in which: 
  
  
  
  
P1 = a priori weight of observation i 
| = D,/ 
D. = distance of observation i from center point 
D = mean distance 
P P 
i i 
P: P 
é d Rs + + -— Ra 
5 1 3 
Fig. 6. Modification of weights for Fig. 7. Modification of weights for 
the starting least squares the starting least squares 
planimetric iteration step height iteration step 
The effectiveness of the modifications related to medium sized and large gross errors is shown 
in Example 3. With the relatively bad geometry the gross error of three base lengths at point 
10201 would not be locatable in planimetry without the reduction of the weight in the first least 
squares iteration step. 
9. Reinsertion of observations: 
In two cases it is required to reinsert already eliminated observations. Due to falsified 
Ü-approximations it may happen that an observation is wrongly eliminated. After orientation of the 
mode] the residuals of this observation will become small and a reinsertion is advisable. 
Secondly the result of a least squares adjustment differs from the result of an adjustment with 
robust estimators in the range of 1-20. After the final elimination of the small gross errors at 
the end of the robust adjustment some iteration steps with least squares are performed and small 
gross errors just at the limit of localization will tend to the class of random errors in the 
least squares adjustment and also should be reinserted. 
Therefore the weight function (formula 1) is used in the final least squares iteration steps to 
check for reinsertion of eliminated observations. 
+ 
During the whole procedure of adjustment, as soon as the value of F(V,8, ,.Q) becomes bigger than 
the value 0.01, used for elimination, an already eliminated observation will be reinserted in 
order to stabilize the geometry of the block respectively to be closer to a final result of ad- 
justment. 
10. Conclusion: 
The above described procedure of automatic gross errors localization is a specific developement 
for the blockadjustment by the method of independent models and is not transferable to other 
problems without modifications. 
The procedure covers the full range of occuring gross errors from the small ones, just at the 
C 
dig ones with several base ! hs and shows the power of robust 
(D 
=> 
te 
ct