Digital analysis of satellite data in water resourses studies

Digital analysis of satellite data in water resourses studies

A.K. Chakraborti
Water Resourses Division
Indian Institute of Remote sensing
Dehradun 248001, India

B. Ganbaatar
State Committee for Environment Control
UlaanBaatar, Mongolia

Abstract
Digital analysis of satellite data is studied from the point of view of selecting optimum band (s) and or band combination (s) through several enhancement techniques for water in two states , namely , snow / glacier ice and surface water in a reservoir . study areas are : Gangotri glacier in central himalaya and pong reservoir in Himachal Pradesh , in India . It is seen that , amongst several individual bands & their FCC combinations of 7 bands TM CCT data , FCC (456) provides best information on glacier ice / snow cover condition and secondly , FCC of certain transformed functions ( logarithmic , exponential , principal component ) of IRS-1A LISS-ICCT data provide qualitative aspect of water-depth donation in the reservoir .

1. Introduction
Remote Sensing data have been utilised for water resources studies over the last two-decades . In these studies , both image interpretation techniques and digital analysis methods are employed . But yet there is not enough comprehensive documentation on methodology and analytical tool which should minimise human effort , computer time & cost while dealing with remote sensing data to solve a typical water resource problem .

The utility of digital analysis of satellite remote sensing data in water recourses studies is now well recognised . Because , this mathod is fast , produces , accurate mapping , and can depict subtile changes in water & land regimes vital for water recourses monitoring & decision making . However , with so many new types of satellites / sensors now available , it is not very definitely known to a water recourses investigator , which spectral band or combination or bands are most useful , which enhancement or classification procedure are to be applied, what level of accuracy one can be discerned while addressing a particular water recourse problem ?

Water in two states , snow / ice field and surface water body are of absorbing interest since the early days of satellite remote sensing , because of their unique spectral characteristics in the entire range of electromagnetic wavelength regions , temporal variation within day / season opportunity to capture synoptic view of the entire water body/ice mass and easy way of digital manipulation of satellite data , compared to other complex terrain features . Over 600 , 000 sq . km . area in the Himalayas is snow-bond , situated at 2500 metre above mean sea level . More then 900 large & medium size reservoirs with a storage capacity of 17 million hectare metre have been built in India . Therefore , studies of water in these two states are important .

Two independent studies in Indian situations with limited objective of using digital analysis of satellite data to select . optium band / band combinations are presented in this paper : i) glacier ice / snow cover identification in gangotri glacier in Himalayas ii) water-depth zonation in pong reservoir in himachal pradesh . At this stage , no effort is made to collate digital data analysis & finding with ground-truth information .

2. Study-I : Glacier Ice / Snow Cover Identification

2.1 Objective
Snow in its various forms like fresh snow , aged snow and with its properties like crystal size , density , temperature , thickness , snow cover area , liquid water content etc . are rapidly variable phenomena both in space and time . Optium satellite / sensor / data processing technique to study these phenomena is a promising research area . Since at-present Landsat TM alone provides spectral response in 7 bands of visible , near-IR , middle-IR & thermal IR , thus promising added information on snow , single objective taken up in this study is to find out optium single band image /3-band combination false colour composite (FCC) through digital manipulation of TM CCT data to maximize glacier ice / snow cover identification .

2.2 Study Area & Satellite Data
Southern slope of the gangotri glacier in central himalaya is the locale for this study . Ganga , the most important river in India , originates in this glacier . It is Known that during the month of march , maximum snow accumulation takes place in this glaciated region . A data window of 512 x 512 (pixel : 1-511 , scanline : 2000-2511) of Landsat TM CCT data of path-row 146-039 of 03 March 1987 is , therefore , selected for this purpose .

2.3 Analysis technique
CCT data are analysed in IMAVISION image processing system with EASI / PACE software . analysis scheme adopted in this study is shown in Figure 1. To find out optimum single band image which provides maximum information on snow , histogram of windowed data set in all bands are generated. linear stretched enhancements is performed for single band data . To select which 3-band combination will be useful to emphasize snow classification , an algorithm , called , optium Index Factor (OIF) is utilised (Ref . 1):

Sk = standard deviation of band k

Rj = absolute value of Correlation coefficient between any two of three bands data being evaluated .

After arriving at optimum 3-band combination to create a color composite , the scene is contrast enhanced with linear stretching and studied for glacier ice /snow cover identification .

Figure 1. Selection of Optimum Band / Band Combination for Snow / Glacier Ice Classification
2.4 Result
Histograms of single band date in all the 7 bands go windowed data set show that response of snow & glaciated ice in band 1, 2, 3, & 4 are all –together saturated and do not show any significant frequency distribution whereas in band 5,6, & 7, there is significant spectral response as depicted by the histograms, thereby indicating that band 5,6 & 7 are significant bands for snow cover classification. Correlation matrix of the 4 bands TM data is presented in Table-1.

Table 1 : Class correlation Matrix of 4 Tm Band Data For Gangotri Glacier

	B4	B5	B6	B7
B4	1.000
B5	0.715	1.000
B6	- 0.091	0.041	1.000
B7	0.686	0.967	0.017	1.000

Table-1 shows that band 5 data is having very high correlation with band 7 data indicating that one of then is spectrally redundant . Also band 5 & 7 data offer low correlation with band 6 data . Also band 5 & 7 data is highly correlated with band 4 data . Negative correlation exists between band 6 & 4 . Therefore , band 4 ,5 & 6 combination should be optium combination for snow & ice classes discrimination . This is also confirmed by the OIF value for 456 as the highest of the three 3-band combination sets examined in this study (Table-2) .

Table 2 : Standard Deviation , Correlation Coefficient & optimum Index Factor of Landsat TM data set for gangotri Glacier

SK	RJ	OIF
S4= 73.804	R45 = 0.715	OIF456 = 124
S5 = 18.952	R46 = 0.091	OIF457 = 43
S6 = 12.166	R47 = 0.686	OIF458 = 40
S7 = 9.631	R56 = 0.041
	R57 = 0.967
	R58 = 0.017

Colour composites of different band combinations as well as single band enhanced image present some interesting results, when further visually interpreted. While standard FCC 432 dies not give much information a bout snow cover classification, situation dramatically change, if other combinations 456 or 457 are studied (Table-3) . Whereas, identification of features in single band data is not very encouraging because spectral discriminabilty does not exist for all snow cover features .

Table 3 : Results of Snow Classification in Gangotri Glacier Through Digital Image Processing

3-Band Combination FCC	Feature	colour in FCC image
457	Fresh snow	White
	Glacier snow	Bluish grey
	Melting snow	Bright red
	Transient snow	Greenish blue
	Vegetation	Dark blue
	Mountain shadow	Black
456	Fresh snow	Bright yellow
	Glacier snow	Reddish grey
	Melting snow	megneta
	Transient snow	Greenish blue
	Mountain shadow	Dark blue
576	Fresh snow	Bright yellow
	Glacier snow	Reddish grey
	Transient snow	White
	Vegetation	Magneta
	Mountain shadow	Dark blue
432	Fresh snow	white
	Glacier/snow-melt	Greenish blue
	Mountain shadow	Dark blue

3. STUDY- II : Reservoir Water-Spread Estimation

3.1 Objective
Storage reservoir created on river valley are important water supply source for irrigation , hydro-power generation, industrial and domestic water supply , recreational use etc . Definitive indication of total water volume is, therefore, very important to balance supply & demand schedule by the water resources managers . Remont sensing From satellite altitude provides some measures of this water availability by acquiring a synoptic view of the entire water – spread of the reservoir, its fluctuation in time, and qualitatively evaluate water depth zonation in the reservoir .

3.2 Study Area & Satellite Data
Study site is the pong reservoir on Beas river , located in western Himalaya in Himachal Pradesh in India. The reservoir serve the multipurpose use of water as stated above. IRS-1A LISS-I CCT 4 band data of path-row 031-045 dated 11 February 1989 with . window area of pixel 1280-1792 & scanline 660-1172 is taken up for digital analysis . Spectral response of water is a function of two broad physical variables: i} water depth { shallow/ deep} , ii} water quality { suspended sediment distribution in a reservoir}. Since the scene is four months well past the monsoon period { june-September}, it is assumed that effect of later will be minimal than the former one.

3.3 Analysis Technique
The CCT data is analysed in a super- mini computer based image processing system configured around VAX 11/780 hardware and VIPS 32 software. Reservoir water area is studied mainly from the point of view pre-processing of data by i} band rationing ii} logrithmic function iii} exponential function iv} principal component analysis. Firstly, a pre-processed image is created out of 4 band CCT data manipulation using established algorithm; then, its physical information are discerned through visual interpretation.

3.4 Result
Band Ratioing: Since it is Known that ratio of infrared to visible reflectance provides maximum contrasting information between water bodies & surrounding land eatures, besides eliminating other noise factors, only ratioed image of B4/B2 is created. The histogram f digital number (DN) of ratioed image along a line covering reservoir water and shore area rode 3 distinct class levels: Reservoir water, expose shoe, sandy river bed.

When this data is plotted between brightness vale and frequency (Figure 2), it gives very interesting information. For water, the histogram has 3 peaks which could have been 3 classes of water. Secondly, for other 2 classes (exposed shore and sandy river bed), rationing shows that the data is completely saturated, that is, very little information about these land features is available.

Figure 2. Histogram of DN values of IRS – IA LISS- I Band Ratioed (B4/B2) Data for Water and Land features of Pong Reservior
Logarithmic Function/Exponential Function/Principal Component Function

Table 4: The original 4 bands (B1, B2, B3, B4) IRS-1A LISSS-1 data are transformed into following functions

Logarithmic Function	Exponential Function	Principal Component Function
log10 B1 (DN+1)	eB1 (DN x 0.1)	PC1, PC2, PC3, PC4
log10 B2 (DN+1)	eB2 (DN x 0.1)
log10 B3 (DN+1)	eB3 (DN x 0.1)
log10 B4 (DN+1)	eB4 (DN x 0.1)

When these transformed values are studied individually as well as in FCC combination to derive physical information of reservoir water and shore area into 5 classes : very shallow water , shallow water , deep water , exposed shore , sandy bed of influent streams , following FCC combination are found to be better : FCC (log B4 , log B3 , log B2) ; FCC (eB1 , eB2 , eB3) ; FCC (PC1 , PC2 , PC3) .We are very familiar with standard FCC (B4 , B3 , B2) image . However , for water conditions (shallow/turbid , deep/clear) ,it is observed that , standard FCC image does not provide much information except the water- spread .

4. Conclusion

Digital analysis of 7 bands TM FCC data indicate that amongst several individual bands & their FCC combinations , FCC (456) provides best information on ice/snow cover conditions .
Digital analysis of 4 bands IRS-1A LISS-I CCT data indicate that instead of single band or their transformed function , FCC of some transformed functions, namely , FCC (log B4 ,log B3 , log B2) , FCC (eB1 , eB2 , eB3) , FCC (PC1 , PC2, PC3) may provide qualitative aspect of information about water depth zonation besides water- spread of a reservoir .

5 . Acknowledgement
Authors like to give credit to Mr . D . Vijayan and Mr . M . Kudrat for assistance in using image processing system and to Ms . Sangeeta for nearly typing manuscript of this paper .

6 . Reference

Jensen , J . R . 1986 . “Introductory Digital Image Processing : A Remote sensing Perspective” . Prentice Hall .