of 
Manfred Sties 
  
  
  
Evaluation of data: 
elevation differences 
difference 
  
  
  
event 
  
  
+0 
  
| -100 0,0 10,0 
density function 
20,0 30,0 40,0 50,0 
difference (m) 
evaluation of data: 
Radar - Ground All test areas Radar - Ground bare soil 1 
* outhers deleted 
number average 
urban* 1.776 0,9648 
vegetated area * 1.487 -1,4410 
bare soil * 363 -0,3063 
vineyard* 1.311 0,1959 
forest” 14.453 19,1508 
total number 19.390 minimum -3,834 average 14,2600 stan. Dev. 10,2503 number: 84 minimum -0,803 average -0,0599 stan. dev 0,4671 
maximum 41,408 median 17,324 variance 105,0693 outliers 2 maximum 1,213 median -0,125 variance 0,2182 
elevation differences 
  
difference 
  
[EN 
  
  
distribution function 
  
  
| 15 40 05 0,0 0,5 1,0 15 
| density function 
| 
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 
event 
20 
difference (m) 
  
  
  
| 
006 | b op— - A — : 140 | 
| L 120 
[n 
| 080 
| 0,60 
| 040 
| 020 
+ 0,00 
15 20 
  
  
  
10.0 20,0 30,0 40,0 50,0 | 15 
  
  
  
  
  
  
5 
Figure 2: ’Original elevation measurements’: statistics of 
test area 1 (class "agriculture, bare soil’) 
Figure 1: "Original elevation measurements: statistics of 
elevation differences selected for ’SAR - Ground’ 
elevation differences: evident statstical outliers have already been deleted. A positive difference signifies that the elevation 
measured by the SAR system was greater than the elevation measured by the laser scanner. The overall average amounts 
to 414.26 m with a standard deviation of 10.25 m. For the different land cover types included, the respective number 
of point pairs and the average value of the elevation differences are given. The histogramm of the differences conveys 
that the statistical values are dominated by the large number of positive differences in point pairs for the class forested 
areas’. The density function of the elevation differences indicates that a minority of differences (belonging to classes 
‘urban’, "agriculture", "vineyard", and ’water’) accumulates in the 0 m range, whereas a majority of differences (belonging 
to the class "forest') accumulates in the 20 m range. A theoretical gaussian distribution is overlayed as a dotted line. 
A graphical representation of the elevation difference statistics of a single test area of the class "agriculture, bare soil’ is 
shown in Fig. 2. A total of 84 point pairs were selected from the SAR and and the laser 'ground' data sets. The difference 
values cover the range between -0.8 m and «1.2 m, while the average difference is -0.06 m with a standard deviation of 
0.467 m. The density function reveals a better approximation of the theoretical gaussian distribution (dotted line). 
5.2 Karlsruhe Results Comparing Co-located Elevation Rasters in Areas of Pre-selected Land Cover Type 
An overview of all elevation difference statistics for the resampled 4.5 m rasters is presented in Tab. 1. In total, 5 urban test 
areas. 7 forest test areas, 2 water test areas, 8 vineyard test areas, 5 agricultural (bare soil) test areas, and 12 agricultural 
(Vegetation cover) test areas were included in these calculations. The size of each test area is given. Elevation differences 
between the resampled rasters of the SAR and the laser ’ground’ data were calculated for each area: the number of 
differences calculated, the average difference and the standard deviation are shown. For the difference statistics betwen the 
SAR and the laser ’vegetation’ data set, only a few test areas qualified (land cover types ‘urban’, "forest, and " vineyard"). 
In addition. for urban test area 5 the differences have been calculated between the resampled rasters of the SAR and the 
complete laser (integrating 'ground' and ' vegetation") data set. 
In detail. one can contrast the results for the original’ measurements in the agricultural test area I as shown in Fig. 2 with 
the results in Tab. 1. For the agricultural test area 1, the ’original’ measurements show an average difference of -0.06 m 
with a standard deviation of 0.467 m (calculated from 84 events), while - in the raster differences - the average amounts 
to +0.05 m with a standard deviation of 0.493 m (calculated from 2.796 events). For the majority of test areas, the results 
from both assessment methods were found to be consistent. For test area 1 of class ’urban’, a graphical representation of 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 869