International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
  
   
     
    
   
   
  
4.2 Determination of search radius 
Tests were carried out choosing neighbours within a circular 
area of variable radius, keeping the number of neighbouring 
points fixed at 30 with at least 12 within, using selection 
without quadrant, over the whole sector (table 9). 
  
    
  
  
  
   
  
      
  
    
  
  
  
   
  
    
  
   
  
  
  
Method Semiaxis RMS 
0.50 0.025 
IDW 0.10 0.025 
0.05 0.025 
0.01 0.028 
0.5 0,041 
Krieim 0.1 0.041 
gme 0.05 0.042 
0.01 0.021 
  
  
   
   
   
  
  
  
  
     
    
    
   
    
   
    
    
   
   
constant. 
4.3 Choice of optimum exponent 
4.4 Statistical evaluation 
Table 9. RMS according to neighbours' selection radius. 
In the light of results, we can affirm that the method most 
homogeneous to the selection circumference is IDW, knowing 
that varying the radius, the model's mean square error remains 
[t can be due to the fact that all points that take part in the 
interpolation have already been selected within a 5-cm radius; 
thus an increase in radius doesn't alter the interpolation at all. 
Hence the RMS varies when the radius decreases to 1 cm. 
The use of the IDW method implies choosing the optimum 
exponent of the weighting functions. In our case -having 
analysed all the data- the exponent with minimum RMSPE (i.e. 
the optimum) has a value of 1.257; which not only means that it 
is not a linear relationship, but that more importance is given to 
far away points than is this exponent is quadratic. 
Comparing among models using the value of a 10-cm radius 
without quadrants, we obtained (table 10): 
  
   
  
   
  
   
  
  
   
  
  
Method RMS (m) 
IDW 0.025 
RBF 0.046 
Kriging 0.041 
  
  
   
    
  
   
   
     
    
Table 10.. Comparison among the different models. 
4.5 Visual evaluation 
Visual evaluation shows little differences between three 
analized methods. The model using IDW method (figure 11), 
the model with RBF method (figure 13) and the model 
generated by kriging method (figure 8). 
E igure 11; IDW model 
  
Figure 12: RBF model 
  
Internation 
  
  
  
Results hi 
accomplist 
generated | 
accuracy 
different n 
the simple: 
RMSE val 
more soph 
previous I 
method is 
and in the | 
Los result: 
utiles para 
puntos que 
valores de 
aunque los 
el método 
menores v; 
por tanto, 
analizados 
método IL 
selección y 
Aguilar, 
Evaluaciór 
la genera 
Mapping ir 
Felicisimo. 
deteccion 
Photogram 
García-Lec 
experiment 
with autoir