The International Archives ofthe Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
solar energy. 3D modeling of the aggregated yearly amount of 
solar energy for each point on the roof is a promising way to get 
precise results. 
4.4 Calculating the Solar Potential 
Once the amount of square meters on a roof which are suitable 
for solar exploitation is known, a prognosis on the solar harvest 
of a building can be made. For Cologne, a solar harvest of ca. 
400 kwh/m 2 per year for solar thermal collectors and ca 100 
kwh/m 2 per year for photovoltaic modules can be simulated. 
The results have to be modified by a factor (Peuser et al, 2001) 
resulting from the derived information on slope and aspect of 
each roof segment. 
5. RESULTS & DISCUSSION 
The results of this field test make it clear that simple roof 
structures and high measure point densities generally lead to 
precise results. For the LIDAR data used in this study, after the 
filtering, there remains an average density of 1 measure point 
per 3 m 2 on an orthogonal projected surface. An average point 
density of 1 point per 4 m 2 results for sloped roof patches, 
which means that most of the small roof elements, e.g. dormers, 
cannot be identified. 
Figure 4 shows that the classification according to the 
exposition of the different roof patches can lead to good results, 
even for more complex superstructures, if LIDAR data with a 
sufficient point density can be obtained. This roof for example 
is represented by 1 LIDAR measure point per 3m 2 . However, it 
can also be seen that the roof patch at the building’s southern 
end cannot be detected as there were no measure points 
available for this particular area. 
In Figure 4-7, the yellow roof outlines are not identical with 
the contours in the corresponding ortho images because the 
aerial imagery is not orthorectified. 
Figure 4. Segmentation of roof areas with an 
azimuthal exposition between 90° and 270° 
(left: evaluation data; right: LIDAR data) 
The main problems that have to be considered when modeling 
LIDAR data on such a large scale are the limitations of the 
resolution of the DEM measure points and a possible 
inaccuracy of the points’ position. 
Figure 5. DSM measure points 
representing the roofs height level which 
are eliminated by masking as they are 
positioned outside the building's outline 
(blue points) 
Figure 6. DSM measure points that 
represent the ground surface height 
inside the roof s outline are eliminated 
by a 3m threshold value (blue points) 
Inside and outside of the buildings’ outlines an approx. 1.5 m 
wide band can be detected where displaced points are located. 
Figure 5 shows displaced measure points (blue points) that 
represent the roofs surface height. But as they lie outside the 
roofs outline they are eliminated by the masking. Figure 6 
illustrates the same inaccuracy of the data inside the roofs’ 
outlines. The blue measure points are actually at ground surface 
height or at a particular height level of the building’s vertical 
facade, where the laserbeam incidentally hit the building’s wall. 
They are eliminated by a 3 m threshold value segmentation. 
Brenner & Haala (1999) also confirm this problem of misplaced 
points in their contributions to 3D city modeling. 
Misplaced DSM points do not only affect the modeling of the 
buildings’ outer borders. They also cause problems as they 
impede a precise delimitation of different inner roof segments. 
This especially affects the- modeling of different flat roof