Results on the basis of Chapter 7
8 Results on the basis of Chapter 7
In this chapter, a closer look will be taken at the accuracy which can be achieved in the calculations, when taking into consideration the lessons learned in Chapter 7.
Even if quite a few gross errors are found in an automatic generation, for instance 30 %, it still means that 70% of the measuring has gone well. This implies that the method of automatic generation of elevations from digital images is still a relevant method for use in the Danish landscape. The question is, how the gross errors can be located. The experience learned from Chapter 7 shows that a part of the gross errors is due to the landscape type over which the automatic generation is done and that, in some of the calcula-tions, there were not enough pixels per mesh. In this chapter, these experiences will be taken into con-sideration.
The first investigation is into which accuracy can be achieved by using the minimum mesh size, and ex-cluding those landscape types which contribute to the greatest number of gross errors. Then it is investi-gated whether there are still gross errors in the remaining landscape types.
8.1.1 Results without gross errors
The data material consists of the landscape types flat and hilly terrain for the images in scale 1:5,000. For the images in scale 1:15,000 and 1:25,000, the data material consists of the landscape types flat and hilly terrain and village. As the same landscape types are not included in all three scales, the results are pre-sented in two tables. The iterative process described in section 6.3 is done to isolate the remaining gross errors. The results are shown in table 8.1 for scale 1:5,000, and in table 8.2 for scales 1.15,000 and 1:25,000 in metres and ‰ of the flight altitude.
Resolutionµm Mesh size Accuracy ‰ of H for 1:5,000
15µm
5 x 5 m 0.08 m 0.11
12.5 x 12.5 m 0.10 m 0.13
25 x 25 m 0.16 m 0.21
30µm
5 x 5 m 0.10 m 0.13
12.5 x 12.5 m 0.13 m 0.17
25 x 25 m 0.18 m 0.24
60µm
5 x 5 m 0.17 m 0.23
12.5 x 12.5m 0.17 m 0.23
25 x 25 m 0.24 m 0.32
Table 8.1: The accuracy without gross errors for scale 1:5,000, the three resolutions and mesh sizes. The landscape types included are flat and hilly terrain.
Resolution Mesh size Accuracy for 1:15,0000
‰ of H for 1:15,000
Accuracy for 1:25,000
‰ of H for 1:25,000 15µm
5 x 5 m 0.29 m 0.13 m Should not be calculated!
12.5 x 12.5 m 0.27 m 0.12 m 0.42 m 0.11 m
25 x 25 m 0.30 m 0.13 m 0.47 m 0.12 m
30µm
5 x 5 m Should not be calculated! Cannot be calculated!
12.5 x 12.5 m 0.41 m 0.18 m 0.59 m 0.16 m
25 x 25 m 0.46 m 0.21 m 0.60 m 0.16 m
60µm
5 x 5 m Cannot be calculated! Cannot be calculated!
12.5 x 12.5 m Should not be calculated! Should not be calculated!
25 x 25 m 0.62 m 0.27 m 0.75 m 0.20 m
Table 8.2: The accuracy without gross errors for scale 1:15,000 and 1:25,000, the three resolutions and mesh sizes. The landscape types included are flat terrain, hilly terrain and village.
Table 8.1 and 8.2 show that the best results are achieved by using 15µm images in all three scales. Here the accuracy for the landscape type open, flat terrain is close to 0.1‰ of the flight altitude, if a small mesh size is used. For images with resolution 30µm, the accuracy is reduced by a small amount and lies be-tween approx. 0.13‰ and approx. 0.25‰ of the flight altitude. For images with resolution 60µm, the accu-racy is again a little reduced and lies between 0.2‰ and approx. 0.3‰ of the flight altitude.
8.1.2 Percentage of eliminated points
After the landscape types gravel pit and woods are no longer part of the data material, and the type vil-lage has been excluded in scale 1:5,000, it is now investigated how many points still have to be elimi-nated during the iteration process. The number of elimielimi-nated points is given as a percentage.
Results on the basis of Chapter 7
Image scale 1 : 5,000
Resolution Mesh sizes Flat terrain Hilly terrain Total
15µm
5 x 5 m 3.25 % 7.69 % 5.95 %
12.5 x 12.5 m 2.53 % 11.44 % 7.99 %
25 x 25 m 0.45 % 11.61 % 7.22 %
30µm
5 x 5 m 3.11 % 7.61 % 5.81 %
12.5 x 12.5 m 0.83 % 11.95 % 7.59 %
25 x 25 m 0.23 % 19.60 % 11.98 %
60µm
5 x 5 m 2.34 % 10.68 % 7.40 %
12.5 x 12.5 m 1.67 % 9.65 % 6.52 %
25 x 25 m 0.45 % 14.16 % 8.77 %
Table 8.3: Number of eliminated points given as a percentage for each landscape type, and for the total number of points in scale 1:5,000.
It will be seen from table 8.3 that very few points have to be eliminated in flat terrain (approx. 0.25% - 3.25 %). In hilly terrain, between approx. 8% and approx. 20% of points have to be eliminated. Of the total number of points, between approx. 6% and approx. 12% are eliminated.
In scale 1:15,000, the landscape types flat and hilly terrain and village are included. The number of elimi-nated points given as a percentage is shown in table 8.4.
Image scale 1 : 15,000
Resolution Mesh sizes Flat terrain Village Hilly terrain Total 15µm
5 x 5 m 11.97 % 19.69 % 18.16 % 16.62 %
12.5 x 12.5 m 12.69 % 16.65 % 16.25 % 15.24 %
25 x 25 m 8.44 % 11.92 % 17.97 % 13.47 %
30µm
5 x 5 m Should not be calculated!
12.5 x 12.5 m 2.04 % 14.01 % 10.91 % 8.90 %
25 x 25 m 1.49 % 8.81 % 11.56 % 7.72 %
60µm
5 x 5 m Cannot be calculated!
12.5 x 12.5 m Should not be calculated!
25 x 25 m 1.24 % 12.79 % 9.50 % 7.74 %
Table 8.4: Number of eliminated points given as a percentage for each landscape type, and for the total number of points in scale 1:15,000.
Table 8.4 shows that the number of eliminated points in flat terrain is strongly reduced by using a lower resolution from approx. 13% to approx. 1 %. In the landscape types village and hilly terrain, between approx. 10% and 20% of the points are eliminated.
The number of eliminated points given as a percentage in scale 1:25,000.
Image scale 1 : 25,000
Resolution Mesh sizes Flat terrain Village Hilly terrain Total 15µm
5 x 5 m Should not be calculated!
12.5 x 12.5 m 64.40 % 71.44 % 82.25 % 73.91 %
25 x 25 m 50.62 % 59.09 % 67.81 % 60.16 %
30µm
5 x 5 m Cannot be calculated!
12.5 x 12.5 m 18.3 % 47.3 % 40.6 % 34.7 %
25 x 25 m 17.02 % 45.64 % 43.29 % 35.15 %
60µm
5 x 5 m Cannot be calculated!
12.5 x 12.5 m Should not be calculated!
Table 8.5 shows that in scale 1:25,000, very many points are still eliminated in the chosen landscape types. With resolution 15µm, from 50% to more than 80% are eliminated. With resolution 30µm, from approx. 20% to approx. 50% are eliminated. With resolution 60µm, from approx. 10% to approx. 35% are eliminated.
8.1.3 Summation
After leaving out the problematic areas and elimination of gross errors, it can be seen from tables 8.1 and 8.2 when compared with the results in Chapter 7, tables 7.4 – 7.6, it can be seen that by excluding the problematic landscape types, an improvement in accuracy of approx. 0.10m to 0.20m can be achieved.
It has turned out in this investigation that there is a very great difference in the number of points it has been necessary to eliminate. Tables 8.3 and 8.4 show that there are still up to 20% gross errors, even if the landscape types which contribute the largest number of gross errors are not included in the investiga-tion. Furthermore, it will be seen that the number of eliminated points vary with the resolution and the mesh size.
Table 8.5 shows that there are still very many gross errors in scale 1.25,000. It should therefore be inves-tigated more closely, why the elevation determination in this scale is so markedly poorer than in the two other scales. Unfortunately this has not been possible as the Match-T programme is no longer available.
The results from scale 1:25,000 will still be included in this project, though with an eye on the circum-stance that the elimination of gross errors has been very extensive, and therefore deviates from the gen-eral tendency.
In scale 1:5,000 in flat terrain, the accuracy without gross errors lies between approx. 0.08 m and 0.24 m, cf. table 8.1. By the elimination process, a maximum of 12% of the points are excluded, cf. table 8.3. This shows that there are few gross errors in this landscape type. This is also valid for the two other scales, though not to such an unambiguous degree. However, it is seen clearly from the tables 8.3 – 8.5 that there are still some gross errors, even if the most problematic landscape types are no longer included in the test material.
Therefore, a method should be found which can locate these gross errors from the existing data material without involving the frame of reference.