km in 2002. This class wasmainly changed into planted vegetation (FPO),irrigated land (WET), Brush (HIB) and Built-up(BUP) accounting for almost 22% of the totaloccurred changes. Ham Thuan Bac, Bac Binhand Duc Linh are the districts mainlyaffected by this kind of conversion.

The area of water bodies plays an importantrole in development of aquaculture for thestudy area. During the studied period, waterbody increased by about 6.5 times of theoriginal area. Water occupied 13.2 km (0.2%of total studied area) in 1990 and finallyreached 80.5 km (1.3%) in 2002. This wasconverted mainly from natural forest,plantations (FPO), cropland (CLD), bush(HIB), lowland and paddy fields (WET).The most plausible explanation of thisconversion into water bodies was mainly due tohydroelectric dam constructions (Bac Binh,Ham Thuan Bac Tanh Linh districts) and alsoprobably to expansion of small reservoirs forirrigation.

Lowland with irrigated agriculture fields(WET) also showed increasing trends in sizefrom 1990 to 2002. The area of paddy fieldswas only 296 km (4.9% of total area) in 1990,increased to 637 km (10.5%) in 2002. This

class extends on the foothill of the mountainchains and chiefly into hilly valleys followinggeographical orientation of the alluvial basins.

This was converted mainly from natural forests(FPO), cropland (CLD) and HIB classvegetation and some from water bodies andmixed orchard.

Even though many hectares of natural forestswere converted into planted vegetation andother agricultural land from 1990 to 2002, therewas also conversion from barren lands tocropland and forests. This change occurred ondune and sandbak (DSB) class where decreaseoccurred from 141.9 km to 110.4 km. Theseareas are used primarily for horticulturalactivity, with only very small areas of rain-fed

rice grown in low-lying areas. Anotherexplanation of this reverse trend is that therewas a wide application of vetiver grass(Vetiveria zizanioides) planting that has realimpact on land stabilization / reclamation.Although the concept of using vetiver forvarious applications has only been introducedinto Vietnam in 1999 (Le Van and Truong,2003) vetiver has become widely knownthroughout the country with numeroussuccessful applications for natural disaster

mitigation and environmental protection.Typical examples include road batter stabilityenhancement, erosion / flood control ofembankments, dykes, riverbanks, sand dunefixation. However, we can unfortunately noticea persistence of sandbank and dune roving,resulting probably of land degradation in BacBinh (9.7 km), Ham Thuan Bac (8.4 km)districts and Phan Thiet (10.3 km).

Discussion

Within a few years, Vietnam was not onlyproducing enough food for domesticconsumption but became one of the worlds

leading rice exporters (World Bank, 1996).Unfortunately, in some regions, farmersprofited much less from the reforms thanothers. There is wide disparity in regionaldevelopment trends following land reformspartly because the reforms were notimplemented homogeneously throughout thecountry and also because of the tremendous

diversity of natural and human environmentsthey were applied to.


11/15

345

Table-5 Area and percentage of change in each LULC class and annual rate of change between1990 and 2002

Change class Districts and city affected by changesTotalarea

(km)

Change rate(%)

From To

Bac

Binh

Duc

Linh

Ham

ThuanBac

Ham

ThuanNam

Phan

Thiet

Tanh

Linh

FPO

HIB 91.3 46.6 40.6 76.4 0.5 237.5 492.8 21.1

49.1WET 136.2 8.1 75.6 82.7 2.0 20.1 324.6 13.9CLD 98.3 34.8 57.7 30.3 2.4 58.4 282.0 12.1BUP 4.5 15.6 3.6 2.0 1.5 6.1 33.4 1.4WAT 2.3 2.7 4.6 0.8 0.2 3.9 14.4 0.6

CLD

FPO 57.6 55.9 89.7 36.4 30.5 35.1 305.2 13.1

24.1

WET 36.3 7.5 36.1 29.7 1.3 9.6 120.5 5.2HIB 9.7 7.1 16.1 5.0 0.6 8.3 46.8 2.0BUP 9.2 12.6 7.2 2.6 2.7 9.4 43.7 1.9

DSB 9.7 0 8.4 0.3 10.3 0.0 28.7 1.2WAT 2.3 3.0 3.7 0.5 1.2 6.2 17.0 0.7

HIB

FPO 70.2 11.7 55.3 27.1 2.0 44.0 210.3 9.0

14.4WET 30.9 0.1 18.0 5.4 0 3.2 57.7 2.5CLD 25.0 0.3 5.9 1.3 1.1 4.1 37.7 1.6WAT 3.4 1.4 16.7 0.3 1.5 6.7 30.0 1.3

WET

CLD 30.4 0.6 19.2 33.7 0.9 1.3 86.1 3.7

7.0

FPO 8.9 2.8 8.5 10.2 0.2 2.8 33.5 1.4HIB 6.9 0.4 6.5 13.2 0.1 6.2 33.3 1.4BUP 2.5 0.5 0.9 1.4 0.1 0.4 5.8 0.2WAT 1.9 0.7 1.6 0.6 0.2 0.4 5.5 0.2

DSB CLD 29.7 0.2 5.8 1.7 13.2 0.2 50.7 2.2 2.7FPO 4.7 0.8 2.4 0.5 3.6 0.4 12.4 0.5

BUPFPO 7.8 6.4 6.7 3.2 1.2 1.1 26.4 1.1

1.7CLD 4.2 0.7 3.5 2.0 0.8 1.3 12.5 0.5Total (km) 683.9 220.5 494.5 367.0 78.1 466.8 2310.9 99.0 99.0

This resulted in different ways of interpretinglocal success or failure of the land policy to liftmarginal farmers out of poverty or reverse landdegradation trends (Ducourtieux and Castella,

2006).Land use in the study area has undergonedramatic change and was impacted by humanresource development. Our findings from thesatellite-image interpretation suggest a decreasein closed canopy forest due to conversion intorain-fed and irrigated agricultural land. Opencanopy forest cover (including bush andfallows) increased during the 1990sapproximately by 4%, mainly due to the naturalregeneration of mixed grassland. Followedfields formerly used for shifting cultivation maybe largely abandoned during the last decade andregenerated to become open canopy secondaryforest. Overall, rain-fed mixed agriculture

decreased slightly in the 1990s, as the reductionin shifting cultivation was compensated by anexpansion of irrigate cropping area. Waterbodies also showed increasing areas from 1990

to 2002, but the overall change was not largefor these types. Increases in irrigated fields andwater bodies represent the possibility ofincreasing aquaculture or increasing thesuitability of the study area for futureaquaculture. It has been demonstrated that interms of economic returns, aquaculture oftengives higher returns than rice culture (Salam etal., 2003). However, the decision to convertpaddy fields to fish ponds is often related tofood security and social aspects. In order tomaintain a balance between social andeconomic aspects, integrated rice-fish culturesystems should be promoted (Gregory and


12/15


346

Guttman, 2002). Outreach and extensionservices to local farmers should be enhanced,and extension information should be produced.Over time, agricultural production becamemore locally concentrated. These changes inland use suggested that shifting cultivation asthe traditional farming system practiced by thepopulation in the research area almost entirelydisappeared in its traditional form during thelast decade. On the other hand, provincial anddistrict authorities generally lack of manpower,resources, capacity and experience to put intopractice consistent and participative land-usepolicy. They should be allocated moreresources and more time to implement a policythat would satisfy the different objectives

expressed at different hierarchical levels.Experience has shown that promoting the linksbetween land allocation and subsequentextension activities is indispensable for closeinteraction with farmers and when directingresearch to development activities that arerelevant and acceptable to local people. In abroader context, some studies are summarizingthe general factors that promote a successfultransition towards sustainable management ofnatural resources (Raquez and Lambin, 2007).

Three broad groups of factors that have beenidentified: information regarding the state of theresources and their degradation, motivations tosearch for solutions, and capacity to implementeffective solutions. Environmental degradationof places with which people maintain anemotional connection can lead these people toadopt a more ecological worldview and toreassess their involvement in conservationactivities (Marshall et al., 2005; Rogan et al.,2005). The perception of the capacity to predictand improve the state of the environment alsomediates behavior (Ostrom, 1999).

Overall, natural forests declined in area, whilethe most dramatic increase was for irrigatedfields. During the 12-years span, we canobserve a pathway of land expansion intopreviously uncultivated areas. At the same time,due to market development, agriculturalintensification is concentrated in the mostsuitable regions. Large areas of land marginallysuitable for agriculture are therefore abandoned

and left to forest regeneration. In this forestscarcity path, political and economic changeswill arise as a response to the growing scarcity

of forest products and decrease in the provisionof ecosystem services following. We think it ispossible that intensification of agriculturecombined with the (enforced) protection offorested areas can reduce the pressure onforested land, and slow down or even halt theexpansion of agricultural land if coupled withthe right policy instruments.

Concerning the undesirable changes in land useand the need for more sustainable developmentin the region, we propose that the governmentshould reconsider the policies applied to theregion of the study area, as well as the policiesof the surrounding regions that may directly orindirectly affect the development of the study

area. For instance, the modeling power of GISto evaluate land suitability for development ofagricultural activities chiefly in the basins andwatershed ponds must be developed. It isknown that sustainable agricultural systems arebased on managing soils according to theircapabilities and environmental constraints.According to this, GIS information could allowthe landholder to move from regional or districtland suitability recommendations and landmanagement practices to soil-specific

management, thus improving productivity,profitability and sustainability. An economiccomponent should be incorporated into GISapplications to determine economic suitabilityin addition to physical suitability. This wouldreduce the cost of the rehabilitation of degradedand barren areas and, consequently, facilitatethe adoption of more ecological worldview andless vulnerable of land conservation activities.In addition, to lessen the degree of degradation,the free water areas should be reduced, or atleast consolidated. This could be achievedthrough the improvement of the drainagenetwork system in the waterlogged areas and bypreventing drainage water from overflowinginto lowlands. These actions could protectcropland areas from soil salinization and,consequently, increase agricultural production,especially in order to adapt to future impacts ofclimate change (Doutreloup et al., 2011).

Conclusion

The results demonstrate that Landsat imageclassifications can be used to produce accuratelandscape change maps and statistics. General


13/15

347

patterns and trends of land use change in BinhThuan Province were evaluated by:

classifying the amount of land in sixdistricts area that was converted from

forest, agricultural, and wetland use tourban use during three periods from1990 to 2002;

comparing the results of Landsat-derived statistics;

quantitatively assessing the accuracyof change detection maps; and

analyzing the major land use changepatterns.

In addition to the generation of information tiedto geographic coordinates (i.e., maps), statistics

quantifying the magnitude of change, andfromto information can be readily derived

from the classifications. The results quantifythe land cover change patterns in Binh ThuanProvince and demonstrate the potential ofmultitemporal Landsat data to provide anaccurate, economical means to map and analyzechanges in land cover over time that can beused as inputs to land management and policydecisions. While it is a site specific comparison,it is also useful to compare the Landsat

classification estimates to another, independentinventory such as the Natural ResourcesInventory. Perfect agreement would not beexpected due to the differences in the dates ofdata collection, as well as differences in classesbetween the two surveys.

Acknowledgement

The research is funded by BELSPO (BelgianScience Policy Office) and executed in theframework of a research project entitled

Impact of Global Climate Change andDesertification on the Environment and Societyin Southern Central Vietnam-Case Study inBinh Thuan Province.

REFERENCES

ADAMO S.B. & CREWS-MEYER K.A. (2006)ARIDITY AND DESERTIFICATION: EXPLORINGENVIRONMENTAL HAZARDS IN JCHAL,ARGENTINA. APPLIED GEOGRAPHY VOL. 26,NO.1,PP.61-85.ANDERSON J.R.,HARDY E.E.,ROACH,J.T.&WITMER,R.E.(1976)A LAND USE AND LANDCOVER CLASSIFICATION SYSTEM FOR USE WITH

REMOTE SENSOR DATA. U.S. GEOLOGICALSURVEY PROFESSIONAL PAPER, NO. 964.USGS,WASHINGTON,D.C.27P.Bauer M.E, Sersland C.A., & Steinberg S.J.

(1996) Land-cover classification of the TwinCities metropolitan area with Landsat TM

data. Proceedings, Pecora 13 symposium.August 2022, Sioux Falls, South Dakota :138145.Chen X., Vierling L. and Deering D. (2005)A simple and effective radiometric correctionmethod to improve landscape change detectionacross sensors and across time. RemoteSensing of Environment Vol.98, No.1, pp. 63-79.Coppin, P., Jonckheere, I., Nackaerts, K.,

Muys, B. & Lambin, E. (2004) Digital

change detection methods in ecosystemmonitoring: a review. International Journal ofRemote Sensing, Vol.25, pp.15651596.Doutreloup S., Erpicum M., Fettweis X., &

Ozer P. (2011) Analysis of the past (1970-1999) and future (2046-2065 and 2081-2100)evolutions of precipitation and temperature, inthe Province of Binh Thuan, South EastVietnam, based on IPCC models", paperpresented at the 1st International Conference onEnergy, Environment and Climate Changes, HoChi Minh City, Vietnam, 27th and 28th August2011.Ducourtieux O. & Castella J.-C. (2006)

Land reforms and impact on land use in theuplands of Vietnam and Laos: Environmentalprotection or poverty alleviation? Colloqueinternational Les frontires de la question

foncireAt the frontier of land issues, 17-19mai 2006, Montpellier, France. 22p.ENVI (2006) ENVI 4.4 Help. ITT IndustriesInc., Boulder, CO, USA.Foody G. M. (2002) Status of land cover

classification accuracy assessment. RemoteSensing of Environment, Vol.80, No.1, pp.185-201.


14/15


348

Gao J. & Liu Y. (2010)Determination of landdegradation causes in Tongyu County,Northeast China via land cover changedetection. International Journal of AppliedEarth Observation and Geoinformation, Vol.12,No.1, pp.9-16.General Statistical Office (1994) VietnamCatalogue of the Administrative Divisions,Socialist Republic of Vietnam, StatisticalPublishing House, Hanoi.Gregory G. & Guttman H. (2002)Developing appropriate interventions for rice-fish cultures. In: P. Edwards, D.C. Little, andH. Demaine (Editors), Rural Aquaculture.CABI Publishing, Wallingford, UnitedKingdom. pp.1528.

Imbernon J. (1999) Changes in agriculturalpractice and landscape over a 60-year period inNorth Lampung, Sumatra. Agriculture,Ecosystems and Environment. 76: 61-66.Jensen J.R. (2004)Introductory digital imageprocessing, 3rd Eds. Publisher: PrenticeHall, 544p.Kamusoko C. & Aniya M. (2007) Landuse/cover change and landscape fragmentationanalysis in the Bindura district, Zimbabwe.Land Degradation & Development Vol.18, pp.

221223.Marshall B.K., Picou J.S & Bevc C.A. (2005)Ecological disaster as contextualtransformation: Environmental values in aRenewable Resource Community.Environment and Behavior, Vol.37, No.5,pp.706-728Le K.T., Nguyen T.H., Nguyen D.D., Konda

A., Kajiwara K. & Honda Y. (2002)Automated Classification for Vegetation ofNinh Thuan, Binh Thuan and Lam DongProvinces in Vietnam by Simulated GLI Datafrom Landsat TM. Journal of the JapanSociety of Photogrammetry and RemoteSensing, Vol.41, No.6, pp. 4-13.Le Van D. & Truong, P. (2003). VetiverSystem for Erosion Control on Drainage andIrrigation Channels on Severe Acid Sulfate Soilin Southern Vietnam. Proceedings 3rdInternational Conference on Vetiver grass(ICV3), Quangzhou, China, Oct., 2003.Lu D., Mausel P., Brondizio E. & Moran, E.(2004) Change detection techniques.

International Journal of Remote Sensing,Vol.25, No.12, pp.2365-2407.

Nagendra H., Pareeth S. & Ghate R.(2006)People within parks: forest villages andfragmentation in the Tadoba Andhari TigerReserve, India. Appl. Geog. Vol.26, pp.96112.Ostrom, E. (1999)Self-governance and forestresources. CIFOR Occasional Paper No. 20.Center for International Forestry Research,Bogor, Indonesia.Raquez P. & Lambin E.F. (2007)Conditions for a sustainable land use: casestudy evidence. Journal of Land Use Science,Vol.1, No.2-4, pp.109125.Rogan R., OConnor M. & Horwitz P. (2005)Nowhere to hide: Awareness and perceptionsof environmental change, and their influences

on relationships with place. Journal ofEnvironmental Psychology, Vol.25, pp.147-158.Salam M.A., Ross L.G., & Beveridge C.M.M.

(2003) A comparison of developmentopportunities for crab and shrimp aquaculturein southern Bangladesh, using GIS modeling.Aquaculture Vol.220, pp.477494.Schulz J.J., Cayuela L., Echeverria C., SalasJ. & Rey Benayas J.M. (2010) Monitoringland cover change of the dryland forest

landscape of Central Chile (1975-2008). Applied Geography Vol.30, No.3,pp.436-447.Serra P., Pons X. & Sauri D. (2008) Land-cover and land-use change in a Mediterraneanlandscape: a spatial analysis of driving forcesintegrating biophysical and human factors.Applied Geography, Vol.28, No.3, pp.189-209.Singh A. (1989) Digital change detectiontechniques using remotely sensed data.International Journal of Remote SensingVol.10, pp.9891003.Sivrikaya F., Cakir G., Kadiogullari A.I.,

Kele S., Baskent E.Z. & Terzioglu S. (2007)Evaluating land use/land cover changes andfragmentation in the Camili Forest PlanningUnit of northeastern Turkey from 1972 to2005. Land Degradation and Development 18:383396.Wulder M.A., White J.C., Goward S.N.,

Masek J.G., Irons J.R. & Herold M. (2008)Landsat continuity: Issues and opportunitiesfor land cover monitoring. Remote Sensing of

Environment Vol.112, No.3, pp. 955-969.Wyman M.S. & Stein T.V. (2010).Modeling social and land-use/land-cover


15/15

349

change data to assess drivers of smallholderdeforestation in Belize. Applied Geography,Vol.30, No.3, pp.329-342.Yang X. & Lo C. (2002)Using a time seriesof satellite imagery to detect land use and landcover changes in the Atlanta, Georgiametropolitan area. International Journal ofRemote Sensing Vol.23, No.9, pp.775-1798.Yuan D., Elvidge C.D. & Lunetta, R.S.(1998) Survey of multispectral methods forland cover change analysis. In RemoteSensing Change Detection: EnvironmentalMonitoring Methods and Applications, Editedby R. S. Lunetta; C. D. Elvidge, Ed.; AnnArbor Press: Chelsea, MI: 22-23.Yuan F., Sawaya K.E., Loeffelholz B. &

Bauer M.E. (2005)Land cover classificationand change analysis of the Twin Cities(Minnesota) metropolitan area bymultitemporal Landsat remotesensing. Remote Sensing ofEnvironment Vol.98, No.2, pp.317-328.World Bank (1996) Vietnam WaterResources Sector Review. Main Report, WaterResources Group, Institute of Water ResourcesPlanning, World Bank, Washington D.CZhou W., Troy A. & Grove M. (2008).

Object-based land-cover classification andchange analysis in the Baltimore metropolitanarea using multitemporal high resolutionremote sensing data. Sensors, Vol.8, pp. 16131636.
http://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdfhttp://rsl.gis.umn.edu/Documents/TCMA_change_detection--RSE_paper-3.pdf