In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds). IAPRS. Vol. XXXVIII. Part 3A - Saint-Mandé, France. September 1-3. 2010 
Figure 2. Concept of DSM generation considering different data 
properties (layers /] to /„/) and different models (m, to m nn ,). 
The general case is, thus, to define the final surface model as a 
function /of the layers and the primary surface models. 
For evaluating the function at a specific (x.y) location, the 
arguments are the values of the layers /j(.v.v),... l„/(x,y), and the 
values of the models m\{x,y), ... m nn ,(x,y). The return value of f 
has to be metric, too. More specifically, to be interpretable as a 
surface model, /(x,y) has to be within the minimum-maximum 
range of /rt|(x,y),... m n Jx,y). 
The layers and the surface models may be derived as cell 
information (raster data) or as grids (vector data). For the 
computation of DSMs, several interpolation algorithms are 
available. Amongst the interpolation methods considered useful 
for topographic point clouds are Moving Least Squares, Inverse 
Distance Weighting, Kriging, gridding of a triangulation, etc. 
In the specific approach presented in this paper, the function/ 
shall be used to choose the DSM computed with an 
interpolation method suitable for the specific land cover class 
found at the grid post. The surface roughness was chosen, 
because it discriminates between street, house roofs, and open 
areas on the one hand, and strongly vegetated and rocky 
surfaces on the other hand. Thus only 2 groups, each 
representing a number of land cover classes, are formed, and 
therefore, two DSMs are computed. 
linkage. Arbitrary workflows can be constructed by embedding 
the respective OPALS modules in a scripting environment. To 
handle ALS data in the order of >10 9 points, a central data 
management component (OPALS Data Manager. ODM) was 
developed, providing efficient spatial data access and an 
administration concept for storing arbitrary point attributes (e.g. 
echo width, amplitude, classification, normal vector, etc.). In 
this study, the modules opalsCell, opalsGrid and opalsAlgebra 
are used to derive a land cover dependent DSM. 
2.2 Land cover dependent calculation of DSMs 
The module opalsCell is a raster based analysis tool 
accumulating specific features (min, max, mean, etc.) of a 
selected point attribute (z, amplitude, echo width, etc.). For the 
work at hand, first a DSM raster containing the maximum 
elevation of all points within a cell (DSM niax ) was derived. 
The aim of opalsGrid is to derive digital surface or terrain 
models (DSM/DTM) in regular grid structure using simple 
interpolation techniques like moving least squares, nearest 
neighbour or moving average. In our study we used the moving 
least squares interpolation with a plane as functional model, i.e.. 
a tilted regression plane is fitted through the k-nearest 
neighbours (A). Apart from the elevation (DSM mls ), the moving 
least squares interpolation allows for the derivation of 
additional features per grid post. Among these features are the 
standard error of the estimated grid post elevation (a z , 
roughness indicator) and the eccentricity (distance: grid point - 
centre of gravity of input points). These attributes have proven 
their worth in subsequent processing steps, especially to detect 
occluded and vegetated areas. 
2.3 Land cover dependent combination of DSMs 
For the land cover dependent combination of the DSMs the 
module opalsAlgebra is employed to derive a grid or raster 
model by combining multiple input grid and/or raster data sets. 
The cell values are calculated by applying an algebraic formula 
based on the values of the respective input grids. Any 
mathematical formula, and even an entire program code 
returning a scalar value, can be passed. In this study, we assume 
that the o z -layer can be used to classify the area in rough and 
smooth surfaces. Therefore, we combine the DSM niax and the 
DSM m!s depending on the corresponding c z -layer. The heights 
(z) of the land cover dependent DSM are calculated by (pseudo 
code): 
z[DSM] = z[a z ] < 0.2 or not z[DSM max ] ? z[DSM in i s ] : 
z[DSM max ] 
2.1 Implementation within OPALS 
OPALS (Orientation and Processing of Airborne Laser 
Scanning data) is a scientific software project developed at the 
Institute of Photogrammetry and Remote Sensing (I.P.F), TU 
Vienna (Mandlburger et al„ 2009). The aim of OPALS is to 
provide a complete workflow for processing large ALS projects. 
OPALS targets the following topics: processing of raw sensor 
data, quality control, georeferencing, modelling of structure 
lines, filtering of ALS point clouds, DTM interpolation, and 
subsequent applications like city modelling, forestry, hydraulics 
etc. OPALS is a modular system consisting of small units 
(modules), each covering a well defined task. A software frame 
work is responsible for providing each module in three different 
implementations: (i) as command-line executable, (ii) as Python 
module (Phyton. 2010), and (iii) as C++-class library via DLL 
3. STUDY AREA AND DATA 
The proposed work flow is applied to four test sites located in 
Vienna (parts of the Schönbrunn Palace), lower Austria (north 
of the Ötscher mountain), Burgenland (Neusiedler See) and 
Vorarlberg (Montafon region), Austria. For the first three test 
sites full-waveform ALS data sets are available. For the first 
two test sites the ALS data is acquired with a Riegl LMS-Q560. 
For the Burgenland test site a Riegl LMS-Q680 was used 
(Riegl. 2010). The ALS data acquisition took place under leaf- 
off conditions. The point density (echoes per nr) is approx. 60 
and approx. 20 for the Vienna and lower Austria / Burgenland 
test site respectively. For the Vorarlberg test site, ALS data with 
a point density of approx. 5.5 echoes per nr are available and 
the discrete echoes were acquired as first and last echoes. This