phase stripes, mainly due to glacial movement and melting 
snow surface. Non-glacial regions include water, bare soil and 
vegetation. Due to temporal decorrelation, phase of water area 
shows a similar texture feature with glacier, namely, dense noise. 
Bare soil presents obvious interferometric phase fringe, because 
of high coherence. Vegetation regions include forests and 
grasslands. For mountain glaciers are often developed in 
regions with higher elevation and lower temperature, very little 
forest is distributed. The surrounding area is mainly covered by 
grassland and lichen low vegetation. Therefore, in a short time 
interval, these regions are able to maintain high coherence. As 
glacial and non-glacial area have different phase texture, when 
the coherence contrast between the two is not strong, phase 
texture analysis can extract the glacial area. 
2.2 Phase texture analysis based on gray level co- 
occurrence matrix (GLCM) 
In this research, gray level co-occurrence matrix (GLCM) is 
used to analysis the texture feature of interferometric phase. 
GLCM proposed by Haralick is an important space-domain 
method for analyzing image texture features (Haralick, 1973), 
which can reflect the distribution characteristics of gray level. 
GLCM can be expressed as 
P(0,0) P(0,1) P(0, j) P(0,G -1) 
P(,0) T. 
Peg n P(, j) Pü,G-1) 
P(G-1,0) PC. + P(C-LG-D 
(1) 
where C is the maximum gray level, l is the gray level of 
pixel (x, VI is the gray level of pixel with the distance d 
and the direction 0 from the pixel (x,y ), Since each pixel is 
surrounded by eight neighboring pixels, different direction and 
distance will generate different GLCM. Haralick gives 14 
texture features, among them 8 features are commonly used, 
which are mean, variance, homogeneity, contrast, dissimilarity, 
entropy, second moment, and correlation. In these features, the 
mean, variance, contrast and dissimilarity can best distinguish 
glacial areas from non-glacial areas. 3 of them are chosen for 
RGB pseudo-colour composite which is then classified by 
maximum likelihood supervised classification for extracting 
glacial areas. Figure 1 shows the flowchart of the proposed 
algorithm for glacier extraction. 
3. STUDY AREA AND DATASETS 
3.1 Study area 
In this paper, Geladandong, the head of Yangtze River, is 
selected as the study area. Glacial retreat is a big environmental 
issue in this region. The extent of Geladandong is from 33° 
05’ N to 33° 40’ N (latitude) and from 90° 45’ E to 91° 
20' E (longitude), as shown in Figure 2. The altitude ranges 
from 5200m to 6621m. There are 274 modern glaciers 
developed in Geladandong area, covering more than 1000 km?. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
    
  
y SAR data E. 
  
  
  
  
Y eee edre eee n eene enne pent tnn nee nennen 
Interferogram generation * 
Simulated 
* topographic phase 
  
  
Differential phase 
  
W 
  
Phase texture analysis 
  
Y 
  
RGB combination 
  
  
Supervised classifation 
  
  
Glacier area 
  
  
  
Figure l. Flowchart of the proposed algorithm for glacier 
extraction 
3.2 Data sets 
SAR images used in this paper include 2 ERS-2 SAR images 
acquired on August 10 1997 and September 14 1997, and 2 
ENVISAT ASAR images acquired on June 24 2007 and July 29 
2007. The former 2 images form an interferometric pair with 
temporal baseline 35 days and perpendicular baseline 62 meters. 
The latter 2 images generate an interferometric pair with 
perpendicular baseline 59 meters. Table 1 shows their 
interferometric parameters. 
Digital elevation model (SRTM DEM, with a resolution about 
90m) over Geladandong is also collected for removing the 
topographic phase from interfrogram and also for geocoding the 
final extracted glacier areas. Figure 3 shows the SAR amplitude 
image and SRTM DEM. Besides, Landsat TM images acquired 
on May 5 2007 is obtained to validate the extracted result from 
the ENVISAT ASAR images. 
  
  
  
  
  
Master Slave Temporal Perpendicular 
Pair image image Baseline baseline /m 
8 8 /day 
1 19970810 | 19970914 35 62 
2 20070624 | 20070729 35 59 
  
  
  
  
  
Table 1. List of interferometric parameters of SAR images used 
in this paper