Le RA 
  
  
  
  
  
66 LA COMPENSATION DES BLOCS DE BANDES, DISCUSSION 
strips. However, even after iterative adjustment 
on all perimeter control the mean square value 
of the relative error is only 2.8 meters. This is 
also sufficiently accurate for plotting at scale 
1 : 50,000. 
Those are the results we obtained, and 1 
would like to say that they are still to some 
extent spoiled by reading errors on the com- 
parator which we could not avoid completely. 
We could eliminate the gross errors but there are 
quite a number of 1/10th of a millimetre which 
represents about 8 metres on the ground. You 
can see it in the discrepancies between strips, 
but we could not always find out. We did not 
have time to check which points were wrong and 
which were not, so we had to use them all, and 
also some of the residuals on the ground control 
points. 
Dr H. G. JERIE: I should only like to make 
a short remark, to Professor Fórstner. I do not 
think the accuracy of the blocks is really as bad 
as is mentioned, as for instance adjustment in 
the British Ordnance Survey gave a mean square 
error of the co-ordinates of adjusted points of 
about 15 to 25 microns in the picture scale, and 
this comes down nearly to the measuring accu- 
racy in the single model. I do not think any 
block adjustment or any adjustment can bring 
these errors down much more. 
Another thing, I do not believe completely 
that errors in block adjustment are proportional 
to the size of the block. This consideration is 
only valid for strips, that we get larger errors if 
the strip becomes larger. But it has been our 
experience that at least in a certain range — that 
means up to about twelve strips, for instance — 
the absolute error obtained after block adjust- 
ment is nearly constant. 
Mr J. A. WEIGHTMAN: There is just one 
point which I have never been quite certain of 
in this question of adjustment of higher than 
second order. If one has a curve of order higher 
than second, one has cusps and nodes, and that 
sort of thing. Is there any danger that when one 
has an adjustment of order higher than second 
one will have the singular points occurring be- 
tween the points which one uses to make the 
adjustment, so giving trouble? 
Mr A. J. McNaiR: It was precisely this point 
which I was indicating in my comments, that 
with larger blocks and especially with higher- 
order adjustments you may have computing 
difficulties in reaching a maximum or minimum 
condition. 
Mr A. L. NOWICKI: In connection with 
this last question about the order of adjustment 
curves. We have made some tests on the deter- 
mination of the order of equation to use for 
vertical adjustment of specific flights and spe- 
cific control flights, at right angles to the flight 
lines: that is, the number of lines of control 
along a flight line. We have come to the conclu- 
sion after our many years of experience with 
aerial triangulation that the order of the equa- 
tions to be used is a function of the number of 
flight lines and the number of control lines 
perpendicular to those flight lines. Our results 
show that somewhere between five and six lines 
of control seem to be the maximum as far as the 
degree of equation is concerned. We have found 
that in our simultaneous solutions, if the fifth 
or sixth order curve for the vertical seems to 
be most optimum. This is based on very many 
tests. 
In the case of horizontal control and block 
adjustment, I might just add that we are adjust- 
ing on the system very similar to Dr Jerie's 
method, except that we are using the mathe- 
matical adjustment rather than the mechanical, 
and we find that as our computing machines 
become more complex and larger we shall even- 
tually be able to solve these simultaneously 
rather than have to break up the adjustment 
into parts as we now have to do. 
Prof Dr W. SCHERMERHORN: I have only a 
few remarks. On the question of this order of 
correction curve, may I say quite bluntly that it 
is nonsense. What happens — and it has now 
been proved several times — is that by the 
propagation of errors, purely random errors, in 
a strip you can get everything. You get jumps. 
Mr Schlund, for the first time in my life I hear 
that you also get jumps. Now you can correct 
them. Mr Schut who, in another period of his 
life, did quite a lot of this type of work in Delft, 
now says that he will do away with those things 
— he did not express it quite in that way — which 
I did in the past with my thumbs, I do it now 
with a third, or if necessary a fourth, order 
curve. Mr Nowicki says we go to five cross strips 
and have five lines in a strip. Naturally, if you 
go to the order of nine or ten you can always get 
something which covers the strip as well as pos- 
sible. Let us forget this business, it is nothing 
else but trying to explain away the contradic- 
tions which you have in the strip caused by 
propagation of random errors, which is as old 
as aerial triangulation is. We are now, after 
thirty years of aerial triangulation being in the 
world, still discussing how you can get rid of 
son 
eve 
ual 
fun 
the 
anc