Finally the computation of U(r, s) snnooths the values of W and W 

 into three rows or colunnns and assigns weights of 0.54 to the values given 

 along to the axes and 0.23 to the row or column on either side of the axis. 

 Random fluctuations in W are smoothed out so that U(r, s) is more nearly 

 a constant at every point and equal to 1/800 of the white noise variance. 



The other source of error lies in the possibility of background curvature 

 of the plane of the stereo data. It will be recalled that one set of data was so 

 severely distorted by background curvature that it had to be abandoned. Al- 

 though no curvature is detectable in figures 11.4 and 11.5, a very slight 

 amount of curvature would produce high values for the spectral estimates near 

 the origin. 



The effect of pure white noise can be estimated from the information 



given in Part 7. The accuracy of the spot height readings is considered to be 



±0.5 feet. Under the assumption that the errors are normally distributed this 



can be interpreted to mean that 



(11.5) P(-0.5 <Ht - Hq < 0.5) = 0.5 



which can be read that the probability is one half that the difference between the 



true height and the observed height lies between -0.5 and +0.5 feet. 



This implies that 



0..5 



'^ dx= 0.25 



cr = 0.54 (approximately) 

 168 



