residuals is not given because it looks similar to figure 6. The root 
mean square errors are then 
us Hy A Puy 
perimeter 39.29 10.85 18.94 28.82 
inside 36.64 6.82 13.08 26.35 
The results after self-calibration with by and b, are 
M Hy A Hey 
perimeter 39.57 10.57 19.05 28. 96 
inside 36,79 6.77 13.67 26.45 
These results lead to conclusions which are contrary to those described 
in the preceding section, i.e., self-calibration does not improve the 
results of the block adjustment at all in this case. The values of Hy 
and u, even appear to increase when more parameters are used. 
5. CONCLUSIONS 
If we compare the conclusions of the two preceding sections, we see a 
clear contradiction. The first conclusion would be drawn in a practical 
situation because then only the information given by 9. and the self- 
calibration parameters and their test values are available. We see that 
distortions of the ground control network can lead to significant values 
of the parameters of the Ebner set. Use of self-calibration in the block 
adjustment will lead to a lower value of the estimator 8 We also see, 
however, that a, gives hardly any information about the quality of the 
computed block coordinates in this case, and that self-calibration does 
not necessarily compensate the effect of a distorted network. In this 
example, the quality of the final coordinates in fact decreases, in the 
sense that the values of Hy and Hy increase with self-calibration. 
As a result of all these considerations, we may conclude that now, after 
a period of some 10 years during which the flow of papers on self- 
calibration has nearly dried up, we still hardly understand how self- 
calibration works. A proper analysis as proposed in /10/ has never been 
made, so that even after all the promising experiments done in the past 
we still cannot claim that self calibration is a reliable method. This 
is especially true when the block is connected to sparse ground control, 
as had been advocated in the past /1/, /9/. Especially if sparse ground 
control is used to reduce the costs of land surveying fieldwork, we get 
weak networks in which errors easily pass undetected. There is then a 
great risk that problems will arise as outlined in this paper. Thus 
stronger networks are required as advocated in /13/, with observations 
(traverses) through the center of the block. In such networks, there is 
less risk of undetected errors and deformations. With denser networks, 
obviously there would be no need to work with sparse ground control. It 
is very likely that the need for the use of self-calibration will be 
reduced if blocks are better controlled. 
"145 -