Manfred Sties
valuti f
Radar - Ground urban area 1
number 13753 minimum -5,005 average 0,4523 stan. dev 2,0213
outliers 0 maximum 11,981 median 0,14 variance 4,0857
elevation differences
difference
| area | land cover | mean (m) | std. dev. (m)
1 921 1841 2761 3681 4601 5521 6441 7361 8281 9201 10121 11041 11961 12881
| - A |baldeath | -0.47 0.28
Savion function x ; | B crops 0.66 0.34
| ut | | C forest 21.04 2.16
Á | | overall | bald earth -0.50 0.48
i # | | Table 2: Difference Surface Statistics
ene sis =
Il E i
Figure 3: Resampled rasters: statistics of test area 1 (class
urban’)
with a standard deviation of 2.16 m, which reflects the variability of the canopy even in a relatively homogeneous sub-
area. Similar to this procedure, elevation difference statistics were calculated for 67 sub-areas throughout the entire test
area. The results of the aggregated statistical analysis are also shown in Tab. 2, labelled as 'overall'. In this instance, the
standard deviation is 48 cm compared to the typical noise floor of about 30 cm. This is because, over the large test area,
spatially variable systematic errors contribute to the increased standard deviation.
6 COST ESTIMATION FOR THE DEM GENERATION
It is instructive to compare prices of products derived from the two technologies. From previous experience, the cost (in
Germany) of acquiring a data set of coordinated elevation measurements through a single flight airborne mission of a laser
scanner system amounts to approx. US$ 200.- per km?. The cost figure rises significantly for small areas, i.e., less than
100 km?. This does not include resampling to a raster DTM and final editing. In this case, the cost for large area data
production may double to approx. US$ 400.- per km?. Complex, multi-path flights may increase the cost by additional
factors. The STAR-3i system described here is optimized for large area data acquisition. Prices for DEM data that is
licensed DEM data (i.e. non-proprietary to the customer) are published (www.globalterrain.com) and depending on the
type of product, location and volume discounts, are in the range of US$ 15 to 40 per km?. Proprietary data prices are
typically higher (about factor 2) than the licensed prices, but still considerably less than the standard laser scanner prices
that have been observed to date.
7 CONCLUSIONS
All accuracy evaluations confirmed that the elevation differences between the data sets derived from the interferometric
SAR and the laser scanner are very small for bare soil areas: the mean difference is close to 0, while the standard deviation
ranges from 20 cm to 60 cm which mostly reflects the inevitable noise level. For agricultural test areas covered by crops,
the differences varied to a greater extend, but were still in the one sigma range of 50cm to 100cm. In contrast, the
differences for forested and urban areas were significantly greater, which reflect the fact that the radar produced a DSM,
measuring within the canopy or on top of the buildings, whereas the laser was reporting bald-earth, edited elevations. For
872 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
aj
c
IC
