Songkhram Habitat Classification

8/14/2019 Songkhram Habitat Classification

1/11

Page 19

Habitat mapping forfisheries using remote sensing and GIS techniques in

Lower Songkhram River Basin

Ubolratana Suntornratana1, Thiha2, Sekson Daungsri3, Edward L. Webb4 and Anupong

Sanitchon5

1 Surin Inland Fisheries Research and Development Centre, 2 & 4 Asian Institute of Technology, 3 Maha Sarakhram

Inland Fisheries Research and Development Centre, 4 Nongkhai Inland Fisheries Research and Development Centre

ABSTRACT

Remote sensing, a technique of investigating the earth from satellite platform with the use of

Electro-Magnetic Radiation (EMR), has a potential to help support fisheries study through

investigating spatial distribution of habitats important forfi

sheries production and managementin integration with GIS technology. The main objective of this study was to map habitats, which

are essential forfisheries production in the lower part of the Songkhram River Basin. The core

methodological steps involved delineating inundated floodplain, which expands during wet

season and estimating reproductive areas forfisheries production and also other type of habitats

important for aquatic production through ancillary GIS analyses.

Two Landsat ETM images from 17 September 2000 and 18 March 2003 were classified into

thematic land cover maps using supervised classification with maximum likelihood method. A

self-developed classification protocol was used to map the habitat across dry and wet seasons

of the study area. Prior to the image classification, standard preparatory image processing oflayer stacking, image co-registration, subsets and cloud removal were performed using ERDAS

image processing software. The orthorectified TM image from 6 November 1992 was used as

the master image for the image co-registration.

A system of land cover classification using satellite data, then, combined with field

observations presented of habitats types. Ground truth data was collected from several

field visits to the study area (Lower Songkhram River Basin) during July to August 2006

using Garmin Map 60 GPS. Field data collection was supplemented with information from

topographic maps at a scale 1:50,000 and field observations by the research team.

Thematic habitat maps from preliminary analysis to serve fisheries field of studies exhibited

ten major habitats in the lower Songkhram River Basin. Permanent water bodies of the main

Songkhram River and its tributaries, other water bodies, inundated flooded areas mostly

associated with riparian mixed vegetation, paddy fields (lowland) and some agricultural crops

in wet season represent the most important habitats essential forfisheries production inside

study area. Inundated flooded extended in the wet season was more than 50,000 hectare. It is

anticipated that this information combined with spatially explicit GIS maps could help fishery

managers to link with catch information to estimate catch production in the lower basin by

habitat types and/or helping present a scenario offish stock for better management. However,

the completion of habit mapping forfisheries would require an accuracy assessment to evaluate


2/11

Proceedings of the 8th Technical Symposium on Mekong Fisheries

Page 20

the correctness of habitats. In addition, revision of classified habitat maps, by integrating ground

truth data, is recommended for further spatial analyses relative to future fishery management.

INTRODUCTION

Low-land floodplain is the important system of wetland and also a sensitive ecosystem which

requires a proper monitoring system to determine the change. This area can produce big amount

of production and generate the big food chain to serve many living organism from the bottom

to top level (Bayley 153 58; Chapman and Chapman 23 30). However, the large variation

is affect by a seasonal and inter-annually condition, especially fisheries sector. The fisheries

production potential in the floodplain river system is directly related to the duration and extent

offloods and consequently led to potential offisheries production.

The Songkhram is very well known an important floodplain ecosystem with its huge area

covers extensive in the lower part of the basin (David Balke and Rattaphon Pitakthepsombat

118 1; Interim Committee for Coordination of Investigations of the Lower Mekong Basin

1 11). There were record of many Mekong species migrate to this river system and utilized

this area as a spawning ground and nursing ground. This caused of a big amount offisheries

production occurred every year during wet season (AMCF 1 34; Sjorslev et al. 5 19;

Suntornratana et al. 29).

Estimated fisheries production roughly in the lower part of the basin which covers around

4,000 km2

was possible more than 16,000 tonnes (AMCF 1 34; Sjorslev et al. 5 19). Theextrapolation on production conducted by introducing a simple interview survey altogether with

broad inventory on communitys fisheries. The used of Geographic Information System (GIS)

was to estimate area of habitat in order to support assessing fisheries production. However, the

suggestion provided was a proper system to estimate habitat area could help better assessing

fisheries production.

This paper, then, presented a methodology used the technique of Remote Sensing (RS) and

Geographic Information System (GIS) to visualize habitat and monitor the surface changed,

especially flood extended area which is important forfisheries production. The RS technique

considered an observation from space though time series which can help monitoring surfacechange seasonally (Butleret al. 1 165; Dekkeret al. 211 25; Ellis and Woitowich 98 109).

The techniques is also useful in assessing the status of the environmental system and long-term

monitoring habitat change to support decision making forfisheries management foe better

sustainable used.


3/11

Page 21

Habitat mapping forfisheries using remote sensing and GIS techniques in Lower Songkhram River Basin

METHODOLOGY

Habitat classification mapping were generated used of Remote Sensing (RS) and Geographic

Information System (GIS) technologies which basically apply a remote sensed data integratedand organized through a computer based systems. The core analytical procedure included two

phases: habitat classification and spatial analysis. Habitat classification phase concerned with

the classification of remote sensing data into thematic habitat maps. Spatial analysis dealt with

the quantification of the seasonal dynamics of majorfishery habitats through a change vector

analysis.

Framework of the study

Overall simple framework for assessing fisheries sector is the information of habitat and

fisheries should be able to integrate and combine for better assessing this sector. The structure

of integrating the different types of information presented as in Figure 1. Spatial implication

used of remote sensed datasets can visualize and monitor various types of habitat, included

the habitat important forfisheries production. This classified habitat, then, can be merged with

a non-geographic information from other surveys and present scenario offisheries situation

and monitor the trend of change. For the main purpose to conduct habitat classification using

Remote Sensing and GIS techniques, the diagram presented process of Land covered analysis.

Framework for integrated remote sensed data with interview survey for assessingFigure 1.

fisheries and process of land cover analysis for habitat classification.

RS and GIS system to

visualize image and

process spatial data

Image processing

Ground reference

Image classification

Spatial anlysis

Remote sensed data for

habitat classification

Spatial/ Geographic

dataset (Landset ETM

(Sep 2000 and Mar 2003)

Thematic maps

provided detail habitat

class with

boundary/area

Non-geographic

dataset on fisheries

Information in

fisheries which

presented in simple

tabular

Statistic technique

for manipulating

result

Scenario of

integrating data-

sets supported

simple assessing

process


4/11


Page 22

Study area

The lower part of Songkhram River Basin (Figure 2) was selected for study as it is very well

known an important lowlandfl

oodplain in the upper northeast region of Thailand. The areaselected for habitat classification was around one-third of the total basin area which covers

more than 12,700 km. It locates between latitude 14 14' and 18 98', longitude 103 75' and

104 48'.

The ecosystem of this area is mainly a wetland consists of special vegetation and the extent

of inundated flooded occurred annually for two to three month during wet season. This natural

event generate huge fisheries production to serve local communities in the area. Information

from various studies showed big amount offisheries production occurred every year and many

reports showed the evidence offisheries production was generate due to large area offloodplain

in this river system (AMCF 1 34; Yingcharen and Virapat; David Balke and RattaphonPitakthepsombat 118 1).

Study area in lower part the Songkhram River Basin, upper northeast region of ThailandFigure 2.

Data selection for habitat classification and data source

Two Landsat ETM+ images from 18 March 2003 and 28 September 2004 were purchased from

Michigan State University. The orthorectified TM image from 6 November 1992 and ETM+

image from 17 September 2000 were downloaded from the Global Land Cover Facility (http://

glcf.umiacs.umd.edu/index.shtml) (Table 1). Topographic maps at a scale of 1: 50,000 covering

H.NamYam

Nam Oon

H. K

ho

ngH.H

i

NamM

ao

1040'0"E

1040'0"E

1030'0"E

1030'0"E

180'0"N180'0"N

170'0"N170'0"N

River

Flooded area

0 20 4010

km

LAO PDR

THAILAND

Songkhram

River Basin

THAILAND

LAO PDR

CAMBODIA

VIET NAM

Gulf of Thailand

NamSongkhram

Mekong


5/11

Page 23


the lower part of Songkhram River Basin were obtained from Technical Advisor Division

of the Mekong River Commission (TSD-MRC). Those were used as reference and training

materials required for image registration, evaluation of habitat reflectance signatures and image

classification

List of satellite images data used for habitat classification and co-registration parametersTable 1.

Satellite and sensor Acquisition date Sun azimuth

and angle

Ground

resolution

RMS

(root MSE)

Source

Landsat 7, ETM + 17 September 2000 118.62, 60.87 30 x 30 m 12.8 m GLCF

Landsat 7, ETM + 18 March 2003 119.68, 55.73 30 x 30 m 13.2 m MSU

Landsat 5, TM 6 November 1992 137, 46 30 x 30 mOrthorectified

(Master) imageGLCF

Landsat 7, ETM+ 26 April 2003 95, 64 30 x 30 m 9.04 m GLCF

Landsat 7, ETM+ 28 September 2004 125.94, 58.60 30 x 30 m 10 m MSU

Image preprocessing

Prior to the image classification, we performed standard preparatory processes of image

geometric correction, image-coregistration, image subsetting and cloud removal. First, all

Landsat images were geo-referenced into Universal Transverse Mercator (UTM) zone 48, using

an image-to-image registration method. Second, geo-referenced images were co-registered to

the orthorectified TM image from 6 November 1992 for their topographic errors due to terrain

effect. Third, the datum of all coregistered images was transformed into Thailand/Vietnamdatum. Data transformation was necessary because all the topographic maps, thematic GIS

layers and Global Positioning System (GPS) used the Thailand/Vietnam datum while all the

Landsat images were projected in World Geodetic System 84 (WGS 84). Finally all images

were subsetted to exclude the areas outside of the area of interest from being considered.

Ground reference data collection

Ground reference data required for the reflectance signature development for the imageclassification was collected from an intensive field work during the period between July and

August 2006. The research teams recorded approximately fifty ground reference points covering

all major habit types, using a GARMIN GPS 60 receiver. In addition, a self-designed field data

collection sheet was used to record the attribute data such as surrounding land cover, human

activities and other environmental variables (e.g. soil erosion). Reflectance signatures for each

habitat type were developed based on ground reference data, training and reference materials as

well as researchers familiarity with the area.


6/11


Page 24

Image classification

Supervised classification with maximum likelihood classifier classified all image pixels of each

Landsat image into nine habitat types (Table 2).

General description of habitat types in the lower Songkhram River BasinTable 2.

Habitat type Description Additional information of habitat

Permanent water body

(Channel-type) *

Main Songkhram River and its tributaries as well

as other small streams

Considered an important spawning and nursing

ground for many riverine species migrated

from the Mekong to the Songkhram and also

the channel for many species to migrate onto

floodplain of the Songkhram (Suntornratana et

al. 270 80).

Permanent water body

(Non-channel type)*

A Lentic water system such as several sub-

categories of natural and man-made water bodies

such as swamp, reservoir, pond, weir, etc.

Considered an important fishing ground which

fishing activities occurred almost all year round.

This area could be a dry-refuge season area for

stock of many fish species in the Songkhram.

Agricultural crops Man-made plantations mainly Rubber (Havea

brasiliensis) andEucalyptus spp., cash crops such

as Sugarcane

The ages of plantation ranged between 4 to > 10

years. This habitat had dense canopy cover and

uniform pattern.

Dipterocarps-

dominated mixed

vegetation

Majority are natural vegetation mixed sometimes

with enrichment plantations in privately owned

patches and community forests. Usually this

vegetation had two layers, with the Dipterocarps

being at the canopy layer

Undergrowths include bushes, herbs and shrubs.

Riparian mixed

vegetation*

Wetland natural vegetation with non-canopy

trees. Majority of vegetation are dwarf and bushy,

and highly associated with bamboos, shrubs and

grasses.

Mainly found on river banks and inundated

lowland areas. This zone also considered a

riparian flooded forest which important as a

spawning and nursing ground area for many fish

species. It was also important fishing habitat

during wet season.

Paddy field * Includes two types of rice fields: wetland rice

fields and upland rice field. Wetland rice fields are

abandoned in the wet season and rice is grown

when water recedes in dry season. In the upland

rice fields, irrigated rice is grown throughout the

year.

Considered an important habitat forfisheries

production, especially for lowland paddy which

normally flooded during .

Barren land Man-made clearings for housing complex and

other infrastructure such as roads and bridges.

Mixed vegetation Majority is mixed vegetation in relativelyhigher ground, rarely mixed with riparian mixed

vegetation. Majority appears natural vegetation,

rarely mixed with plantations, especially private

owned plots and community forests.

Settlement Built-up areas such as highways, feeder roads,

villages and town, rarely mixed with home garden

and road-side vegetation

This category was spectrally highly variable due

to the association with diversified home garden

species and agricultural crops.

* Considered an important habitat potentially generated fisheries production.

Spatial analysis began with performing a neighbourhood analysis. Neighbourhood analysis

with three by three pixel kernel function eliminated the isolated image pixels that were likely


7/11

Page 25


to be misclassified and maintained the spatial cohesiveness among image pixels. Subsequently,

non-channel type and channel type permanent water bodies and inundate flooded area were

extracted from the dry and wet season habitat maps and were vectorized. A GIS overlay

between wet and dry season habitat GIS maps produced a map showing inundated flooded

bodies in the wet season. Area statistics of the extended water body was calculated from theattribute table of the output map.

Spatial analysis

Supervised classification was one of the post classification techniques that have been widely

used in several land cover/use and habitat mapping studies. The method has also proven to be

comparable with other classification methods, such as unsupervised classification and hybrid

method, in terms achieving high classification accuracies.

Result of habitat classifi

cation

Followed the RS and GIS techniques, images have been identified and estimated for area

covered in the lower part of the Songkhram River Basin into different categories of nine major

habitat types. Output from image classification included dry and wet season maps showing

major habitats in the lower part of the basin (Figure 3 and 4).

Thematic mapping classified habitat into nine categories in the lower part of theFigure 3.

Songkhram in dry season


8/11


Page 26

Thematic mapping classified habitat into nine categories included inundated floodedFigure 4.

extended area in the lower part of the Songkhram in wet season.

During peakflooded period in September 2000, the image presented clearly offlood-pulse

extent covered lowland area of the basin. This area determined a large area of habitat essential

forfisheries production which possibly changed every year due to the rate of precipitation, run

off from Songkhram sub-catchments and also the water level in the mainstream Mekong.

Beside major output of two thematic maps showing the habitat types in the lower

Songkhram River Basin, an ancillary statistics: area statistics for each habitat types and

seasonal change was also summarized. The statistics outputs were exported to excel format and

delimited formats for further calculation. Habitat-wise area statistics across two seasons are

given in Table 3. The most important output from the spatial analysis was the map showing the

spatial distribution of inundated water bodies in the wet season. Clearly shown of an extended

inundated flooded was approximately 53,000 hectare.

Even though the different of area in all habitat types could not be compared directly due tothe limitation to get the images in between two seasons the same period offlooded cycle in a

year. This flooded area extent presented could possible determine the area of paddy field which

clearly different between the two seasons for around 36 percent of the total area of paddy field

inside study area in dry season. However, the change of paddy area was under the assumption

of no change over time in two years.

In relevant to the extent flooded area and all water bodies of both channel and non-channel

types, it was considered that the class of Riparian Mixed Vegetation was also the important

habitat forfisheries production, especially during wet season. This area cover found along the


9/11

Page 27


river/stream channel and scatter around an inundated flooded area. Statistic showed more than

ten percent of the total area and seemed to be decrease by time.

Area statistics of nine habitat types in the lower part of the Songkhram River Basin. (*)Table 3.

indicated the extent water body during peakflooded in the wet season

Habitat type Wet season (2000) Dry season (2003) % different

between

season *Area (ha) % Area (ha) %

Dipterocarpus-dominated forest 15,291.4 4.4 14,097.2 4.0

Paddy 91,838.7 26.2 144,226.2 41.1 -36.3

Permanent water body (river and stream) 6,883.0 2.0 2,037.3 0.6 -237.8

Permanent water bodies (non-channel type) 12,232.3 3.5 11,406.1 3.3 -0.1

Riparian mixed vegetation 45,414.1 13.0 39,993.2 11.4

Mixed vegetation 90,858.8 25.9 96,254.1 27.5

Agricultural crops 25,697.3 7.3 18,969.2 5.4

Settlement 10,340.0 3.0 11,408.7 3.3

Barren land 0.0 0.0 722.0 0.2

Inundated flooded extended* 51,803.9 14.8 0.0 0.0

Cloud 11,446.7 3.3

Total area 350,359 100 350,560.6 100

* The different of area changed was under the assumption of land used pattern in the lower part of the Songkhram River Basin

was not significantly change during three years.

Recommendation

The result obtained of this trial study have demonstrated the possibility of habitat classification.

Application of RS and GIS can help visualizing various types of habitat included habitat

essential forfisheries production like an inundated flooded area, riparian mixed vegetation,

paddy and water bodies inside study area. However, the difficult was to encounter for

identification ground control points on the image due to the low resolution.

With the completion of habitat classification and mapping forfisheries field of study,following technical recommendation are made for the improvement of habitat maps for further

spatial analyses.

The classified habitat maps are subject to accuracy assessment to evaluate the correctnessof each habitat type.

Revision of classified maps could be necessary using ground truth data. Ground truth datacollection can be done during accuracy assessment in which randomly selected habitat

mapping places are field-visited.


10/11


Page 28

Additional spatial information should be integrated to rectify classification error. Forexample, existing land use maps, cadastral maps and topographic data are useful to

improve the accuracy.

Required more satellite images of different seasons in the same period of year (fl

ooded cycle) for better comparison of the change over time and seasonal affect of change.

According to the dense of riparian vegetation affected to reflectance, therefore, area offlooded extent underneath of bushy dense could not be clearly detected. Field survey,

then, required in order to be able to indicate flooded extent in this habitat class.

REFERENCES

AMCF. 2003. Baselinefi

sheries the Lower Songkhram Basin; Draft report. Udonthani,Thailand. 34 pp.

Bayley, P. B. 1995. Understanding large river-floodplain ecosystems. Bioscience 45 p.

153 158.

Butler, M. J. A et al. 1988. The Application of Remote Sensing Technology to Marine Fisheries:

an Introductory Manual. Rome: FAO.

Chapman, L. J and C. A Chapman. 1993. Fish populations in tropical floodplain pools: a re-

evaluation of holdens data on the River Sokoto. Ecology of Freshwater Fish 2.2 (1993):p. 23 30.

David Balke and Rattaphon Pitakthepsombat. 2006. Situation Analysis: Lower Songkhram

River Basin, Thailand. Mekong Wetland Biodiversity Conservation and Sustainable Used

Programme (MWBP). Bangkok. 118 pp.

Dekker, A. J et al. 1992. The effect of spectral bandwidth and positioning on the spectral

signature analysis of inland waters. Remote Sensed Environment. p. 211 225.

Ellis, T. J and W. A Woitowich. 1989. An Overview of the Use of Remote Sensing for the Studyof Rivers and River Systems. Canadian Special Publication of Fisheries and Aquatic

Sciences.

Khon Kean University. 1984. Interim Committee for Coordination of Investigations of

the Lower Mekong Basin. Study of Fisheries Impacts of Nam Songkhram Basin

Development. pp 488.


11/11

Page 29


Sjorslev, J. G et al. 2001. Combining Sampling and Census Approaches for Fisheries Habitat

Assessment in Songkhram River Basin, North East Thailand. Phnom Penh, Cambodia:

Mekong River Commission.

Suntornratana, U et al. 2002. Migration Onto the Floodplain of the Songkhram River Basin.Phnom Penh, Cambodia: Mekong River Commission, 2002.

Suntornratana, U et al. 1999. Preliminary Study on Migration and Spawning of Fish in the

Songkhram River, Thailand. Phnom Penh, Cambodia: Mekong River Commission

Secretariat.

Yingcharen and Virapat, C. 1998. Final Report Aquatic Ecology and Fisheries

Surveys in the Songkhram River Basin, Nakorn Phanom Province, Thailand.

Documents

Songkhram Habitat Classification