Tài liệu Science Direct pdf

Status and changes of mangrove forest in Mekong Delta:
Case study in Tra Vinh, Vietnam
Phan Minh Thu
a,
*
, Jacques Populus
b
a
Institute of Oceanography, 01 Cau Da, Nha Trang, Khanh Hoa, Vietnam
b
IFREMER, Centre de Brest, Technopole Brest-Iroise, BP 70, 29280 Plouzane, France
Received 9 August 2006; accepted 10 August 2006
Available online 28 September 2006
Abstract
Because shrimp culture in the Mekong Delta develops rapidly, it has negatively impacted the environment, socio-economics and natural re-
sources. In particular, mangrove forests have been altered by the shrimp culture. The area of mangrove forests in the region has been reduced and
this is seen especially in Tra Vinh province. The results obtained from GIS (Geography Information System) and RS (Remote Sensing) show the
status of mangrove forests in Tra Vinh province in 1965, 1995 (Northeastern part of Tra Vinh Province) and 2001. In 1965, the area of mangrove
forests was 21,221 ha making up 56% of total land-use, while in 2001 it was 12,797 ha making up 37% of total land-use. Also based on GIS
analysis, over the 36 years (1965e2001), the total coverage of mangrove forests have decreased by 50% since 1965. However, the speed of
mangrove forest destruction in the period from 1965 to 1995 was much less than that in the period from 1995 to 2001. The average annual
reduction in mangrove forest coverage in the first period (1965e1995) was 0.2% whereas it was 13.1% in the later period (1995e2001). For
the long time, mangrove deforestation has been caused by war, collection of firewood and clearing for agriculture, and recently, shrimp farming
has significantly contributed rate of mangrove destruction.
Ó 2006 Elsevier Ltd. All rights reserved.
Keywords: mangrove forest; GIS; remote sensing; Mekong; mangrove changes; mangrove management
1. Introduction
Tropical mangrove forest ecosystems play an important
role in coastal zones, not only in the biogeochemical cycle
but also in the economic life of the region through activities
such as aquaculture and fishing. Mangrove forests in the Me-

kong Delta used to cover more than 250,000 ha (Hong and
San, 1993). War, forest fire, collection of fuel wood and other
human activities have resulted in the reduction of the man-
grove forests in the Mekong Delta. Especially, since the end
of the 1990’s, mangrove forests have been cleared for shrimp
farming in many areas (Hong and San, 1993; Hong, 1995;
Hao, 1999).
Despite the many factors that have affected the mangroves of
the Mekong Delta, the most important factor that has contrib-
uted to mangrove destruction is the shrimp culture activities.
The herbicides sprayed by the USA in the war (1962e1971)
destroyed about 104,939 ha, about 36% of the total mangrove
area in South Vietnam (NAS, 1974). Population pressure has
led to an increased need for land for agricultural production.
In addition, environmental degradation and sedimentation
have also negatively affected mangrove forests (Macintosh,
1996; Le and Munekage, 2004).
Earlier studies (Hong, 1995; Macintosh and Zisman, 1995;
Vits and Tack, 1995; Macintosh, 1996; Phuong and Hai, 1998;
Lakshmi and Rajagopalan, 2000; Lin, 2000; Srinath et al.,
2000; Yap, 2000) have demonstrated that mangrove and
shrimp farming have shown a complex relationship. Mangrove
forests serve as nurseries and food-supply base for marine and
brackish water animals. The mangroves also absorb waste
* Corresponding author.
E-mail address: (P.M. Thu).
0272-7714/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ecss.2006.08.007
Estuarine, Coastal and Shelf Science 71 (2007) 98e109
www.elsevier.com/locate/ecss

generated by shrimp farming (Hong, 1995; Macintosh and Zis-
man, 1995; Macintosh, 1996; Lin, 2000; Gautier et al., 2001;
Wosten et al., 2003). Shrimp culture supplies nutrients for
mangrove forests through water and sediment discharge into
mangrove areas. Nevertheless, the high economic returns in
shrimp farming have resulted in thousands of hectares of man-
grove forest being converted to shrimp ponds and the natural
waterways blocks. The pattern of land-use in the Mekong
Delta has been changed significantly over decades, and this
has consequently affected the economic development in the
region.
Mangrove habitat maps have been used for three general
management applications: resource inventory, change detec-
tion and the selection and inventory of aquaculture sites.
The mangrove distribution maps can be made from investiga-
tion in situ or obtained from remote sensing images and GIS
techniques (Aschbacher et al., 1995; Blasco et al., 1998;
Dahdouh-Guebas et al., 2000; Kairo et al., 2002). Images
used for the present study include SPOT XS (Multispectral
mode imagery from Satellite Pourl’ Observation de la Terre),
SPOT XP or SPOT Pan (Panchromatic mode imagery from
SPOT), Landsat TM (Landsat Thematic Mapper), Landsat
MSS (Landsat Multispectral Scanner), MOS-1 MESSR (Mul-
tispectral Electronic Self-Scanning Radiometer carried out on
the Marine Observation Satellite), JERS-1 (Japanese Earth
Resources Satellite), ERS-1 SAR (Synthetic Aperture Radar
carried on the European Remote Sensing Satellite), MK6 (Rus-
sian Multispectral camera carried on the Salyut-7 Satellite),
and KATE-140 (Soviet panchromatic large format camera).
Aerial Photos were also involved. These together were used

to map mangrove habitat with different image processing tech-
niques, including Visual interpretation, Vegetation index
(NDVI e Normalized Difference Vegetation Index, and BI e
Brightness Index), Unsupervised classification, Supervised
classification, Band ratios and Resolution merge between
Landsat TM with SPOT Pan, Leaf area index (LAI) (Lorenzo
et al., 1979; Bina et al., 1980; Untawale et al., 1982; Patterson
and Rehder, 1985; Blasco et al., 1986; Ranganath et al., 1989;
Roy, 1989; Chaudhury, 1990; Dutrieux et al., 1990; Gray et al.,
1990; Vibulsresth et al., 1990; Jensen et al., 1991; Kay et al.,
1991; Populus and Lantieri, 1991; Eong et al., 1992; Gang
and Agatsiva, 1992; Loo et al., 1992; Mohamed et al., 1992;
Palaganas, 1992 e pers. comm; Long and Skewes, 1994; Asch-
bacher et al., 1995; Vits and Tack, 1995; Rasolofoharinoro
et al., 1998; Blasco et al., 1998; Green et al., 1998;
Dahdouh-Guebas et al., 2000; Kairo et al., 2002; Tong et al.,
2004; Kovacs et al., 2005). These processing methods have
been acceptable for application on mangrove habitat maps in
management, including mangrove inventory and mapping,
change detection and management of aquaculture activities
(Blasco et al., 1986; Ranganath et al., 1989; Chaudhury,
1990; Palaganas, 1992 e pers. comm; Long and Skewes,
1994; Vits and Tack, 1995; Rasolofoharinoro et al., 1998;
Green et al., 1998; Tong et al., 2004; Son and Thu, 2005). It
is recognized that SPOT images can be classified for mangrove
forest identification achieving an accuracy of from 81 to 95%
(Palaganas, 1992 e pers. comm; Vits and Tack, 1995).
The present study provides an overview of the mangrove
forest distribution and changes in Tra Vinh province by using
Remote Sensing (RS) and Geographical Information Systems

(GIS).
2. Materials and methods
2.1. Materials
Tra Vinh Province belongs to the Mekong Delta (Fig. 1),
which is situated from 9

31
0
Nto10

04
0
N and from
105

57
0
E to 106

36
0
E. With a total shoreline of 65 km, it
lies between two branches of the Mekong River (Co Chien
River and Bassac River) and flows into the Bien Dong (South
China Sea). The economy in Tra Vinh is based mainly on ag-
riculture and aquaculture. Shrimp farming areas have devel-
oped significantly and the mangrove forest has also changed
accordingly. In 1943 the area of mangrove forest was about
65,000 ha (Hong and San, 1993), however by 1995 it was
hard reduced to 6678 ha (Phuong and Hai, 1998).

Data have been made available to this study from different
sources. There were topographical maps in 1965 from US
Navy maps which were established in 1967 (Scale map:
1:50,000 and UTM: Indian 1960, Zone 48 in Southern), and
remote sensing images e SPOT image on February 04, 1995
with 3 bands and 20 m resolution (however, the 1995 image
only intercepts the northeastern part of the study area) and
SPOT4 image on January 22, 2001 with 4 bands and 10 m res-
olution. These are images in medium resolution. So, they
could help to recognize the distribution of mangrove forests
with the high accuracy (Vits and Tack, 1995).
2.2. Field trips
Four field trips were carried out at 20 stations, September
10e20, 2000; March 14e28, 2001; September 6e23, 2001
and March 2e20, 2002. At each station, one water and one
sediment sample were collected for environmental factors an-
alyzed in every survey, including salinity, the color of water
and turbidity. In these surveys, salinity and turbidity were
measured by YSI multi-parameter, and the color of water
was measured by color scales. In addition, land-use classifica-
tion and the structure of mangrove forests were identified.
Structure, density, height, floristic composition and standing
biomass of mangrove forests were studied, which helped rec-
ognize training areas and to access accuracy ratio after analyz-
ing the results of remote sensing to find out the distribution of
mangrove forests at the studied areas.
2.3. Methodology to identify mangrove forest by GIS
and RS
The processing of the identification of mangrove forests
was carried out step by step as shown in Fig. 2. This process-

ing was implemented by ArcView 3.2 and ENVI 3.4.
99P.M. Thu, J. Populus / Estuarine, Coastal and Shelf Science 71 (2007) 98e109
2.3.1. Image registration
Image registration is a process whereby an image is re-
sampled to conform to another image or topographical map
(e.g. US Navy maps). In this stage, the varying pixel sizes of
the different images were changed into a common map grid
based on a reference image/map. Evenly-distributed GCPs
(Ground Control Points) were selected in the different images
and registered with the reference images/maps. A RMS (Root
Mean Square) error of less than 0.5 pixels was accepted for
the transformation. Resampling is preformed by converting dif-
ferent pixel sizes to the same final image pixel sizes.
2.3.2. Preliminary analysis
After satellite images were geometrically corrected, prelim-
inary analysis methods could be applied for image enhance-
ment, filtering, unsupervised classification and NDVI
computation (Son and Thu, 2005).
For vegetation areas, including mangrove forest, NDVIs al-
lows cataloguing into 3 classes: low, moderate and high den-
sity. Blasco et al. (1986) and Chaudhury (1990) indicated
that the classification of mangrove forests could be identified
by NDVIs. According to Guyot and Gu (1994), NDVI of man-
grove forest was higher than 0.13. Based on values of NDVIs
Fig. 1. Study area (left) and SPOT image (January 22, 2001) displaying false color composite in Tra Vinh study area (right).
100 P.M. Thu, J. Populus / Estuarine, Coastal and Shelf Science 71 (2007) 98e109
(Table 1), the training areas of different mangrove layers were
selected to support for field trips.
2.3.3. Training area selection
The training areas were selected based on prior informa-

tion, including the result of a preliminary analysis, topograph-
ical maps and information gathered during the field trips, and
on the experience gained from the visual image interpretation.
Each parcel was captured from homogeneous areas and en-
coded. Several parcels were selected per code. These training
sites, therefore, were determined by the numbers of groups
that retained to define the spectral space. The spectral
signature for each group was defined by the means of each
band reflectance and their standard deviation. Each training
area was larger than 200 pixels (20,000 m
2
for image with
10 m in resolution and 80,000 m
2
for image with 20 m in
resolution).
2.3.4. Realization of ground data
Comparison between the training areas and the actual dis-
tribution of the themes in the field trips was an essential ele-
ment of any remote sensing work. By the end of this step,
the whole spectral space was split into classes and each class
represented one or several training areas, and each training site
was assigned with a thematic code. The fieldwork also helped
to ascertain training areas.
2.3.5. Supervised classification
For any given theme the pixels of training sites were used to
calculate a mean spectral reference value. A standard Maxi-
mum Likelihood Classification with Bayesian variation was
Geometric correction
Preliminary analysis

Definition of training area
Field trip
Supervised classification
Post-classification
Accuracy assessment
Input of GIS Ground truth
Overlay map in GIS
Digital analysis
Input information
Labeling
Mangrove classification
maps in 1995 and 2001
SPOT image
Topographical map
Mangrove map in
1965
Mangrove changes
Processing
Material or production
Fig. 2. Processing flowchart to map mangrove change by GIS and RS.
Table 1
NDVI value of the training areas
NDVI Mangrove class Code
0.13e0.42 Low density I
0.43e0.71 Moderate density II
0.72e1.00 High density III
101P.M. Thu, J. Populus / Estuarine, Coastal and Shelf Science 71 (2007) 98e109
performed on the image. Some post classification steps such as
checking information and layers were carried out.
2.3.6. GIS database

The contour of the dykes separating the area between ma-
rine and freshwater were digitized from the topographical
maps and imported to the images. Similarly, the networks of
road and estuaries were digitized. The results from this proce-
dure were used to eliminate interpretation/classification errors
thus providing a more accurate stratification of mangrove,
non-mangrove and other land-use areas.
2.3.7. Evaluation of classification results
Results were calculated from the images obtained using
field observations as reference. The accuracy of the classifica-
tion results was evaluated by comparing the geographical data
derived from ground truth. Randomly selected reference pixels
(about 200 pixels) were inspected at the corresponding sites to
verify the classification results derived earlier.
Fig. 3. Mangrove forest in Tra Vinh province from 1965 to 2001. (A) In 1965 of all region with 21,221 ha of mangrove forest; (B) 1995 at the sourthern part with
2596 ha of low density, 3343 ha of moderate density and 1301 ha of high density of mangrove forest; (C) 2001 of all region with with 8666 ha of low density,
2347 ha of moderate density and 1784 ha of high density of mangrove forest; and (D) Comparing of mangrove forest during 1965e2001.
102 P.M. Thu, J. Populus / Estuarine, Coastal and Shelf Science 71 (2007) 98e109