In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
635 
with similar shapes and spectral properties. Finally, the means 
and standard deviations of the geometric and spectral variables 
were calculated for each segment. 
3.3 Tree cover 
The extraction of the area covered by trees is required for the 
area-wide mapping of the classified tree species. Tree cover and 
non-tree area masks were generated as described in detail in 
Waser et al. (2008). Briefly summarized: First, digital canopy 
height models (CHM) were produced subtracting the LiDAR 
DTM from the three DSMs. In a second step, pixels with CHM 
values > 3 m were used to extract potential tree areas according 
to the definition in the Swiss NFI (Brassel and Lischke, 2001). 
In a third step, non-tree objects, e.g. buildings, rocks, and 
artifacts were removed using spectral information from the 
ADS40-SH40 and ADS40-SH52 RGB images (low IHS pixel 
values) as well as information (curvature) about the image 
segments (e.g. segments on buildings have lower curvature 
values and ranges than trees or large shrubs). These four steps 
resulted in three canopy covers providing sunlit tree area for 
each study area. 
3.4 Classification of tree species 
3.4.1 Evaluation of modelling procedures: Image segments 
representing single trees were to be assigned to classes (species) 
by predictive modelling. The classes were given by a field 
sample from the 7 dominant tree species of the study area as 
described in section 2.2. As the response variable has more than 
two possible states, a multinomial model had to be applied. The 
logistic regression model is a special case of the generalized 
linear model (GLM) and described in e.g. McCullagh and 
Nelder (1983). Combination of logistic models was 
implemented by fitting a binomial logistic regression model to 
each class (species) separately and assigning the respective 
segment to the species with the highest probability. For details 
on the logistic regression function with quadratic terms see e.g. 
Hosmer and Lemeshow (2000). The explanatory variables as 
given in section 3.1 were used. 
In a first run, a single classification was performed using each 
set of variables separately. Then the explanatory variables from 
both the 2007 May and July images were tested together within 
a logistic regression model since the same flight path was used 
and the shadows were quite similar. Due to large differences in 
the flight paths and shadows between 2007 and 2008 the 2008 
data had to be used separately in a separate logistic regression 
model (see Fig. 2). In total, tree species were classified four 
times using different logistic regression models and input 
imagery (see also table 3). 
Figure 2. Example of the same area of trees acquired by 
different flight paths between 2008 August (left) and 2007 July 
(right) images. 
3.4.2 Validation: In order to validate the predictions of tree 
species, the digitized reference tree data (see section 2.2) had to 
be assigned to the corresponding image segments. Since the 
delineations of the field samples were not always congruent 
with the automatically generated image segments each of the 
230 digitized reference trees was assigned to an image segment 
using the following rule: If one segment contained more than 
one digitized field sample, the segment was assigned to the field 
sample covering the greater part of the segment. If less than 
10% of the image segment was covered by the sample polygon, 
the segment was not assigned at all. The predictive power of the 
models was verified by a 5-fold cross-validation. The statistical 
measures used to validate the results were: producer’s (PA)- and 
user’s accuracy (UA), correct classification rate (CCR), and 
kappa coefficient (K). 
4. RESULTS 
4.1 Confusion matrices 
The classification of the seven tree species was achieved semi- 
automatically and, depending on the image data used, quite 
high accuracies were obtained. The overall accuracies for tree 
species classification obtained by the different input imagery 
are summarized in table 3. The confusion matrices of the May, 
July and August classifications with best CCR and K are 
summarized in Tables 4-6. The classified main tree species 
are: Abies alba (Aa), Picea abies (Pa), Pinus sylvestris (Ps), 
Larix decidua (La), Acer sp. (Ac), Fagus sylvatica (Fs), and 
Fraxinus excelsior (Fe). 
Input data sets 
CCR 
K 
05-2007 
0.798 
0.691 
07-2007 
0.668 
0.598 
05 and 07-2007 combined 
0.691 
0.632 
08-2008 
0.757 
0.667 
Table 3. Overall accuracies for four different tree species 
classifications. 
Table 4 shows that five of seven tree species are classified with 
accuracies > 73% when using the May 2007 images. Best 
agreements are obtained for Picea abies (92%) and Fagus 
sylvatica (86%). The most frequent failures happen in 
classifying the non-dominant tree species Acer sp. (43%), and 
Larix decidua (56%) which are often misclassified either as 
Fagus sylvatica or Picea abies. 
May 2007 Classified as 
Field 
Aa 
Pa 
Ps 
La 
Ac 
Fs 
Fe 
PA 
Aa 
29 
2 
— 
— 
— 
— 
4 
0.83 
Pa 
— 
77 
— 
2 
2 
2 
1 
0.92 
Ps 
- 
~ 
14 
— 
— 
— 
— 
0.76 
La 
— 
12 
1 
10 
— 
— 
— 
0.43 
Ac 
1 
2 
- 
— 
19 
9 
3 
0.56 
Fs 
1 
3 
— 
— 
3 
55 
2 
0.86 
Fe 
3 
4 
— 
— 
2 
2 
54 
0.83 
UA 
0.85 
0.76 
0.93 
0.67 
0.73 
0.81 
0.84 
Table 4. Confusion matrix for tree species classification using 
the explanatory variables from May 2007 ADS40-SH40