vegetation cover change analysis from multi-temporal...
Post on 10-Mar-2020
3 Views
Preview:
TRANSCRIPT
Indian Journal of Geo-Marine Sciences Vol. 42 (3), June 2013, pp. 331-342
Vegetation cover change analysis from multi-temporal satellite data in Jharkhali
Island, Sundarbans, India
1Sudip Manna*, 2Partho Protim Mondal, 3Anirban Mukhopadhyay, 4Anirban Akhand, 5Sugata Hazra & 6Debashis Mitra
1, 3, 4 & 5: School of Oceanographic Studies, Jadavpur University, Jadavpur, Kolkata-700 032, India. 2 & 6: Indian Institute of Remote Sensing, (Indian Space Research Organization),
Government of India, 4 Kalidas Road, Dehradun-248001, India.
[E-mail: sudipmarine@gmail.com , sudipmanna.ju@gmail.com]
(Received 12 September 2011; revised 25 April 2012)
Present study intends to quantify change of natural vegetation cover (mainly of mangrove forest) in Sundarbans Island between the time span of 2004-2010, when sustained efforts of a forestation and conservation has been in vogue. Vegetation indices like Normalized Difference Vegetation Index (NDVI), Global Environmental Monitoring Index (GEMI), Optimized
Soil Adjusted Vegetation Index (OSAVI) and Transformed Difference Vegetation Index (TDVI) have been used to decipher the measure of vegetation cover in this island and its changes during the period. Radiometric normalization technique is used to nullify various imaging condition anomalies while comparing multi-temporal data for change detection analysis. TDVI has been found to be more effective in vegetation cover change detection in such deltaic environment. Present study shows an overall net increase of vegetation cover in the island as a result of sustained conservation and plantation efforts.
[Key words: Vegetation, Conservation, Sundarbans, Mangrove, Spatiotemporal]
Introduction
Jharkhali being part of Sundarban group of islands (Delta of Ganges) is undergoing continuous changes.
Indian part of Sundarbans, the largest Mangrove
Forest on earth with an area of 9,630 km2, lies
between 21°32′–22°40′ N and 88°05′–89°00′ E.1 It
hosts a wide and diverse range of flora and fauna. The
Island is dominated by mangroves at periphery and also in the creeks with regular tidal influxes; mostly
comprised of Aveccenia sp., Aegiceras sp., Aegialites
sp., Bruguiera sp. etc. Along with mangroves, various
back mangroves and xerophytes e.g. Exoecaria sp., Thespesia populnia etc. cover the vegetated area.
Jharkhali island, previously part of the Namkhana
Reserve Forest, is presently habited by fishermen and farmers. A strong population of 1, 28,802
2 in the year
2001 might have increased to 1, 58,092 by the year
2011 in Jharkhali. Anthropogenic impact caused by incessantly increasing population might caused
around 16 km2
mangrove forest deforestation3, for the
purpose of settlement, agriculture and construction of
bheries (shallow water bodies for brackish water aquaculture). In spite of a significant reduction in land
area (around 86 km2) in Sundarban island system
during the past three decades4 the island Jharkhali
remained quite immune to coastal erosion, being 50 km inland from the southern sea front and
protected from all sides by other land masses.
Sustained conservation effort was taken at both government and community level in the form of
wide-spread plantation, reclamation of land by
natural succession and sincere effort to stop further degradation. There has been a reverse trend in
Jharkhali Island regarding declining forest cover
during the time period of 2008-2010 where the
mangrove area increased by 265 ha by plantation in and around the island, and in the Indian part
of Sundarbans 1610 ha of mangroves were planted
at 70 locations during 2007-2010.5 Though being
young plantations the increase doesn’t contributes
much to vegetation cover. Also an increase of
16 km2
of mangrove cover in entire West Bengal during 2005–2007 is reported by Forest Survey
of India6.
Our present study is aimed at capturing that
conservation effort in terms of increase in natural vegetation cover through multi-temporal change
detection study. Under the present circumstances
continuous spatiotemporal monitoring of the forest cover becomes a critically important element for
—————— * Corresponding author: Phone and fax no. 033-24146242
INDIAN J. MAR. SCI., 42, NO.3 JUNE 2013
332
sustainable management of such susceptible
ecosystem. Remote sensing can act as an effective
tool here, to facilitate collection of ground truth data and conducting scientific study over such relatively
inaccessible parts of the Sundarban delta. This study
is an effort to assess vegetation cover change with
special emphasis on mangrove vegetation, in the Jharkhali island of Indian Sundarbans, from the year
2004 to 2010.
Materials and Methods
The study was carried out at Jharkhali
(88°38′14.83″E to 88°47′38.74″E and 22°00′18.80″N to 22°12′29.18″N, Fig. 1) an island, part of the
Sundarbans deltaic ecosystem. For studying
the change in vegetation cover, cloud free images
of Landsat 5 TM (thematic mapper) sensor (Path 138/Row 45) dated 4
th November 2004 and
6th February 2010 were used. Satellite data used
have a spatial resolution of 30 meters (120 meters for Thermal band) with spectral resolution of
seven bands (0.45-0.52 µm Blue, 0.52-0.60 µm
Green, 0.63-0.69 µm Red, 0.76-0.90 µm Near IR, and 1.55-1.75 µm Mid-IR, 10.4-12.5 µm Thermal-IR and
2.08-2.35 µm SWIR).
Data Pre-processing
Calibration of raw sensor data to meaningful
physical units (e.g. reflectance) prior to any multi-temporal analysis was strongly recommended by
Duggin and Robinove, (1990)7. Raw images were
first calibrated to radiance and then reflectance with
the help of the satellite and sensor calibration parameters
8. For this study 2004 image was taken
as slave image and 2010 image as master. Both were
georeferenced to UTM projection with WGS 84 datum at zone north 45 and co-registered with
accuracy of 0.5 pixels. Though both the images
were taken by same sensor thus having similar
spectral response, relative radiometric normalization is necessary for multi-temporal change detection
study. Radiometric normalization is a linear first order
data transformation, which is applied to reduce the sensor and annual variability effects between multi
temporal datasets over the same geographic area9.
The 2004 image was radiometrically normalized with reference to the 2010 image following the
method of Jenson (1983)10
. Though a very old
method with certain drawbacks, but it is effective
for coastal, estuarine delta and islands where, the entire geomorphologic system is very dynamic and
permanent features are hard to find. Subsequently
Fig. 1—Study area
MANNA et al.: VEGETATION COVER CHANGE ANALYSIS FROM MULTI-TEMPORAL SATELLITE DATA
333
four vegetation indices (VIs) namely Global
Environmental Monitoring Index (GEMI),11
Normalised Difference Vegetation Index (NDVI),12,13
Transformed Difference Vegetation Index (TDVI),
14
and Optimized Soil Adjusted Vegetation Index
(OSAVI)15
were computed for assessment of
vegetation cover and various other land cover classes. All the indices used for the study were
developed based on differential spectral responses off
vegetation over Red and near Infrared spectrum. Individually they have different supremacy over
each other at different percentage vegetation cover
and soil types. GEMI reduces the effect of
atmospheric interferences and retains the information about vegetation cover with fewer or no anomalies
11.
NDVI is the most widely used VI for vegetation
cover and biophysical parameter analysis.16
OSAVI is having benefit that its formulation is simple
with no initial knowledge of soil type required and
the soil background variation is annulated17
whereas, TDVI performs better than NDVI and shows a linear
relation with vegetation cover percentage at high
canopy density.14
The mathematical formulation for them
Classification and accuracy assessment
Unsupervised classification was applied on the
vegetation indices images of both the years to
produce multiple user-defined classes. Initial
clustering was done using ISODATA (Iterative Self
Organizing Data Analysis Technique) algorithm to clump the images into 100 classes, which were
further generalized into 4 classes viz. (A) natural and
planted vegetation, (B) water, (C) mudflats, low-
lying area & inter-tidal zones and (D) settlement, agricultural field, aquaculture farms & unclassified
pixels. Classification accuracy was also estimated
using field verification data collected at 50 sites for four land-cover classes (Figure 2a). Application of a
SR normalization10
enabled estimation of the
classification accuracy of the 2004 image using 2010
ground data. An error matrix was formed calculating user’s and producer’s accuracy with an overall
accuracy for each of the four vegetation index
images for both years. Based on the classified maps, change detection matrix was generated to quantify
the change in natural vegetation cover within the
study period.
Results
Different vegetation indices (VI) calculated from the reflectance value showed different sensitivity in
demarcation of natural vegetation from other
land-use land-cover classes. The classified VI images are shown in Figure 3 and Figure 4 for year 2004
and 2010 respectively. Table 1 summarizes the
Fig. 2—a: Ground verification locations; b: Vegetation cover change map
INDIAN J. MAR. SCI., 42, NO.3 JUNE 2013
334
Fig. 3—Classified vegetation Index images for year 2004
MANNA et al.: VEGETATION COVER CHANGE ANALYSIS FROM MULTI-TEMPORAL SATELLITE DATA
335
Fig. 4—Classfied vegetation Index images for year 2010
INDIAN J. MAR. SCI., 42, NO.3 JUNE 2013
336
Table 1—Classified image based land cover composition for the
years 2004 and 2010
Class Indices 2004 AREA in ha
2010 AREA in ha
Vegetation GEMI NDVI
OSAVI TDVI
3523.05 2881.08
3287.97 3146.31
3648.78 3095.37
3393.45 3274.11
Water
GEMI NDVI
OSAVI TDVI
1481.31 1486.26 1474.02 1539.27
1281.15 1113.03 1115.28 1202.94
MUD flat, low lying
area, intertidal zones.
GEMI
NDVI OSAVI TDVI
1145.70
1197.54 1192.86 1887.3
1551.51
1155.42 1529.28 1915.65
Settlement, agricultural field, aquaculture farms and unclassified zones.
GEMI NDVI
OSAVI TDVI
11675.0 12260.2 11870.2 11248.8
11343.6 11787.0 12089.5 11432.3
Table 2—Classification accuracy and Kappa Statistics
Vegetation indices
Overall classification accuracy
Overall Kappa Statistics KHAT (K^)
GEMI 86.00 % 0.8108
NDVI 80.00 % 0.7247
OSAVI 84.00 % 0.7827
TDVI 88.00 % 0.8373
classification result for the both years. The table
shows that TDVI performs better than NDVI in
extraction of vegetation cover as NDVI tend to saturate at high density canopy cover, a typical
feature forest. The error matrix result is summarized
in Table 2. TDVI also fared superior in terms of
classification results with an overall classification accuracy of 88.00% and kappa statistics of 0.8373.
For the purpose of multi-temporal change detection
post-classification comparison approach was taken which is sometimes referred to as delta-classification.
It involves independently produced spectral
classification results of the time interval of interest, followed by a pixel-by-pixel or segment-by-segment
comparison to detect changes in cover type.
By adequately coding the classification results, a
complete matrix of change is obtained, and change classes can be defined by the analyst. As the two
dates of imagery are classified separately, thereby
minimizing the problem of radiometric calibration between dates.
18 Figure 5 (a), 5 (b), 5 (c) and 5 (d)
shows the results of change detection analysis
for the four VIs: GEMI, NDVI, OSAVI and TDVI
respectively. Change detection matrix for TDVI is given in Table 3. Though different results have been
obtained using different indices but net increase in
natural vegetation and plantations compensating
the forest loss by anthropogenic actions over the study period has been confirmed by all. A detailed
analysis of Table 3 shows that wherever vegetation
cover has decreased, it is due by conversion of mangrove forest zones into aquaculture farms and
in marginal areas by water ingression and coastal
erosion. Also in few places, vegetation cover
has decreased due to replacement by agricultural fields and human settlements indicating direct
anthropogenic intervention. Increased vegetation
cover indicates plantation and natural colonization by mangroves reclaiming swampy areas, accreted
mud flats and other sites supporting the conservation
efforts. An increase of 127.80 Ha in vegetation cover is found by change detection (TDVI). Few
areas with changes are marked in Fig. 5. A complete
change dynamic on the Island, depicted by the
vegetation in Table 4.
Discussion
Change detection study of the vegetation cover
in deltaic ecosystem is an important issue for
sustainable development with reference to the present day scenario of climatic change and
global warming. Our study is based on the central
question that whether there is any increase or decrease in vegetation cover in Jharkhali Island
within the study period. The outputs show the
varied sensitivity of different indices in
discriminating vegetation from other features. TDVI proved to be the most accurate method
for physical assessment of vegetation cover in
mangrove ecosystem due to its less sensitivity to saturation over NDVI. Soil background interference
in signature assortment was minimised by OSAVI,
hence the vegetation cover can be separated from others with good accuracy index. However,
GEMI and OSAVI seem to be over-estimating
vegetation coverage over other indices. As OSAVI
is better in delineating crop canopy cover, it might confuse agricultural land with natural mangrove
forest. GEMI tends to saturate at high LAI values
and perform poorly with respect to soil noise at low vegetation covers.
19 Its capability towards
MANNA et al.: VEGETATION COVER CHANGE ANALYSIS FROM MULTI-TEMPORAL SATELLITE DATA
337
Fig. 5a—GEMI based Change detection map
INDIAN J. MAR. SCI., 42, NO.3 JUNE 2013
338
Fig. 5b—NDVI based Change detection map
MANNA et al.: VEGETATION COVER CHANGE ANALYSIS FROM MULTI-TEMPORAL SATELLITE DATA
339
Fig. 5c—OSAVI based Change detection map
INDIAN J. MAR. SCI., 42, NO.3 JUNE 2013
340
Fig. 5d—TDVI based Change detection map
MANNA et al.: VEGETATION COVER CHANGE ANALYSIS FROM MULTI-TEMPORAL SATELLITE DATA
341
Table 3—Change detection matrixof TDVI based classification (2004-2010)
ClassesArea in ha. 2010 2004
Vegetation Water Mudflat, low lying areas, intertidal zones
Settlement, agriculture, aquaculture, unclassified
Vegetation Vegetation 929.7
Erosion and forest cover depletion 0.81
Erosion and forest cover depletion 97.29
Vegetation destruction and depletion. 2118.51
Water Accretion and natural
colonisation. 15.66 +ve
Water 1048.32
Accretion 374.58
Accretion, Anthropogenic
encroachment. 100.71
Mudflat, low lying areas, intertidal zones
Accretion and natural
colonisation. 91.53 +ve
Erosion 136.62
Mudflat, Low Lying Areas,
Intertidal Zones 863.46
Accretion, Anthropogenic encroachment. 795.69
Settlement, agriculture, aquaculture, unclassified
Natural colonisation, Plantation and
conservation. 2237.22 +ve
Unused land, unmanaged and low yielding Aq. farms. 13.86
Erosion, unmanaged and low yielding
Aq. farms. 580.32
Settlement, agriculture, aquaculture, unclassified.
8417.43
Table 4—Change in vegetation cover (2004-20100 derived from different vegetation indices
Vegetation
indices
No Change Vegetation
Area in ha.
Decreased Vegetation
cover Area in ha.
Increased Vegetation
cover Area in ha.
Net Change in Vegetation
cover. Area in ha.
GEMI 1091.34 2431.71 2557.44 + 126.00
NDVI 850.68 2030.40 2244.69 + 214.29
OSAVI 930.60 2297.79 2403.27 + 105.48
TDVI 929.70 2216.61 2344.41 + 127.80
neutralizing the atmospheric interferences is
proved effective in multi-temporal change analysis.
Result of this study shows that overall natural vegetation cover in the island has increased
offsetting the anthropogenic loss due to sustained
conservation efforts. Natural colonization of
mangroves in charlands and naturally accreted areas in and around the island is noticed.
Observations show that a good measure of mudflats
were colonised by mangroves naturally during the course and the process is continuing.
This increase is evident clearly and marked in
Figure 5d. In addition to that, plantation is done
by state forest department. Reduction of the vegetation cover in some parts of the island
during the study period was due to expansion
of aquaculture activities in violation to CRZ regulation and activities related with increasing
population pressure along with sporadic cases of
water ingression/erosion. Multi-vegetation-index approach adopted here for change detection study
of sensitive and dynamic ecosystems is found
to be appropriate as it requires relatively less amount
of field data, and similar assessment can be done on other important but inaccessible parts of
Sundarbans.
Acknowledgement
First author, Mr. Sudip Manna is grateful to
Department of Science and Technology for providing PURSE fellowship to carryout the research work.
References 1 Nandy, S. & Kushwaha, S. P. S., Study on the utility of IRS
1D LISS-III data and the classification techniques for
mapping of Sunderban mangroves, J. Coast. Conserv.,
15(2011) 123–137.
2 Indian Census, 2001, http://www.censusindia.gov.in/ accessed 15 June 2011.
3 Manna, S., Chaudhuri, K., Bhattacharyya, S., & Bhattacharyya, M.,. Dynamics of Sundarban estuarine
ecosystem: eutrophication induced threat to mangroves. Saline systems, 6:8 (2010) 1-16.
4 Hazra, S., Ghosh, T., Das Gupta, R., & Sen, G., Sea level changes in the Sundarbans, Sci. Cu,. 68(9-12): (2002) 309-321.
5 Sundarban Development Board (2010), Programme for regeneration of mangroves in the charlands of Sunarban by
the Social Forestry Division of Sundarban Development Board. Department of Sundarbans affairs. Government of West Bengal. Unpublished data.
6 India State of Forest Report, 2009, Forest survey of India, Ministry of Environment & Forests., Government of India.
7 Duggin, M. J., & Robinove, C. J., Assumptions implicit in
remote sensing data acquisition and analysis. Int. J. Remote.
Sens., 11 (1990) 1669–1694.
INDIAN J. MAR. SCI., 42, NO.3 JUNE 2013
342
8 Chander, G., Markham, B. L., & Helder, D. L., Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sens. Environ. 113(2009) 893–903.
9 Ya’allah, S.M., & Saradijian, M. R., Automatic normalization of satellite images using unchanged pixels within urban areas., Inf. Fusion. 6(2005) 235–241.
10 Jensen, J.R., Urban/Suburban Land Use Analysis. In R.N. Colwell (editor-in-chief), Manual of Remote Sensing, Second Edition, American Society of Photogrammetry, Falls Church, USA, (1983)1571-1666.
11 Pinty, B., & Verstraete, M.M., GEMI: a non-linear index to monitor global vegetation from satellites. Vegetatio. 101(1992) 15-20.
12 Rouse, J.W., Haas, R.H., Shell, J.A. & Deering, D.W., Monitoring vegetation systems in the Great Plains with ERTS. Third earth resources technology satellite-1 symposium.,1(1974) 10-14.
13 Leprieur, C., Kerr, Y. H., Mastorchio, S., & Meunier, J. C. Monitoring vegetation cover across semi-arid regions:
Comparison of remote observations from various scales, Int. J. Remote. Sens., 21(2000) 281−300.
14 Bannari, A., Asalhi, H., & Teillet, P.M., Geoscience and
Remote Sensing Symposium, IGARSS IEEE International,
5(2002) 3053 – 3055.
15 Rondeaux, G., Steven, M., & Baret, F., Optimization of Soil-Adjusted Vegetation Indices. Remote. Sens.Eenviron., 55(1996) 96-106.
16 Jiang, Z., Huete, A.R., Chen. J., Chen, Y., Li, J., Yan, G. &
Zhang, X., Analysis of NDVI and scaled difference vegetation index retrievals of vegetation fraction, Remote.
Sens.Eenviron., 101 (2006) 366–378.
17 Steven, M. D., The Sensitivity of the OSAVI Vegetation Index to Observational Parameters Remote. Sens.Eenviron., 63(1998) 49–60.
18 Coppin, P., Jonckheere, I., Nackaerts, K., Muys, B. & Lambin, E., Digital change detection methods in ecosystem monitoring: a review. Int. J. Remote. Sens., 25(2004) 1565-1596.
19 Qi, J., Kerr., Y., & Chehbouni, A., External factor consideration in vegetation index development. Proceedings
of Physical measurements and signatures in Remote sensing, Int. Soc. Photogramm. Remote. Sens., (1994)723-730.
top related