WETLAND AND WATER RESOURCE MODELING AND ASSESSMENT: A Watershed Perspective - Chapter 4 pot

4 Poyang Lake Wetland Based on Spectral Library and Spectral Angle Mapping Technology Shuisen Chen, Liangfu Chen, Xiaobo Su, Qinhuo Liu, and Jian Li 4.1 INTRODUCTION Located in the nort

4

Poyang Lake Wetland

Based on Spectral Library and Spectral Angle

Mapping Technology

Shuisen Chen, Liangfu Chen, Xiaobo Su,

Qinhuo Liu, and Jian Li

4.1 INTRODUCTION

Located in the northern Jiangxi Province, the Poyang Lake wetlands have played an important role in controlling floods, providing habitation, purifying toxicants, and adjusting climate This valuable resource has been severely depleted in recent years due to excessive exploitation, resulting in the decline of the wetland’s functions and self-restoration ability (Zhang 2004, Zhu et al 2004) The inaccessibility of the area makes it difficult and expensive to monitor and assess the dynamics of this fresh-water lacustrine system Thus, remote sensing technology has become a necessary and efficient tool for this task (Nepstad et al 1999) Remotely sensed data have been extensively used to map land cover (including wetland areas) for the purpose of envi-ronmental conservation, for example, identifying areas demanding protection and monitoring important habitats (Steininger et al 2001, Turner et al 2003)

Remotely sensed data has previously been used for acquiring information about the environment and resources of the Poyang Lake wetland, including specific char-acteristics analysis of healthy vegetation (Xiaonong et al 2002) and multitemporal analysis of land cover changes However, the wetland mapping in Poyang Lake was not completely automatic (Jian et al 2001, Zhao et al 2003), which hindered quick acquisition of the biological resource information necessary for utilization and pro-tection of the wetland The purpose of this study is: (1) to develop an automatic method for mapping the wetland using Landsat ETM+ or TM images, and (2) to map

the distribution of Carex based on the spectral library and the spectral angle

map-ping approach, and (3) to assess the accuracy of the presented approach

4.2 DATA AND METHODS

4.2.1 DATA

Two Landsat images, TM (July 15, 1989, flood season), and ETM+ (December 10,

1999, dry season), were used in this study The first was used for automatic mapping

of the flooding area and the second for the extraction of beach vegetation information within the Poyang Lake cofferdam Before subsetting images, radiometric calibration was performed to transform the digital number (DN) values into image reflectance

4.2.2 FLOODING AREA EXTRACTION

The flood season image was used to create a mask of two primary classes—water and land The flood plain extent of Poyang Lake during the flood period was used for the automatic mapping of the lake beach vegetation distribution area in the low

water season, limiting the Carex mapping area within the lake beach wetland and

water area The identified flood plain area from the July image was used as a mask

to obtain the lake beach wetland and water area from the December image, that is, our study area From Figures 4.1 and 4.3, it is obvious that the water body and land have different reflectance values This allows the setting of a threshold to distinguish the water body from the land using reflectance values As a comparison, in previous research (Niu and Zhao 2004), only the DN values of Landsat imagery were used, which reduced the precision of wetland vegetation mapping using spectral informa-tion Based on the statistical analysis of reflectance of the water body and land region

on Band 4 (Figures 4.2 and 4.4), a threshold value of 0.12 was applied to distinguish water and nonwater areas on the image of July 15, 1989, because Band 4 (near-infra-red band) provided the best differentiation between water and land Image pixels with a reflectance value less than 0.12 were classified as a water body (including both lake and reservoir waters), and the rest as land It can be seen from Figure 4.4 that the reflectance of lake water is smaller than 0.12

FIGURE 4.1 Location of reference reservoir transect in the lake area

0.20

0.15

0.10

0.05

10 Location of Transect across Reservoir

20 30 40 50 60

TM4

TM3 TM2

10 Location of Transect across Reservoir

0.25

0.05 0.10 0.15 0.20

TM4

TM3 TM2

FIGURE 4.2 Reflectance transect of ETM+ Band 4 across a reservoir

FIGURE 4.3 Location of the lake water reference transect in the lake area

4.2.3 SPECTRAL ANGLE MAPPING(SAM)

One of the most commonly used methods in spectrometry is the comparison of tral angles among different land covers Figure 4.5 illustrates the difference in spec-tral angles among natural materials (lake water, reservoir water, land) spectra and the lake water endmembers, with the analysis results inTable 4.1 Traditionally, the spectral distance and probability-based classification methods do not consider the linear scaling of overall reflectance patterns Spectral angle mapping techniques, on

0.25

0.20

0.15

0.10

0.05

Reflectance TM2

TM3 TM4

Location

FIGURE 4.4 Reflectance transect of ETM+ band 4 across Poyang Lake

25

15 10 5 0 0.105 0.155 0.205 0.255 0.305 0.354 Reflectance of Land Near Lake Water Boundary

2500

1500

1000

500

300

250 200 150 100 50 0

Reflectance of Lake Water

Reflectance of Reservoir Water

C

FIGURE 4.5 Spectral angles of lake water endmembers, lake water, reservoir, and land

the other hand, incorporate linearly scaled reflectance patterns to avoid the misclas-sification of land use and land covers, which are the linearly scaled versions of a particular reflectance pattern

The angle that defines a spectral signature or class does not change, and the vectors forming the angle from the origin delineate and contain all possible positions for the spectra (Sohn and Rebello 2002) These parameters encompass all the possible combi-nations of illumination for the spectra Changes in reflectance due to the effect of illu-mination are still within the class angle (Sohn et al 1999) The fact that the spectra of the same type are approximately linearly scaled versions of one another due to illumi-nation and topographic variations is utilized to achieve accurate classification results (Sohn et al 1999) Because the spectral angle classifier utilizes the shape of the pattern for the clustering and classification of multispectral image data (Sohn and Rebello

2002, Luo and Chen 2002, An et al 2005), the analyst’s ability to relate field informa-tion to spectral characteristics and spectral shape patterns for different land cover and land use types is an important factor for achieving accurate mapping results

In this study, spectral angle–based statistical analysis could better quantify

Carex vegetation than other wetland types It is more convenient and useful than

traditional supervised and unsupervised methods of classification

4.3 RESULTS AND DISCUSSION

The comparison of reflectance range and frequency between different land covers helps to distinguish various water bodies from land in the study area (Figure 4.5) Table 4.1 further explains the reflective difference between water bodies and land The multispectral composite image of water bodies in the study area indicates that the extent of light-colored water represents more turbid lake water (Figure 4.6);

Figure 4.7 depicts the spatial distribution of the reflection for different water bodies

Figure 4.8 is a vector map of water body distribution within the boundary of 14 lake-shore towns in Poyang Lake wetland area Figures 4.8 and 4.9 show the distribution

of dense Carex based on the spectral angles of image pixels with reference to the

Carex endmember The total area of Carex is 166 km2 According to the formula for calculating landscape fragmentation index:

C = ∑Ni/A

TABLE 4.1

Various ranges of different water and nonwater bodies in reflection.

Sample land cover Minimum Maximum Scope of most pixels

Nonwater body 0.1045 0.3650 0.1403–0.3650

(98.7%) 0.1245–0.3650 (99.6%)

where C is the landscape fragmentation index of Carex, ∑Ni stands for the total number

of Carex landscape type polygons, and A is the total area of the Carex landscape.

The fragmentation index of the Carex landscape in the Poyang Lake wetland area is

0.6041 (0 represents landscape that has not been depleted, and 1 represents landscape that has been completely destroyed), which is higher than the Honghu Lake wet-land, the other important wetland in China, with a landscape fragmentation index of 0.4207 (Wang et al 2005) This figure indicates that the landscape in Poyang Lake is influenced by human activities more severely than that of the Honghu Lake wetland The area of Poyang Lake’s water body within the boundary of 14 lakeshore towns

Water Body

N

S

E W

FIGURE 4.6 Multispectral component image of extracted water body in study area (R, G,

B = TM 4, 3, 2) (See color insert after p 162.)

inFigures 4.6 through 4.8 is 4,209 km2 while the central lake water area is 3,340

km2(Figure 4.10) The precision of the dense Carex mapping was validated by the

field of investigation (Figure 4.11) According to the 11 field sites, there were 10 sites

that had growth in dense Carex, with the remaining site having growth of Arte misia

selengensis However, there was Carex growing under the Arte misia selengensis A

high precision of 91% for Carex mapping was achieved.

FIGURE 4.7 Water bodies of different reflection in the study area (See color insert after

p 162.)

4.4 CONCLUSION

The automatic and quick mapping method for discriminating land and water, and different turbidities of water, has proved to be an effective tool for wetland

monitor-ing Specifically, it was possible to accurately detect the dense Carex by the spectral angle mapping approach based on the comparison of the reflection of Carex

end-members and the image pixels The method used to extract the flood plain could also

be used to automap the other wetland types This study shows that multitemporal images with their near-infrared reflection of water bodies and spectral library, are valid choices for automatic and quick wetland classification

E W

N

S

Water Body

FIGURE 4.8 Vector map of Poyang Lake wetland

FIGURE 4.9 Distribution of dense Carex in Poyang Lake (See color insert after p 162.)

ACKNOWLEDGMENTS

This research was supported by the open fund of the Key Lab of Poyang Lake Eco-logical Environment and Resource Development of the Chinese Ministry of Educa-tion housed in Jiangxi Normal University, China’s 863 High-Tech Research Plan Project (No: 2002AA130010), and the 2005 Science and Technology Plan Fund of Jiangxi Province Thanks also to two anonymous reviewers for their helpful com-ments and suggestions

E W

N

S

Dense Carex

FIGURE 4.10 Vector map of Carex distribution focused on the central Poyang Lake

wet-land area

FIGURE 4.11 Sites of field investigation in the study area (See color insert after p 162.)

REFERENCES

An, Bin, Shu-hai Chen, and Wei-dong Yan 2005 Application of SAM algorithm in

multi-spectral image classification [in Chinese] Chinese Journal of Stereology and Image Analysis 11(11):55–61.

Jian, Yongxing, Rendong Li, Jianbo Wang, and Jiakuan Chen 2001 Acta Phytoecologica, Sinica 25(5):581–587.

Lan, Tianwei 2004 Poyang Lake wetland is formally applied to be protected as world’s natu-ral heritage http://news.xinhuanet.com/house/2004-11/04/content_2175744.htm Luo, Yu-xia, and Huan-wei Chen 2002 Comparing degree classification with distance

clas-sification—with salt soil as an example [in Chinese] Remote Sensing for Land and Resources 2:46–48.

Nepstad, D C., A Verissimo, A Alencar, C Nobre, E Lima, P Lefebvre, P Schlesinger,

C Potter, P Moutinho, E Mendoza, M Cochrane, and V Brooks 1999 Large-scale

impoverishment of Amazonian forests by logging and fire Nature 398:505–508.

Niu, Ming-xiang, and Geng-xing Zhao 2004 Study on remote sensing techniques on

wet-land information extracting in Nansihu area Territory and Natural Resources Study

4:51–53

Sohn, Y., E Moran, and F Gurri 1999 Deforestation in north-central Yucatan (1985–1995): Mapping secondary succession of forest and agricultural land use in Sotuta using

the cosine of the angle concept Photogrammetric Engineering and Remote Sensing

65(8):947–958

Steininger, M K., C J Tucker, J R G Townshend, T J Killeen, A Desch, V Bell, and P

Ersts 2001 Tropical deforestation in the Bolivian Amazon Environmental Conserva-tion 28:127–134.

Turner, W., S Spector, N Gardiner, M Fladeland, E Sterling, and M K Steininger 2003

Remote sensing and biodiversity science and conservation Trends in Ecology and Evo-lution 18:306–314.

Wang, Xi, Xianyou Ren, and Fei Xiao 2005 Remote sensing and GIS-based landscape structure analysis of Honghu Lake wetland, www.sdinfo.net.cn/xinxizhuanti/2005/xxzt-78.html Wessels, K J., R S De Fries, J Dempewolf, L O Anderson, A J Hansen, S L Powell, and E F Moran 2004 Mapping regional land cover with MODIS data for biological conservation: Examples from the Greater Yellowstone Ecosystem, USA and Pará State, Brazil.Remote Sensing of Environment 92(1):67–83.

Xiaonong, Zhou, Lin Dandan, Yang Huiming, Chen Honggen, Sun Leping, Yang Guojing, Hong Qingbiao, Leslie Brown, and J B Malone 2002 Use of Landsat TM satellite surveillance data to measure the impact of the 1998 flood on snail intermediate host

dispersal in the lower Yangtze River Basin Acta Tropica 82(2):199–205.

Youngsinn, Sohn, and N Sanjay Rebello 2002 Supervised and unsupervised spectral angle

classifiers Photogrammetric Engineering and Remote Sensing 68(12):1271–1280.

Zhao, Xiaomin, Mengxian Yuan, and Hu Wang 2003 Study on remote sensing investigation

and comprehensive utilization of low-grassland in Poyang Lake Region Acta Agricul-turae Universitatis Jiangxiensis 25(1):84–87.

Zhang, Juntao 2004 Elementary accounting of resources and environment loss in Poyang

Lake wetland [in Chinese] Statistical Research 8:9–12.

Zhu, Lin, Yingwei Zhao, and Liming Liu 2004 Protective utilization and function estimate

of wetlands ecosystem in Poyang Lake [in Chinese] Journal of Soil and Water Con-servation 18(2):196–170.