Predicting water table depth using Remotely sensed data
Om Prakash Dubey
Department of Civil Engg. University of Roorkee, Roorkee, India.
Sriniwas, A. K. Awasthi
Department of E.sc., University of Roorkee, Roorkee, India.
Abstract
In the present study a simple method for delineating ground water source areas has been presented. Aerial photographs at 1:50,000 scale, Landsat MSS band 5 and 7 blow up at 1:250,000 scale, Landsat CCT and ground truth has been utilized for preparation of base map land cover map, slope map, grain size map and water table map. Multi-vitiate optimization technique in form of PCA considering Rainfall, elevation above mean sea level ground slope, vegetal cover , drainage density and texture of surface material has been carried out. It has been found that PCI accounts for about 928 of total variation. Bivariate plot of PCI and depth of water table clearly brings out clusters for different land uses. This bivariate plot can be utilized for predicting ground water table.
Introduction
If today we are suffering from problems associated with provision of suitable supply of oil at a price we can afford. We are moving towards a state when we shall be suffering from comparable problems associated with provision of water. Several surface and subsurface methods are available for exploration of ground water resources. These methods require large data set making them very costly and time consuming. For optimal exploration of ground water, it is essential to delineate promising areas.
Keeping above in view a study has been carried out in a part of Genetics plain to develop a simple methodology for delineating the promising areas.
Development of Methodology
It is a established fact that ground water availability depends upon several factors viz. Rainfall, elevation above mean sea level, vegetal cover, drainage density, texture of surface material, ground slope etc. The techniques commonly adopted for the analysis of multivariate data are cluster analysis, discriminate analysis, multivariate regression and principal component analysis (PCA) etc. (Dunteman, 1984).
PCA involves the selection of a set of weights to the variance. The problem is to find
b' = [b
11, b
12, ......... b
ip] such that the variance of b', x= b
11 x
1+b
12x+............maximized subject to the constraints that
The problem can be stated as: y= b' v b-(b'b-1)
Where y is the function to be maximized, V is the pxp covariance (correlation) matrix of the original variable, and b' vb is the variance of the composite to be maximized.
l is largance multiplier. b'b-1 =10.
The next step is to find b' and
l such that the y is maximised.
(¶y/¶b) = 2vb-2 l b =0
or, (v-lI)b= 0; since b¹0;
The matrix V-
lI must be singular (V-
lI) b=0; If we premultiply by b' b' (V-
lI) b=0
or,
b'Vb = l b'b
or,
b'Vb = l
i.e. we want to maximize b'Vb=
l (that is the variance of the composite)
